DNA Microarray

Data Analysis

TOMI PASANEN, JANNA SAARELA, ILANA SAARIKKO, TEEMU TOIVANEN,

MARTTI TOLVANEN, MAUNO VIHINEN AND GARRY WONG
EDITORS JARNO TUIMALA AND M. MINNA LAINE
CSC
DNA Microarray Data Analysis
DNA Microarray Data Analysis

Editors
Jarno Tuimala
M. Minna Laine

CSC, the Finnish IT center for Science

CSC – Scientific Computing Ltd. is a non-profit organization for high-
performance computing and networking in Finland. CSC is owned by the
Ministry of Education. CSC runs a national large-scale facility for compu-
tational science and engineering and supports the university and research
community. CSC is also responsible for the operations of the Finnish Uni-
versity and Research Network (Funet).

All rights reserved. The PDF version of this book or parts of it can be
used in Finnish universities as course material, provided that this copyright
notice is included. However, this publication may not be sold or included
as part of other publications without permission of the publisher.

c The authors and
CSC – Scientific Computing Ltd.
2003

ISBN 952-9821-89-1

http://www.csc.fi/oppaat/siru/

Printed at
Picaset Oy
Helsinki 2003
DNA microarray data analysis 5

Preface

This is the first edition of the DNA microarray data analysis guidebook. Although
invented in the mid-90s, DNA microarrays are still novelties as biomedical research
tools. DNA microarrays generate large amounts of numerical data, which should
be analyzed effectively.
In this book, we hope to offer a broad view of basic theory and techniques
behind the DNA microarray data analysis. Our aim was not to be comprehensive,
but rather to cover the basics, which are unlikely to change much over years. We
hope that especially researchers starting their data analysis can benefit from the
book.
The text emphasizes gene expression analysis. Topics, such as genotyping,
are discussed shortly. This book does not cover the wet-lab practises, such as sam-
ple preparation or hybridization. Rather, we start when the microarrays have been
scanned, and the resulting images analyzed. In other words, we take the files with
signal intensities, which usually generate questions such as: “How is the data nor-
malized?” or “How do I identify the genes which are upregulated?”. We provide
some simple solutions to these specific questions and many others.
Each chapter has a section on suggested reading, which introduces some of
the relevant literature. Several chapters also include data analysis examples using
GeneSpring software.
This edition of the book was written by M. Minna Laine (chapters 4, 8 and
14), Tomi Pasanen (chapter 11), Janna Saarela (chapters 2 and 3), Ilana Saarikko
(chapter 8), Teemu Toivanen (chapter 14), Martti Tolvanen (chapter 12), Jarno Tu-
imala (chapters 4, 6, 7, 8, 9, 10, 13 and 15), Mauno Vihinen (chapters 10, 11 and
12), and Garry Wong (chapters 1 and 5).
Juha Haataja and Leena Jukka are warmly acknowledged for their support
during the production of this book.
We are very interested in receiving feedback about this publication. Especially,
if you feel that some essential technique has been missed, let us know. Please send
your comments to the e-mail address [email protected].

Espoo, 19th May 2003

The authors
6 DNA microarray data analysis

List of Contributors

M. Minna Laine Martti Tolvanen

CSC, the Finnish IT center for Science Institute of Medical Technology
Tekniikantie 15 a D Lenkkeilijänkatu 8
02101 Espoo 33520 Tampere
Finland Finland

Tomi Pasanen Jarno Tuimala

Institute of Medical Technology CSC, the Finnish IT center for Science
Lenkkeilijänkatu 8 Tekniikantie 15 a D
33520 Tampere 02101 Espoo
Finland Finland

Janna Saarela Mauno Vihinen

Biomedicum Biochip Center Institute of Medical Technology
Haartmaninkatu 8 Lenkkeilijänkatu 8
00290 Helsinki 33520 Tampere
Finland Finland

Ilana Saarikko Garry Wong

Centre for Biotechnology A. I. Virtanen -institute
Tykistökatu 6 University of Kuopio
20521 Turku 70211 Kuopio
Finland Finland

Teemu Toivanen
Centre for Biotechnology
Tykistökatu 6
20521 Turku
Finland
Contents 7

Contents

Preface 5

List of Contributors 6

I Introduction 14
1 Introduction 15
1.1 Why perform microarray experiments? . . . . . . . . . . . . . 15
1.2 What is a microarray? . . . . . . . . . . . . . . . . . . . . . . 15
1.3 Microarray production . . . . . . . . . . . . . . . . . . . . . 16
1.4 Where can I obtain microarrays? . . . . . . . . . . . . . . . . 17
1.5 Extracting and labeling the RNA sample . . . . . . . . . . . . 19
1.6 RNA extraction from scarse tissue samples . . . . . . . . . . . 19
1.7 Hybridization . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.8 Scanning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.9 Typical research applications of microarrays . . . . . . . . . . 21
1.10 Experimental design and controls . . . . . . . . . . . . . . . . 22
1.11 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Affymetrix Genechip system 25

2.1 Affymetrix technology . . . . . . . . . . . . . . . . . . . . . 25
2.2 Single Array analysis . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Detection p-value . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Detection call . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.5 Signal algorithm . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.6 Analysis tips . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Comparison analysis . . . . . . . . . . . . . . . . . . . . . . 27
2.8 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.9 Change p-value . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.10 Change call . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.11 Signal Log Ratio Algorithm . . . . . . . . . . . . . . . . . . . 29

3 Genotyping systems 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
8 DNA microarray data analysis

3.2 Methodologies . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Genotype calls . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 33

4 Overview of data analysis 34

4.1 cDNA microarray data analysis . . . . . . . . . . . . . . . . . 34
4.2 Affymetrix data analysis . . . . . . . . . . . . . . . . . . . . 35
4.3 Data analysis pipeline . . . . . . . . . . . . . . . . . . . . . . 35

5 Experimental design 38
5.1 Why do we need to consider experimental design? . . . . . . . 38
5.2 Choosing and using controls . . . . . . . . . . . . . . . . . . 38
5.3 Choosing and using replicates . . . . . . . . . . . . . . . . . . 39
5.4 Choosing a technology platform . . . . . . . . . . . . . . . . 39
5.5 Gene clustering v. gene classification . . . . . . . . . . . . . . 40
5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.7 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 41

6 Basic statistics 42
6.1 Why statistics are needed . . . . . . . . . . . . . . . . . . . . 42
6.2 Basic concepts . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.1 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.2 Constants . . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.3 Distribution . . . . . . . . . . . . . . . . . . . . . . . . . 42
6.2.4 Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3 Simple statistics . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3.1 Number of subjects . . . . . . . . . . . . . . . . . . . . . 43
6.3.2 Mean (m) . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3.3 Trimmed mean . . . . . . . . . . . . . . . . . . . . . . . 43
6.3.4 Median . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6.3.5 Percentile . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.3.6 Range . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.3.7 Variance and the standard deviation . . . . . . . . . . . . 44
6.3.8 Coefficient of variation . . . . . . . . . . . . . . . . . . . 44
6.4 Effect statistics . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.4.1 Scatter plot . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.4.2 Correlation (r) . . . . . . . . . . . . . . . . . . . . . . . 45
6.4.3 Linear regression . . . . . . . . . . . . . . . . . . . . . . 46
6.5 Frequency distributions . . . . . . . . . . . . . . . . . . . . . 47
6.5.1 Normal distribution . . . . . . . . . . . . . . . . . . . . . 47
6.5.2 t-distribution . . . . . . . . . . . . . . . . . . . . . . . . 49
6.5.3 Skewed distribution . . . . . . . . . . . . . . . . . . . . . 49
6.5.4 Checking the distribution of the data . . . . . . . . . . . . 50
Contents 9

6.6 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . 51
6.6.1 Log 2-transformation . . . . . . . . . . . . . . . . . . . . 52
6.7 Outliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.8 Missing values and imputation . . . . . . . . . . . . . . . . . 53
6.9 Statistical testing . . . . . . . . . . . . . . . . . . . . . . . . 54
6.9.1 Basics of statistical testing . . . . . . . . . . . . . . . . . 54
6.9.2 Choosing a test . . . . . . . . . . . . . . . . . . . . . . . 55
6.9.3 Threshold for p-value . . . . . . . . . . . . . . . . . . . 55
6.9.4 Hypothesis pair . . . . . . . . . . . . . . . . . . . . . . . 55
6.9.5 Calculation of test statistic and degrees of freedom . . . . 56
6.9.6 Critical values table . . . . . . . . . . . . . . . . . . . . . 57
6.9.7 Drawing conclusions . . . . . . . . . . . . . . . . . . . . 57
6.9.8 Multiple testing . . . . . . . . . . . . . . . . . . . . . . . 57
6.10 Analysis of variance . . . . . . . . . . . . . . . . . . . . . . . 58
6.10.1 Basics of ANOVA . . . . . . . . . . . . . . . . . . . . . 58
6.10.2 Completely randomized experiment . . . . . . . . . . . . 58
6.11 Statistics using GeneSpring . . . . . . . . . . . . . . . . . . . 60
6.11.1 Simple statistics . . . . . . . . . . . . . . . . . . . . . . 60
6.11.2 Tranformations . . . . . . . . . . . . . . . . . . . . . . . 60
6.11.3 Scatter plot and histogram . . . . . . . . . . . . . . . . . 60
6.11.4 Correlation . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.11.5 Linear regression . . . . . . . . . . . . . . . . . . . . . . 61
6.11.6 One-sample t-test . . . . . . . . . . . . . . . . . . . . . . 62
6.11.7 Independent samples t-test and ANOVA . . . . . . . . . . 62
6.12 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 64

II Analysis 65
7 Preprocessing of data 66
7.1 Rationale for preprocessing . . . . . . . . . . . . . . . . . . . 66
7.2 Missing values . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.3 Checking the background reading . . . . . . . . . . . . . . . . 68
7.4 Calculation of expression change . . . . . . . . . . . . . . . . 69
7.4.1 Intensity ratio . . . . . . . . . . . . . . . . . . . . . . . . 69
7.4.2 Log ratio . . . . . . . . . . . . . . . . . . . . . . . . . . 70
7.4.3 Fold change . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.5 Handling of replicates . . . . . . . . . . . . . . . . . . . . . . 71
7.5.1 Types of replicates . . . . . . . . . . . . . . . . . . . . . 71
7.5.2 Time series . . . . . . . . . . . . . . . . . . . . . . . . . 71
7.5.3 Case-control studies . . . . . . . . . . . . . . . . . . . . 72
7.5.4 Power analysis . . . . . . . . . . . . . . . . . . . . . . . 72
7.5.5 Averaging replicates . . . . . . . . . . . . . . . . . . . . 72
7.6 Checking the quality of replicates . . . . . . . . . . . . . . . . 72
10 DNA microarray data analysis

7.6.1 Quality check of replicate chips . . . . . . . . . . . . . . 73

7.6.2 Quality check of replicate spots . . . . . . . . . . . . . . 73
7.6.3 Excluding bad replicates . . . . . . . . . . . . . . . . . . 73
7.7 Outliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
7.8 Filtering bad data . . . . . . . . . . . . . . . . . . . . . . . . 74
7.9 Filtering uninteresting data . . . . . . . . . . . . . . . . . . . 76
7.10 Simple statistics . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.10.1 Mean and median . . . . . . . . . . . . . . . . . . . . . . 77
7.10.2 Standard deviation . . . . . . . . . . . . . . . . . . . . . 77
7.10.3 Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
7.11 Skewness and normality . . . . . . . . . . . . . . . . . . . . . 77
7.11.1 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.12 Spatial effects . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.13 Normalization . . . . . . . . . . . . . . . . . . . . . . . . . . 81
7.14 Similarity of dynamic range, mean and variance . . . . . . . . 81
7.15 Examples using GeneSpring . . . . . . . . . . . . . . . . . . 82
7.15.1 Importing data . . . . . . . . . . . . . . . . . . . . . . . 82
7.15.2 Background subtraction . . . . . . . . . . . . . . . . . . 82
7.15.3 Calculation of expression change . . . . . . . . . . . . . 82
7.15.4 Replicates . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.15.5 Checking linearity . . . . . . . . . . . . . . . . . . . . . 83
7.15.6 Normality . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.15.7 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
7.16 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 84

8 Normalization 85
8.1 What is normalization? . . . . . . . . . . . . . . . . . . . . . 85
8.2 Sources of systematic bias . . . . . . . . . . . . . . . . . . . 85
8.2.1 Dye effect . . . . . . . . . . . . . . . . . . . . . . . . . . 85
8.2.2 Scanner malfunction . . . . . . . . . . . . . . . . . . . . 85
8.2.3 Uneven hybridization . . . . . . . . . . . . . . . . . . . . 86
8.2.4 Printing tip . . . . . . . . . . . . . . . . . . . . . . . . . 86
8.2.5 Plate and reporter effects . . . . . . . . . . . . . . . . . . 86
8.2.6 Batch effect and array design . . . . . . . . . . . . . . . . 87
8.2.7 Experimenter issues . . . . . . . . . . . . . . . . . . . . 87
8.2.8 What might help to track the sources of bias? . . . . . . . 87
8.3 Normalization terminology . . . . . . . . . . . . . . . . . . . 87
8.3.1 Normalization, standardization and centralization . . . . . 88
8.3.2 Per-chip and per-gene normalization . . . . . . . . . . . . 89
8.3.3 Global and local normalization . . . . . . . . . . . . . . . 89
8.4 Performing normalization . . . . . . . . . . . . . . . . . . . . 89
8.4.1 Choice of the method . . . . . . . . . . . . . . . . . . . . 89
Contents 11

8.4.2 Basic idea . . . . . . . . . . . . . . . . . . . . . . . . . . 90

8.4.3 Control genes . . . . . . . . . . . . . . . . . . . . . . . . 90
8.4.4 Linearity of data matters . . . . . . . . . . . . . . . . . . 91
8.4.5 Basic normalization schemes for linear data . . . . . . . . 91
8.4.6 Special situations . . . . . . . . . . . . . . . . . . . . . . 91
8.5 Mathematical calculations . . . . . . . . . . . . . . . . . . . . 92
8.5.1 Mean centering . . . . . . . . . . . . . . . . . . . . . . . 92
8.5.2 Median centering . . . . . . . . . . . . . . . . . . . . . . 92
8.5.3 Trimmed mean centering . . . . . . . . . . . . . . . . . . 92
8.5.4 Standardization . . . . . . . . . . . . . . . . . . . . . . . 92
8.5.5 Lowess smoothing . . . . . . . . . . . . . . . . . . . . . 93
8.5.6 Ratio statistics . . . . . . . . . . . . . . . . . . . . . . . 94
8.5.7 Analysis of variance . . . . . . . . . . . . . . . . . . . . 94
8.5.8 Spiked controls . . . . . . . . . . . . . . . . . . . . . . . 94
8.5.9 Dye-swap experiments . . . . . . . . . . . . . . . . . . . 94
8.6 Some caution is needed . . . . . . . . . . . . . . . . . . . . . 95
8.7 Graphical example . . . . . . . . . . . . . . . . . . . . . . . 95
8.8 Example of calculations . . . . . . . . . . . . . . . . . . . . . 95
8.9 Using GeneSpring for normalization . . . . . . . . . . . . . . 96
8.10 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 98

9 Finding differentially expressed genes 100

9.1 Identifying over- and underexpressed genes . . . . . . . . . . 100
9.1.1 Filtering by absolute expression change . . . . . . . . . . 100
9.1.2 Statistical single chip methods . . . . . . . . . . . . . . . 100
9.1.3 Noise envelope . . . . . . . . . . . . . . . . . . . . . . . 101
9.1.4 Sapir and Churchill’s single slide method . . . . . . . . . 101
9.1.5 Chen’s single slide method . . . . . . . . . . . . . . . . . 102
9.1.6 Newton’s single slide method . . . . . . . . . . . . . . . 103
9.2 What about the confidence? . . . . . . . . . . . . . . . . . . . 104
9.2.1 Only some treatments have replicates . . . . . . . . . . . 104
9.2.2 All the treatments have replicates: two-sample t-test . . . 105
9.2.3 All the treatments have replicates: one-sample t-test . . . 106
9.3 GeneSpring examples . . . . . . . . . . . . . . . . . . . . . . 106
9.4 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 107

10 Cluster analysis of microarray information 108

10.1 Basic concept of clustering . . . . . . . . . . . . . . . . . . . 108
10.2 Principles of clustering . . . . . . . . . . . . . . . . . . . . . 108
10.3 Hierarchical clustering . . . . . . . . . . . . . . . . . . . . . 109
10.4 Self-organizing map . . . . . . . . . . . . . . . . . . . . . . . 110
10.5 K-means clustering . . . . . . . . . . . . . . . . . . . . . . . 111
10.6 Principal component analysis . . . . . . . . . . . . . . . . . . 112
12 DNA microarray data analysis

10.7 Pros and cons of clustering . . . . . . . . . . . . . . . . . . . 113

10.8 Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
10.9 Programs for clustering and visualization . . . . . . . . . . . . 116
10.10 Function prediction . . . . . . . . . . . . . . . . . . . . . . . 117
10.11 GeneSpring and clustering . . . . . . . . . . . . . . . . . . . 117
10.11.1 Clustering tool . . . . . . . . . . . . . . . . . . . . . . . 117
10.11.2 Principal components analysis tool . . . . . . . . . . . . . 118
10.11.3 Predict parameter value tool . . . . . . . . . . . . . . . . 119
10.12 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 119

III Data mining 120

11 Gene regulatory networks 121
11.1 What are gene regulatory networks? . . . . . . . . . . . . . . 121
11.2 Fundamentals . . . . . . . . . . . . . . . . . . . . . . . . . . 121
11.3 Bayesian network . . . . . . . . . . . . . . . . . . . . . . . . 123
11.4 Calculating Bayesian network parameters . . . . . . . . . . . 124
11.5 Searching Bayesian network structure . . . . . . . . . . . . . 126
11.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
11.7 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 128
12 Data mining for promoter sequences 129
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
12.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
12.3 Finding promoter region sequences . . . . . . . . . . . . . . . 130
12.4 Using EnsMart to retrieve promoter regions . . . . . . . . . . 133
12.5 Comparison of EnsMart and UCSC searches . . . . . . . . . . 135
12.6 Pattern search without prior knowledge . . . . . . . . . . . . . 137
12.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
12.8 GeneSpring and promoter analysis . . . . . . . . . . . . . . . 138
12.9 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 139
13 Annotations and article mining 140
13.1 Retrieving annotations from public databases . . . . . . . . . 140
13.2 Retrieving annotations using BLAST . . . . . . . . . . . . . . 141
13.3 Article mining . . . . . . . . . . . . . . . . . . . . . . . . . . 141
13.4 Annotation and gene ontologies using GeneSpring . . . . . . . 142
13.4.1 Annotations . . . . . . . . . . . . . . . . . . . . . . . . . 142
13.4.2 Ontologies . . . . . . . . . . . . . . . . . . . . . . . . . 142

IV Tools and data management 144

14 Reporting results 145
14.1 Why the results should be reported . . . . . . . . . . . . . . . 145
14.2 What details should be reported: the MIAME standard . . . . 145
Contents 13

14.3 How the data should be presented: the MAGE standard . . . . 147
14.3.1 MAGE-OM . . . . . . . . . . . . . . . . . . . . . . . . . 147
14.3.2 MAGE-ML; an XML-translation of MAGE-OM . . . . . 147
14.3.3 MAGE-STK . . . . . . . . . . . . . . . . . . . . . . . . 148
14.4 Where and how to submit your data . . . . . . . . . . . . . . 148
14.4.1 ArrayExpress and GEO . . . . . . . . . . . . . . . . . . . 148
14.4.2 MIAMExpress . . . . . . . . . . . . . . . . . . . . . . . 148
14.4.3 GEO . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
14.4.4 Other options and aspects . . . . . . . . . . . . . . . . . 149
14.5 MIAME-compliant sample attributes in GeneSpring . . . . . 150
14.6 Suggested reading . . . . . . . . . . . . . . . . . . . . . . . . 150
15 Software issues 152
15.1 Data format conversions problems . . . . . . . . . . . . . . . 152
15.2 A standard file format . . . . . . . . . . . . . . . . . . . . . . 152
15.3 Programming . . . . . . . . . . . . . . . . . . . . . . . . . . 153
15.3.1 Perl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
15.3.2 Awk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
15.3.3 R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
15.4 Freeware software packages . . . . . . . . . . . . . . . . . . . 154
15.4.1 Cluster and treeview . . . . . . . . . . . . . . . . . . . . 155
15.4.2 Expression profiler . . . . . . . . . . . . . . . . . . . . . 155
15.4.3 ArrayViewer . . . . . . . . . . . . . . . . . . . . . . . . 155
15.4.4 MAExplorer . . . . . . . . . . . . . . . . . . . . . . . . 155
15.4.5 Bioconductor . . . . . . . . . . . . . . . . . . . . . . . . 155
15.5 Commercial software packages . . . . . . . . . . . . . . . . . 156
15.5.1 VisualGene . . . . . . . . . . . . . . . . . . . . . . . . . 156
15.5.2 GeneSpring . . . . . . . . . . . . . . . . . . . . . . . . . 156
15.5.3 Kensington . . . . . . . . . . . . . . . . . . . . . . . . . 156
15.5.4 J-Express . . . . . . . . . . . . . . . . . . . . . . . . . . 156
15.5.5 Expression Nti . . . . . . . . . . . . . . . . . . . . . . . 157
15.5.6 Rosetta Resolver . . . . . . . . . . . . . . . . . . . . . . 157
15.5.7 Spotfire . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

Index 158
 Part I

Introduction
1 Introduction 15

1 Introduction
Microarray technologies as a whole provide new tools that transform the way sci-
entific experiments are carried out. The principle advantage of microarray tech-
nologies compared with traditional methods is one of scale. In place of conducting
experiments based on results from one or a few genes, microarrays allow for the
simultaneous interrogation of hundreds or thousands of genes.

1.1 Why perform microarray experiments?

The answers to this question span a wide range from the formally considered, well-
constructed hypotheses with elegant supporting arguments to “I can’t think of any-
thing else to do, so lets’s do a microarray experiment”. The true motivation for
performing these experiments lies likely somewhere between the two extremes. It
is this combination of generating a scientific hypothesis (elegant or not), and at the
same time being able to produce massive amounts of data that has made research
in microarrays so attractive. Nonetheless, the production and use of microarrays is
set with high technical and instrumentation demands. Moreover, the computation
and statistical requirements for dealing with the data can be daunting, especially
to those scientists used to single experiment – single result analysis. So, for those
willing to try this new technology, microarray experiments are performed to answer
a wide range of biological questions to which the answers are to be found in the
realm of hundreds, thousands, or an entire genome of individual genes.

1.2 What is a microarray?

Microarrays are microscope slides that contain an ordered series of samples (DNA,
RNA, protein, tissue). The type of microarray depends upon the material placed
onto the slide: DNA, DNA microarray; RNA, RNA microarray; protein, protein
microarray; tissue, tissue microarray. Since the samples are arranged in an ordered
fashion, data obtained from the microarray can be traced back to any of the samples.
This means that genes on the microarray are addressable. The number of ordered
samples on a microarray can number into the hundred of thousands. The typical
microarray contains several thousands of addressable genes.
The most commonly used microarray is the DNA microrray. The DNA printed
or spotted onto the slides can be chemically synthesized long oligonucleotides or
enzymatically generated PCR products. The slides contain chemically reactive
groups (typically aldehydes or primary amines) that help to stabilize the DNA
16 DNA microarray data analysis

onto the slide, either by covalent bonds or electrostatic interactions. An alterna-

tive technology allows the DNA to be synthesized directly onto the slide itself by
a photolithographic process. This process has been commercialized and is widely
available. DNA microarrays are used to determine

1. The expression levels of genes in a sample, commonly termed expression

profiling.

2. The sequence of genes in a sample, commonly termed minisequencing for

short nucleotide reads, and mutation or SNP analysis for single nucleotide
reads.

1.3 Microarray production

Printing microarrays in not a trivial task and is both an art and a science. The
job requires considerable expertise in chemistry, engineering, programming, large
project management, and molecular biology. The aim during printing is to produce
reproducible spots with consistant morphology. Early versions of printers were
custom made with a basic design taken from a prototype version. Some were built
from laboratory robots. Current commercial microarray printers are available for
almost every size application and have made the task of printing microarrays feasi-
ble and affordable for many nonspecialized laboratories. The basic printer consists
of: a nonvibrating table surface where the raw glass slides are placed, a moving
head in the x-y-z plane that contains the pins or pens to transfer the samples onto
the array, a wash station to clean the pins/pens between samples, a drying station
for the pins/pens, a place for the samples to be printed, and a computer to con-
trol the operation. Some of these procedures can be automated such as replacing
the samples to be printed, although most complete systems are semi-automated.
Samples to be printed are concentrated and stored in microtitre plates.
The printers are operated in dust-free, temperature and humidity controlled
rooms. Some printer designs have their own self-contained environmental controls.
Printing pen designs have been adapted from ink methods and include quill, ball-
point, ink-jet, and P-ring techniques. The pens can get stuck and need to be cleaned
frequently. Multiple pens placed on a printing head can multiplex the printing
operation and speed up the process. Thousands of samples in duplicate or triplicate
are printed in a single run over perhaps a hundred or more slides; thus, printing
times of several days are common. Since printing times can be long and sample
volumes are small, sample evaporation is a major concern. As a result, hygroscopic
printing buffers, often containing DMSO have been developed and are highly useful
to alleviate evaporation. A typical printer design is shown in Figure 1.1.
1 Introduction 17

Figure 1.1: Typical picture of a printer.

1.4 Where can I obtain microarrays?

Microarrays can be obtained from a variety of sources. Commercial microarrays

are of high quality, good density and available for the most commonly studied
organisms including human, mouse, rat, and yeast. Many companies also have
specialized arrays available for more focused investigations. Also microarray core
facilities located in academic and government institutions produce microarrays that
are available for use (Table 1.1). The microarrays from these centers have definite
cost advantages and excellent quality standards, but tend to have a more restricted
focus in terms of the types of arrays and species covered. Custom microarrays that
are designed and produced for individual or limited use are a growing application.
Custom fabrication of a microarray still requires a microarray printer or spotter
facility to produce the array, but has the advantage of producing smaller batches of
slides and more flexibility in the genes on array. If you are fortunate to have your
own facility, microarrays can be produced at your own convenience. Microarrays
once made store well in dark dessicated plastic slide boxes. Some manufacturers
suggest storage at −20 ◦ C while others find room temperature adequate. The shelf
life of microarrays has been claimed to be up to 6 months although this has not
been empiracally tested.
18 DNA microarray data analysis

Table 1.1: Places where microarrays can be obtained.

Center name Website

Biomedicum Biochip Center www.helsinki.fi/biochipcenter
Finnish DNA microarray Center microarrays.btk.utu.fi
Norwegian Microarray Consortium www.med.uio.no/dnr/microarray/
Ontario Cancer Institute www.microarrays.ca

Figure 1.2: Work flow of a typical expression microarray experiment.

1 Introduction 19

1.5 Extracting and labeling the RNA sample

A typical workflow of the microarray experiment has been summarized in Fig-
ure 1.2. Once microarrays have been made and obtained, the next stage is to obtain
samples for labeling and hybridization. Labeling RNA for expression analysis gen-
erally involves three steps:
1. Isolation of RNA.

2. Labeling the RNA by a reverse transcription procedure with fluorescent mark-

ers.

3. Purification of the labeled products.

RNA can be extracted from tissue or cell samples by common organic extrac-
tion procedures used in most molecular biology labs. Many commercial kits are
available for this task. Both total RNA and mRNA can be used for labeling, but the
contaminating genomic DNA must be removed by DNAase treatment. The amount
of total RNA necessary for a single labeling reaction is about 20 µg while the
amount of mRNA necessary is about 0.5 µg. Lesser amounts are known to work,
but require extreme purity and well developed protocols. Thus, while the absolute
amounts may vary, the purity and integrity of the RNA is an absolute must. It is
generally a good idea to check the RNA samples before using them in microarray
experiments. In fact, for many core facilities it is a requirement. This can be done
by assaying the absorption ratio 260/280 lambda and/or running a sample on an
ethidium bromide stained agarose gel.
Direct labeling of the RNA is achieved by producing cDNA from the RNA by
using the enzyme reverse transcriptase and then incorporating the fluorescent la-
bels, most commonly Cy3 and Cy5 . Other fluorophores are available (e.g. Cy3.5,
TAMRO, Texas red) but have not yet found widespread use. In the indirect proce-
dure, a reactive group, usually a primary amine, is incorporated into the cDNA first,
and the Cy3 or Cy5 is then coupled to the cDNA in a separate reaction. The advan-
tage of the indirect method is a higher labeling efficiency due to the incorporation
of a smaller molecule during the reverse transcription step. Once fluorescently
labeled probes are made, the free unincorporated nucleotides must be removed.
This is typically done by column chromatography using convenient spin-columns
or by ethanol precipitation of the sample. Some protocols perform both purification
steps. As a small aside, radioactivity is still around and may even make a comeback
in microarrays. Incorporation of 33 P- or 35 S-labeled nucleotides into cDNAs have
high rates and provide more sensitivity than fluoresecently labeled probes. These
features have been exploited in plastic microarrays that have the gene density and
size of microarrays but require far less instrumentation and are reusable.

1.6 RNA extraction from scarse tissue samples

For many microarray applications there is a scarcity of tissue available for RNA
extraction. This seems to be the case especially in human tissue studies. In re-
sponse, many scientists have developed techniques aimed at getting around this
20 DNA microarray data analysis

problem. These procedures generally involve either PCR amplification of the cD-
NAs made from the original RNAs, or production of more RNA from the original
RNA sample by hybridization of a T7 or T3 promoter followed by RNA sythesis
with RNA polymerase. As usual for any amplification procedure, proper controls
and interpretation of the results need to be considered.
A related issue in isolating tissues for microarray studies is the dissection of
small populations of cells or even single cells. Sophisticated instruments have been
developed for this application and many are commercially available. These laser-
assisted microdissection machine, while expensive, are nonetheless fairly straight-
forward to use and provide a convenient method for obtaining pure cell samples.

1.7 Hybridization
Conditions for hybridizing fluorescently labeled DNAs onto microarrays are re-
markably similar to hybridizations for other molecular biology applications. Gener-
ally the hybridization solution contains salt in the form of buffered standard sodium
citrate (SSC), a detergent such as sodium dodecyl sulphate (SDS), and nonspecific
DNA such as yeast tRNA, salmon sperm DNA, and/or repetitive DNA such as hu-
man Cot-1. Other nonspecific blocking reagents used in hybridization reactions
include bovine serum albumin or Denhardt’s reagent. Lastly, the hybridization so-
lution should contain the labeled cDNAs produced from the different RNA popula-
tions.
Hybridization temperatures vary depending upon the buffers used, but gen-
erally are performed at approximately 15 − 20 ◦ C below the melting temperature,
which is 42 − 45 ◦C for PCR products in 4X SSC and 42 − 50 ◦C for long oligos.
Hybridization volumes vary widely from 20µl to several mLs. For small hybridis-
ation volumes, hydrophobic cover slips are used. For larger volumes, hybridization
chambers can be used.
Hybridization chambers are necessary to keep the temperature constant and
the hybridization solution from evaporating. Hybridization chambers vary substan-
tially from the most expensive high-tech automated instruments to empty pipette
boxes with a few wet paper towels inserted. The range of solutions for providing
a thermally stable, humidified enviroment for a microscope slide is only virtually
unlimited. Some might even consider a sauna as a potential chamber. In small
volumes, the hybridization kinetics are rapid so a few hours can yield reproducible
results, although overnight hybridizations are more common.

1.8 Scanning
Following hybridization, microarrays are washed for several minutes in decreas-
ing salt buffers and finally dried, either by centrifugation of the slide, or a rinse
in isopropanol followed by quick drying with nitrogen gas or filtered air. Fluores-
cently labeled microarrays can then be “read” with commercially available scan-
ners. Most microarray scanners are basically scanning confocal microscopes with
lasers exciting at wavelengths specifically for Cy3 and Cy5, the typical dyes used
in experiments. The scanner excites the fluorescent dyes present at each spot on the
1 Introduction 21

microarray and the dye then emits at a characteristic wavelength that is captured
in a photomultiplier tube. The amount of signal emitted is directly in proportion
to the amount of dye at the spot on the microarray and these values are obtained
and quantitated on the scanner. A reconstruction of the signals from each location
on the microarray is then produced. For cDNA microarrays one intensity value is
generated for the Cy3 and another for the Cy5. Hence, cDNA microarrays pro-
duce two-color data. Affymetrix chips produce one-color data, because only one
mRNA sample is hybridized to every chip (see chapter 3). When both dyes are
reconstructed together, a composite image is generated. This image produces the
typical microarray picture.

1.9 Typical research applications of microarrays

The types and numbers of applications for microarray experiments are quite vari-
able and constantly increasing. Microarrays used to monitor the expression level of
genes in comparison between two conditions remains one of the most widespread
uses of microarrays. This type of study, termed gene expression profiling, can be
used to determine the function of particular genes during a particular state, such as
nutrition, temperature, or chemical environment. Such results could be observed as
up- or down-regulation, or unchanged during particular conditions. For example,
a group of genes could be up-regulated during heat shock, and as a group, these
genes could be assigned as heat shock responsive genes. Some genes in this group
may have already been identified as heat shock responsive, but other genes in the
group may not have been assigned any function. Based on a similar response to
heat shock, new functions are then assigned to the genes. Therefore, extrapola-
tion of function based on common changes in expression remains one of the most
widespread applications of microarray research. By assumption, genes that share
common regulatory patterns also share the same function.
In agriculture, microarrays have been used to identify genes which are in-
volved in the ripening of tomatos, for example. In this type of study, RNAs are iso-
lated from raw and ripened fruit, and then compared to determine which genes are
expressed during the process. Genes which are down-regulated during the ripening
may also provide useful information about the process.
On a basic scientific level, microarrays have been used to map the cellular, re-
gional, or tissue-specific localization of genes and their respectively encoded pro-
teins. Microarrays have been used: at the subcellular level to map genes that encode
membrane or cytosolic proteins; at the cellular level to map genes that distinguish
between different types of immune cells; at the tissue region level to disinguish
genes which encode hippocampus or cortex brain region specific proteins; and at
the tissue level to identify genes which are expressed in muscle, liver, or heart tis-
sues.
Pharmacological studies have also used microarrays as a means of discerning
the mechanism of action of therapeutic agents and as a corrollary to develop new
drug targets. The guiding principle in this endeavor is that genes regulated by ther-
apeutic agents result from the actions of the drug. Identification of the genes that
are regulated by a certain drug could potentially provide insight into the mechanism
22 DNA microarray data analysis

of action of the drug, prediction of toxicologic properties, and new drug targets.
One of the most exciting areas of application is the diagnosis of clinically
relevent diseases. The oncology field has been especially active and to an extent
successful in using microarrays to differentiate between cancer cell types. The
ability to identify cancer cells based on gene expression represents a novel method-
ology that has real benefits. In difficult cases where a morphological or an antigen
marker is not available or reliable enough to distinguish cancer cell types, gene
expression profiling using microarrays can be extremely valuable. Programs to
predict clinical outcome and to design individual therapies based on expression
profiling results are well underway.
A very recent application of microarrays has been to perform comparative
genomic analysis. Genome projects are producing sequences on a massive level,
yet there still does not exist sufficient resources to sequence every organism that
seems interesting or worthy of the effort. Therefore, microarrays have been used as
a shortcut to both characterize the genes within an organism (structural genomics)
and also to determine whether those genes are expressed in a similar way to a
reference organism (functional genomics). A good example of this is in the species
Oryza sativa (rice). Microarrays based on rice sequences can be used to hybridize
cDNAs derived from other plant species such as corn or barley. The genome sizes
in the latter are simply too large for whole genome projects, so hybridization with
microarrays to rice genes presents an agile way to address this question.
Single nucleotide polymorphism (SNP) microarrays are designed to detect the
presence of single nucleotide differences between genomic samples. SNPs occur at
frequencies of approximately 1 in a 1000 bases in humans and underlie the genomic
differences between individuals. Mapping and obtaining frequencies of identified
SNPs should provide a genetic basis for identifying disease genes, predicting ef-
fects of the environment as well as responses to therapeutic agents. Minisequenc-
ing, primer extension, and differential hybridization methods have been developed
on the microarray platform with all the advantages of expression arrays: high
throughput, reproducibility, economy, and speed.
Indeed, the use of microarrays to determine whether a gene is present and
whether it goes up or down under certain conditions will continue to spawn even
more applications that now depend only upon the imagination of the microarray
researcher.

1.10 Experimental design and controls

Good experimental design in a microarray project requires the same principles and
practices that are part of any scientific investigation. Appropriate controls are the
foundation to any experiment. Both positive controls and negative controls can pro-
vide confidence in the results and even provide insight into the success or failure
of the experimental protocol. Sufficient replicates should be planned to decrease
experimental error and to provide statistical power. Forethought and consultation
on the correct statistical practices and procedures for the design are always advan-
tageous. Attention to experimental parameters is a must, so that whatever treat-
ment, time, dose, individual, or tissue location is being studied, the results will be
1 Introduction 23

interpretable with minimum number of confounders. Indeed, probably the only dif-
ference between good experimental design in microarray and other experiments is
that the time budgeted for data analysis seems always to be underestimated. As a
final suggestion, attention to the mundane but critical statistical and data analysis
elements of a microarray experiment will greatly increase your ratio of joy to pain
at the end of your microarray journey.

1.11 Suggested reading

1. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeck-
ert, C., Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., Gaasterland,
T., Glenisson, P., Holstege, F. C., Kim, I. F., Markowitz, V., Matese, J. C.,
Parkinson, H., Robinson, A., Sarkans, U., Schulze-Kremer, S., Stewart, J.,
Taylor, R., Vilo, J., Vingron, M. (2001) Minimum information about a mi-
croarray experiment (MIAME)-toward standards for microarray data. Nat.
Genet. 29, 365-371.
2. Brown, P. O., and Botstein, D. (1999) Exploring the new world of the genome
with DNA microarrays. Nat. Genet. 21, 33-37.
3. Chee, M., Yang, R., Hubbell, E., Berno, A., Huang, X. C., Stern, D., Winkler,
J., Lockhart, D. J., Morris, M. S., Fodor, S. P. (1996) Accessing genetic
information with high-density DNA arrays. Science 274, 610-614.
4. Eisen, M. B., Spellman, P. T., Brown, P. O., Botstein, D. (1998) Cluster
analysis and display of genome-wide expression patterns. Proc. Natl. Acad.
Sci. U S A 95, 14863-14868.
5. Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov,
J. P., Coller, H., Loh, M. L., Downing, J. R., Caligiuri, M. A., Bloomfield, C.
D., Lander, E. S. (1999) Molecular classification of cancer: class discovery
and class prediction by gene expression monitoring. Science 286, 531-537.
6. Hacia, J. G., Makalowski, W., Edgemon, K., Erdos, M. R., Robbins, C. M.,
Fodor, S. P., Brody, L. C., Collins, F. S. (1998) Evolutionary sequence com-
parisons using high-density oligonucleotide arrays. Nat. Genet. 18, 155-158.
7. Humpherys, D., Eggan, K., Akutsu, H., Friedman, A., Hochedlinger, K.,
Yanagimachi, R., Lander, E. S., Golub, T. R., Jaenisch, R. (2002) Abnor-
mal gene expression in cloned mice derived from embryonic stem cell and
cumulus cell nuclei. Proc. Natl. Acad. Sci. U S A 99, 12889-94.
8. Lockhart, D. J., Dong, H., Byrne, M. C., Follettie, M. T., Gallo, M. V., Chee,
M. S., Mittmann, M., Wang, C., Kobayashi, M., Horton, H., and Brown, E.
L. (1996) Expression monitoring by hybridization to high-density oligonu-
cleotide arrays. Nat. Biotechnol. 14, 1675-1680.
9. Pastinen, T., Kurg, A., Metspalu, A.,Peltonen, L., Syvanen, A-C. (1997)
Minisequencing: a specific tool for DNA analysis and diagnostics on oligonu-
cleotide arrays. Genome Res. 7, 606-614.
24 DNA microarray data analysis

10. Schena, M. Shalon, D., Davis, R. W., and Brown, P. O. (1995) Quantitative
monitoring of gene expression patterns with a complementary DNA microar-
ray. Science 270, 467-470.

This chapter was written by Garry Wong.

2 Affymetrix Genechip system 25

2 Affymetrix Genechip
system
2.1 Affymetrix technology

There are two principal differences between the Affymetrix Genechip system and
“traditional” cDNA microarrays in studying gene expression. First, instead of
hybridizing two RNAs labeled with different fluorophores competitively on one
cDNA microarray, a single RNA is hybridized on the array in the Affymetrix sys-
tem, and the comparisons are then made computationally. Second, in the Affymetrix
arrays each gene is represented as a probe set of 10–25 oligonucleotide pairs instead
of one full length or partial cDNA clone. The oligonucleotide pair (probe pair)
comprises of one oligonucleotide perfectly matching to the gene sequence (Perfect
Match, PM) and a second oligonucleotide having one nucleotide mismatch in the
middle of it (Mismatch, MM) . Probes are designed within 500 base pairs of the 3’
end of each gene to hybridize uniquely in the same, predetermined hybridization
conditions. Some housekeeping genes are represented as three probe sets, one set
designed to the 5’ end of the gene, second set to the middle of the gene and the third
to the 3’end. In addition to species specific genes, some spiked-in control probe sets
are introduced to facilitate the controlling of the hybridization. In the experiment,
biotin-labeled RNA is hybridized to the array, which is stained with phycoerythrin-
conjugated streptavidin after washing and scanned with the Gene Array Scanner. A
grid is automatically laid over the array image and the intensities of each probe pair
are used to make expression measurements with the Affymetrix Microarray Suite ,
version 5 software.

2.2 Single Array analysis

This analysis generates qualitative and quantitative values from one gene expres-
sion experiment and provides initial data required to perform comparisons between
experiments. A quantitative value, a Detection call, indicates whether a transcript is
reliably detected (Present) or not detected (Absent) in the array. The Detection call
is determined by comparing the Detection p-value generated in the analysis against
user-defined cut-offs. A quantitative value, a signal, assigns a relative measure of
abundance to the transcript.
26 DNA microarray data analysis

2.3 Detection p-value

The detection algorithm uses a two-step procedure to determine the Detection p-
value. First it calculates Discrimination score (R) for each probe pair. The Dis-
crimination score measures the target-specific intensity difference of the probe pair
relative to its overall hybridization intensity:
PM − MM
R=
PM + MM
Second, the Discrimination score (R) is tested against the user-defined threshold
value Tau, which is a small positive number (by default 0.015). Those probe pairs
with the discrimination score higher than Tau vote for the presence of the transcript,
while the scores lower than Tau vote for the absence. The voting results of all probe
pairs are summarized as a p-value, which is generated by using the One-sided
Wilcoxon’s Signed Rank test as a statistical method. The higher the discrimination
scores are above Tau, the smaller the p-value and the more likely the transcript
is present. The lower the scores are below Tau, the larger the p-value, and more
likely the transcript is absent. Additionally, prior to the two-step Detection p-value
calculation, the level of photomultiplier saturation is evaluated for each probe pair.
If all probe pairs in a probe set are saturated, the corresponding transcript is given
a Present call. If the Mismatch (MM) probe is saturated, that probe pair is rejected
from further analysis. Tau can be used to modify the specificity and the sensitivity
of the assay: increasing the value of Tau increases the specificity, but lowers the
sensitivity, while decreasing Tau lowers the specificity and increases the sensitivity.

2.4 Detection call

A qualitative value, the Detection call is determined by comparing the Detection p-
value with the user-modifiable p-value cut-offs (alpha1 and alpha2, α1 = 0.04, α2
= 0.06 by default). Any p-value smaller than alpha1 is assigned a Present call, and
that larger than alpha2 an Absent call. Marginal calls are given to values between
alpha1 and alpha2. The p-value cut-offs can also be adjusted to modify specificity
and sensitivity.

2.5 Signal algorithm

Signal is a quantitative value calculated for each probe set and it represents the
relative level of expression of a corresponding transcript. Signal is calculated as a
weighted mean using the One-Step Tukey’s Biweight Estimate. The specific signal
for each probe pair is estimated by taking the log of the Perfect match intensity
after subtracting the stray signal estimate. It does not make physiological sense if
the intensity of the Mismatch probe cell is higher than the Perfect match intensity,
and therefore an imputed value called Change Threshold (CT) is used instead of the
uninformative Mismatch intensity. Three rules are employed to form a stray signal
estimate:
• If the Mismatch value is less than the Perfect match, the Mismatch value is
2 Affymetrix Genechip system 27

considered informative and the intensity is used directly as an estimate of

stray signal.

• If the Mismatch probe cells are generally informative across the probe set
except for a few Mismatches, an adjusted Mismatch value is used for unin-
formative Mismatches based on the bi-weight mean of the Perfect match and
the Mismatch ratio.

• If the Mismatch cells are generally uninformative, the uninformative Mis-

matches are replaced with a value slightly smaller than the Perfect match.
The transcripts represented by such probe sets are generally called Absent
by the Detection algorithm.

Each probe pair is considered as having a potential vote in determining the

Signal value. The probe pair is weighted more strongly if the signal of the probe
pair is closer to the median value for that probe set. Once the weight of each probe
pair is determined, the mean of the weighted intensity values for that probe set is
identified. This mean is the quantitative value Signal.

2.6 Analysis tips

For each experiment check that:

• Background is 40–70 (using current scanner settings).

• Scaling Factor is close to one (using current scanner settings and Target In-
tensity 100) or at least in the same level for those probe arrays you are plan-
ning to compare with each other.

• All positive hybridization controls (BioB, BioC, BioD, CreX) are present.

• Signal ratios of 3’, middle and 5’ probe sets (GAPDH, Beta-actin) are close
to one.

• Percentage of present calls is between 30-50

2.7 Comparison analysis

Comparison analysis provides a computational comparison between two RNA sam-

ples hybridized to two GeneChip probe arrays of a same type. For the analysis
one array is designated as an experiment and the other as the baseline and com-
pared with each other in order to detect and quantify differences in gene expression
between the two RNAs. Two sets of algorithms are used to generate qualitative
(Change, Change p-value) and quantitative (Signal Log Ratio) estimates of the po-
tential change.
28 DNA microarray data analysis

2.8 Normalization
Before a comparison of two probe arrays can be made, variations between the two
experiments caused by technical and biological factors must be corrected by scal-
ing or normalization. Main sources of technical variation in an array experiment
are quality and quantity of the labeled RNA hybridized as well as differences in
reagents, stain and chip handling. Biological variation, though irrelevant to study
question, may arise from differences in genetic background, growth conditions,
time, sex, age, etc. Either scaling or normalization should be used to minimize this
variation.
Scaling (recommended by Affymetrix) and normalization can be made using
either data from user-selected probe sets or all probe sets. When using selected
probe sets (for example a group of housekeeping genes) for scaling/normalization,
it is important to have a priori knowledge of gene expression profiles of selected
genes. Thus in most cases less bias is introduced to the experiment when the data
of all probe sets is used for scaling/normalization. When scaling is applied, the
intensity of the probe sets from experimental and baseline arrays is scaled to a same,
user-defined target intensity (recommendation using current scanner setting is 100).
If normalization is applied, the intensity of the probe sets from experimental array
is normalized to the intensity of the probe sets on the baseline array.
An additional normalization factor, Robust Normalization, is also introduced
to the data. It accounts for probe set characteristics resulting from sequence-related
factors, such as affinity of the probe set to the RNA and linearity of the hybridiza-
tion of each probe pair. More specifically, this factor corrects for the inevitable
error of using an average intensity of all the probes (or selected probes) on the
array as a normalization factor for every probe set. Robust normalization of the
probe set is calculated once the initial scaling/normalization factor is determined,
by calculating a slightly higher and a slightly lower factor than the original. User-
modified parameter, Perturbation, which can vary between 1.00 (no perturbation)
and 1.49, defines the range by which the normalization factor is adjusted up and
down. The perturbation value directly affects the subsequent p-value calculation,
since of the p-values resulting from applying the normalization factor and its two
perturbed variants, the most conservative is used to evaluate whether any change
in level is justified by the data. Increasing the perturbation value can reduce the
number of false changes, but may also block the detection of some true changes.

2.9 Change p-value

Differences between Perfect Match and Mismatch intensities as well as between
Perfect Match and background intensities are used to calculate the Change p-value
by the Wilcoxon’s signed rank test. Also the level of photomultiplier saturation
for each probe pair is evaluated. In the computation, any saturated probe cell is
rejected from the analysis (number of used cells can be determined from the Stat
common pairs column). The Change p-value ranges from 0.0 to 1.0 providing an
estimate of the likelihood and direction of the change. For values close to 0.5, true
change in the level of expression between the two RNA samples is unlikely, while
2 Affymetrix Genechip system 29

values close to 0.0 indicate probability for an increase in gene expression level in
the experiment array compared to baseline, whereas values close to 1.0 indicate
likelihood for a decrease in gene expression level.

2.10 Change call

To make the Change Call, Change p-value is categorized by cutoff values called
gamma1 (γ 1) which distinguishes between Change calls: Increase, Marginal In-
crease and No Change, and gamma2 (γ 2) which draws the line between Change
calls: Decrease, Marginal Decrease and No change (Figure 2.1). These values are
not directly user-definable, but are derived from two user-adjustable parameters,
γ 1L and γ 1H, which define the lower and upper boundaries for gamma1. Gamma
2 is computed as a linear interpolation of γ 2L and γ 2H.
It is possible to compensate for effects that influence calls based on low and
high signals by adjusting the stringency of calls associated with low and high signal
ranges (γ L and γ H) independently. This option is not used by default, since gamma
values are set as γ 1L = γ 1H and γ 2L = γ 2H.

Figure 2.1: A representation of p-values in a data set. The Y-axis is the probe set signal and
the X-axis is the p-values. Values γ 1L, γ 2L, γ 1H, γ 2H describe user-adjustable values
used in making the Change Call (I= Increased, MI = Marginal Increase, MD = Marginal
Decrease, D = Decrease).

2.11 Signal Log Ratio Algorithm

Signal Log Ratio algorithm estimates the measure and the direction of change of a
gene/transcript when two arrays are compared. Each probe pair on the experiment
array is compared to the corresponding probe pair in the baseline arrays in the cal-
culation of Signal Log Ratio. This process eliminates differences due to different
probe binding coefficients. A One-Step Tukey’s Biweight method is used in com-
puting the Signal Log Ratio value by taking a mean of the log ratios of probe pair
intensities across the two arrays. The base 2 log scale is used, translating the Signal
30 DNA microarray data analysis

Log Ratio of 1.0 to a 2-fold increase in the expression level and of -1.0 to a 2-fold
decrease. No change in the expression level is thus indicated by a Signal Log Ratio
value 0.
Tukey’s Biweight method also gives estimate of the amount of variation in the
data. Confidence intervals are generated from the scale of variation of the data.
A 95% confidence interval shows a range of values, which will include the true
value 95% of the time. Small confidence interval implies that the expression data
is more exact, while large confidence intervals reflect more noise and uncertainty
in estimating the true level. Since the confidence intervals attached to Signal Log
Ratios are computed from variation between probes, they may not reflect the full
width of experimental variation.

This chapter was written by Janna Saarela.

3 Genotyping systems 31

3 Genotyping systems
3.1 Introduction

SNP (single nucleotide polymorphism) microarrays provide an efficient and rela-

tively inexpensive tool for studying several genetic variations in multiple samples
simultaneously. Numerous methods have been developed for SNP genotyping and
two of them, namely single base extension (SBE) followed by tag-array hybridiza-
tion and allele-specific primer extension, are shortly described here.

3.2 Methodologies

For the single base extension reaction, multiple genomic DNA-regions flanking
SNP sites are amplified by PCR using SNP-specific primers (Figure 3.1:A). Af-
ter enzymatic removal of the excess primers and nucleotides single base extension
reactions are carried out with detection primers each containing a sequence comple-
mentary to the predefined TAG sequence spotted on the array and the SNP-specific
sequence just preceding the variation. One allele-defining nucleotide, each labeled
with different fluorescent dye, is added to the detection primer in the SBE reaction.
Finally, detection primers are hybridized to the TAG sequences spotted on an array.
Genotypes are determined by comparing the signals of two possible nucleotides
labeled with different fluorescent dyes on each spot.
For the second method, allele-specific primer extension, two amino-modified
detection primers, each containing one of the variable nucleotides of the SNP as
their 3’ nucleotide, are spotted and covalently linked to chemically activated mi-
croscope slides (Figure 3.1:B). Genomic DNA flanking the SNP is amplified with
SNP-specific primers containing T7 or T3 RNA polymerase recognition sequence
in their 5’ end. Multiple SNPs can be amplified in one multiplex PCR reaction
simultaneously. By using T7 (or T3) RNA polymerase, PCR products are subse-
quently transcribed to RNA, which is then hybridized to the SNP array containing
the two detection primers for each SNP. Reverse transcriptase enzyme and fluo-
rescently labeled nucleotides are then employed to visualize the sequence-specific
extension reaction of the allele-defining detection primer(s). Up to 80 different
samples can be analyzed on one slide when as many identical subarrays of de-
tection primers are spotted on a slide, which is partitioned into small hybridization
chambers. Genotypes are determined by comparing the signals of the two detection
primers representing the SNP on the array.
32 DNA microarray data analysis

Figure 3.1: Principles of the two SNP genotyping methods. A. Single base primer exten-
sion followed by TAG-array hybridization. B. Allele-specific primer extension on array.
Traffic lights at the bottom of the figures describe three possible genotypes with each
method.

3.3 Genotype calls

To make the genotype call, signals of the two possible variants are compared. In
case of the SBE reaction, intensities of the two different fluorophores in a spot
representing the SNP are measured. In case of the allele-specific primer extension
method, each SNP is represented by two spots, both putatively labeled (with the
same fluorophore). The intensities of the two spots are measured and compared
to each other to determine the genotype. The same methods can be used for both
comparisons. One method is to calculate the part of the signal of one allele over
the total signal intensity:

signal of allele 1
P=
(signal of allele 1 + signal of allele 2)

All the values are between zero and one (Figure 3.2). Values close to one
represent the homozygote allele 1 genotype, and those close to zero represent the
other homozygote genotypes (allele 2). Those in-between (close to 0.5) represent
the heterozygote genotypes. A scatter plot can be formed having the values for P
in the X-axis and logarithm of the total intensity as the Y-axis. Three clear groups
of signals should show clearly, each representing one of the possible genotypes.
3 Genotyping systems 33

Figure 3.2: An example of genotype calls. TT and CC denote homozygous individual,

and TC denotes heterozygous individuals.

3.4 Suggested reading

1. Syvanen, A-C. (1994) Detection of point mutations in human genes by the
solid-phase minisequencing method. Clin. Chim. Acta. 226, 225-36.

2. Guo, Z., Guilfoyle, R. A., Thiel, A. J., Wang, R., and Smith, L. M. (1994)
Direct fluorescence analysis of genetic polymorphisms by hybridization with
oligonucleotide arrays on glass supports. Nucleic Acids Res. 22, 5456-5465.

3. Pastinen, T., Raitio, M., Lindroos, K., Tainola, P., Peltonen, L., Syvanen,
A-C. (2000) A system for specific, high-throughput genotyping by allele-
specific primer extension on microarrays. Genome Res. 10,1031-42.

4. Hirschhorn, J. N., Sklar, P., Lindblad-Toh, K., Lim, Y. M., Ruiz-Gutierrez,

M., Bolk, S., Langhorst, B., Schaffner, S., Winchester, E., Lander E. S.
(2000) SBE-TAGS: an array-based method for efficient single-nucleotide
polymorphism genotyping. Proc. Natl. Acad. Sci. U S A 97, 12164-9.

This chapter was written by Janna Saarela.

34 DNA microarray data analysis

4 Overview of data analysis

In this book, we emphasize to microarray data analysis after the microarrays have
been hybridized, scanned and the images have been analyzed with an image anal-
ysis software. Before any experiments in the laboratory have been initiated, the
experiment and its analysis should be planned carefully. Chapter 5 describes points
that need to be considered when designing the experiments. The following schematic
workflow outlines the basic analysis steps of exploratory data analysis.

4.1 cDNA microarray data analysis

We start data analysis from the results of scanned images. At this point, images
have been evaluated, bad spots have been investigated and the spots have preferably
been scored with flags indicating whether the spot was good, bad, or borderline.
This is crucial, because in the later stages of the analysis the visual inspection of
individual spots is not possible.
Next steps in the analysis are preprocessing (Chapter 7), normalization (Chap-
ter 8), and quality control (Chapters 6, 7, and 8). The goal of these analyses is to
organize results in a meaningful order: flags, controls, and experiments are pointed
out and checked. Variation due to systematic errors are removed, and data from
different chips is made comparable.
Statistics is needed in many steps during the analysis. Therefore, the whole
Chapter 6 has been dedicated to basics of statistics. Statistical tools are used for
evaluation of the raw data during the above-mentioned steps, and in the further
analysis to find significantly differentially expressed genes.
In these further analysis steps, statistically significant, quality-checked data is
separated from not interesting and not-trustworthy data (Chapter 7). The next step
is to find the differentially expressed genes using statistical tools (Chapter 9), or to
group the good quality data (usually only a small fraction of the original raw data)
into meaningful clusters by e.g. clustering (Chapter 10). The goal of clustering
is to find similarly behaving genes or patterns related to time scale, time point,
developmental phase or treatment of the sample.
At this point, we have already manipulated our data quite a lot, and spent con-
siderable time with computer. Despite that and depending on what we are looking
for, we may be at the very beginning of the challenging part of the data analysis.
Next we need to link the observations to biological data, to regulation of genes,
and to annotations of functions and biological processes. This part, data mining, is
described in Chapters 11, 12 and 13.
4 Overview of data analysis 35

With an enormous amount of data, we need standardized systems and tools for
data management in order to publish the results in a proper and sound way, as well
as to be able to benefit from other publicly available gene expression data. These
aspects are discussed in Chapter 14. Data file manipulation and analysis tools are
introduced in Chapter 15.

4.2 Affymetrix data analysis

Putting model-based methods aside, exploratory data analysis using Affymetrix
chips is very similar to cDNA microarray data analysis. The biggest difference
is normalization. If comparison analysis (see 2.7) is conducted, the Affymetrix
data can be treated similarly to a background corrected cDNA data. If single array
analysis is performed, the basic normalization scheme can be similar to the one
presented in section 8.9.

4.3 Data analysis pipeline

To get an overview of the data analysis pipeline, consult Figure 4.1. It covers the
basic methods introduced in this book. The flow chart also helps to choose the right
method for the situation. The Table 4.1 contains a short list of analysis methods,
and links to the chapters, where more details are available.
The flow chart should be read flexibly, though. For example, there can be
several filtering steps instead of just one shown on the chart. Or, when there are
more than two conditions in the experiment, the data can be analyzed using two
conditions route. Note that all possible orders of analysis have not been shown for
clarity, and the schema is only meant to help you in the analysis process, not to be
taken as an absolute truth.
There are different opinions on which order the steps should be taken, and
which portion of data should be transferred to the next step, so please, study the
relevant literature to find out more about limitations and characteristics of different
analysis steps. We would very much like to encourage you to make your own
educated decisions, and not to take this book with a face value!

This chapter was written by M. Minna Laine and Jarno Tuimala.

36 DNA microarray data analysis

Figure 4.1: Overview of data analysis methods discussed in this book. Note that all
possible orders of analysis have not been shown.
4 Overview of data analysis 37

Table 4.1: Chapters for methods shown in Figure 4.1.

Method Chapter

Experimental design 5
Basic statistics 6
Background correction 7
Calculation of expression change 7
Quality checking 7
Filtering 7
Normalization 8
Single slide methods 9
t-test 9
ANOVA 6
Clustering 10
Classification 10
PCA 10
Function prediction 10
Gene regulatory networks 11
Promoter sequence analysis 12
Annotations 12, 13
Ontologies 12, 13
Article mining 13
Data management 14
MIAME standard 14
MAGE standard 14
38 DNA microarray data analysis

5 Experimental design
5.1 Why do we need to consider experimental design?

Good experimental design, above the myriad of other considerations, will likely
provide you with the greatest amount of satisfaction and least amount of frustra-
tion in executing a microarray project. In the early days of microarray research, a
simple design of control v. treatment and replicates of each would suffice. How-
ever, those days are rapidly going by the wayside and 30–80 chip experiments in
a single design are becoming common. While the design depends primarily on
the scientific question that is being proposed (hypothesis), other considerations,
such as appropriate controls, platforms, and statistical issues merit serious thought.
This is to say nothing about costs! While many projects have similar objectives, it
would be foolish for us to design protocols in a guidebook without intimate knowl-
edge of the system and the intricacies involved in the study. Moreover, many fields
have specialized and/or traditional methodologies that require non-typical designs.
Nonetheless, there are several variables in the experimental design that are common
to most studies, and we will try to describe them here. We hope that the consider-
ation of the strengths and weaknesses of each variable on both the biological and
statistical level would help the individual investigator better design his/her own mi-
croarry study. Most of the discussion concerns expression arrays, although places
where SNP arrays are relevant are pointed out.

5.2 Choosing and using controls

The control serves as a reference for an experiment commonly termed the treat-
ment. The treatment can be chemical (drugs, toxins), biological (viruses, microbes,
transgenes) or environmental (stress, irradiation) in nature. The individual treat-
ments can be given in a time series (time course) or at different doses (dose re-
sponse). In all these cases, the control should be matched as closely as possible
genetically. This can mean that the controls are sibs or that the animal used is an
inbred strain, or a combination of the two. Controls with the same environmental
influences are usually dealt with by using litter siblings that have been raised iden-
tically. Physiological matching can be done by taking the same sex, age, and health
status.
Controls for human tissues are problematic since age, cause of death, and
post-mortum interval are difficult to match. Moreover, human tissues for research
purposes are often stored for unequal intervals, which greatly effect RNA quality.
5 Experimental design 39

Because of this potential confounder, many human studies have been done using
human cell lines rather than tissues. Another solution is to match tissue by taking
different regions from the same human tissue, one that is healthy to serve as the
control, and one that carries the disease as the treatment sample.
Controls for mice studies have a different problem in that some transgenic
mice are crosses between xxx and yyy strains. Therefore, both transgenic and non-
transgenic littermates may have quite different genetic backgrounds. The solution
would be to backcross to one of the parental strains until both the control and trans-
genic mouse have the same genetic background. This is somewhat time consuming
since mice have a reproductive cycle of 2–3 months. Another solution would be
to make sure that the transgenic lines that are produced have a homogeneous back-
ground.
Controls for cell based experiments generally consist of identical cultures
without the physiological, physical, or chemical treatments. The controls may also
include cells derived from other sources such as the equivalent or healthy tissues.
When cells are cocultured, the controls become more complicated. Cultures of
each of the individual constituent cell populations in addition to a coculture could
be used as a control. In practice this means that if a coculture of A and B cells is
made, then the controls will be A alone, B alone, A and B together.
Because of all these potential variables that could effect the microarray results,
it often makes sense to create a design with more than one control within the study.

5.3 Choosing and using replicates

Replicates are repeated experiments with the same sample (Table 5.1). Replicates
provide a measure of the experimental variation, such as in RNA isolation, labeling
efficiency, or in chip quality. It is highly recommended to include replicates in your
design, otherwise you will not know the inherent experimental variation. Dupli-
cates are common practice and triplicates provide even more information. Some
designs incorporate combinations of duplicates and triplicates. In some cases, it
may not be practical to perform replicates due to lack of tissue or other reason.
In these cases, replicated spots on the chip can provide some idea of the varia-
tion in hybridization. Including a large number of both controls and treatments
without replicates is another design strategy that has been used. Confirmation of
results from non-replicated experimental designs using independent methods such
as quantitative PCR and northern blotting is also one way to circumvent the need
for replicates.

5.4 Choosing a technology platform

For expression arrays, there are two principle platforms: spotted arrays and in situ
synthesized arrays such as Affymetrix. The basic principles have been presented
in the introduction and detailed aspects are covered later in the book. Apart from
the experimental design issues presented here, the Affymetrix platform uses one
labeling reaction per chip, so a control reaction is performed and hybridized and
then the treatment reaction is performed on a different chip. The results are then
40 DNA microarray data analysis

Table 5.1: Typical experimental designs and number of microarrays needed.

Experimental design Samples Replicates Cy3/Cy5 Affymetrix

Control v. Treated 1 1 1 2
Control v. Treated 1 3 3 6
Control v. Treated 5 2 10 12
Control v. Treatments (4) 1 1 4 5
Control (2) v. Treatments (2) 1 1 4 4
Control (2) v. Treatments (2) vs. 1 1 16 16
time points (4)
Control (2) v. Treatments (2) vs. 1 3 48 48
time points (4)
Control (2) v. Treatments (2) vs. 4 2 64 96-112
time points (4)
Controls (1) v. Treatments (2) vs. 6 2 96 144
time points (4)

compared after having been normalized by independent spiked standards. This is

in contrast to expression arrays where there is a direct comparison between control
and treated. While not completely obvious, the advantage of the Affymetrix plat-
form is that multiple experiments can be compared directly if they are performed
on the same chip series. Thus, many of the steps that can cause experimental vari-
ation are controlled. The labeling, hybridization, and readout is done on identical
instruments with identical protocols on identically designed chips, so other than
normalization to an external standard, all experiments performed on the same chip
series should be comparable. This adds tremendous power to the experimental de-
sign. Since control and treatment are hybridized on different Affymetrix chips,
the design requires twice as many chips compared to expression arrays. However,
controls can be compared with all other samples so there is a small benefit in this
design if many controls are needed. Moreover, comparison to publicly available
data can be performed. Affymetrix platform has the added advantage of multiple
probes for each gene, so in a sense, you get 16–25 measurements for each gene
rather than the 1 or 2 from typical expression arrays. This is especially useful if
you wish to hybridize to genes with degenerated sequences such as splice forms,
highly polymorphic, or related species. The selection of gene chips available from
Affymetrix is more limited than currently available expression arrays, but contains
the major species used in scientific research including Saccharomyces cerevisea
(yeast), Arabidopsis Thaliana (mustard weed), Caenorhabditis elegans (worm),
Mus musculus (mouse), and Homo Sapiens (human).

5.5 Gene clustering v. gene classification

It often makes more sense to identify genes and their related expression levels that
predict the experimental group they are derived from (gene classification) than to
find groups of genes that act in a particular way after treatment (gene clustering).
5 Experimental design 41

For classification studies, large numbers of samples for each treatment groups are
needed in order to robustly find a set of genes that predicts their original treatment
group. The large number of samples is also necessary to validate the predicted clas-
sification based on training sets derived from the data. In this case of using gene
expression to predict human tumor type as a diagnostic tool, individual variation
from human subjects, and variation in the samples due to which part of the tumor
was dissected, how the tissue was obtained, and the other experimental variables
typically suggests that random variation in the profiling is unavoidable, further in-
dicating that sample sizes should be large.
In contrast, the number of genes profiled need not be large if the genes consis-
tently predict the classification group. This is unfortunately not the usual case.

5.6 Conclusions
The experimentalist is likely to know the most about the scientific question and
therefore have the greatest input into the experimental design. Nonetheless, con-
sultations with a statistical expert prior to experiments could help to decrease exper-
imental variation derived from the limitations listed above. Recently, two excellent
review articles have become available and are listed at the end of this chapter. In
the end, there are many factors influencing the successful outcome of a microarray
study, and attention to as many as possible will greatly aid in the final outcome.

5.7 Suggested reading

1. Churchill, G. A. (2002) Fundamentals of experimental design for cDNA mi-
croarrays. Nature Genetics supplement 32, 490-495.

2. Yang, Y. H. (2002) Design issues for cDNA microarray experiments. Nature

Reviews, 3, 579-588.

This chapter was written by Garry Wong.

42 DNA microarray data analysis

6 Basic statistics
6.1 Why statistics are needed
One important use of statistics is to summarize a collection of data in a clear and un-
derstandable way. There are two basic methods: numerical and graphical. Graphi-
cal methods are best at identifying patterns in the data. Numerical approaches are
more precise and objective. Since numerical and graphical approaches complement
each other, both are needed.

6.2 Basic concepts

6.2.1 Variables
A variable is any measured characteristic that differs between subjects. For exam-
ple, if the weight of 50 individuals were measured, then weight would be a variable.
The numerical values of weight for the subjects are called measurements, scores,
data points or observations.
Variables can be quantitative or qualitative. Quantitative variables describe
certain quantities of subjects, and can be measured on a continuous or discrete
scale. On the continuous scale, there is an infinite number of values variables can
take. Discrete variables can take on only a limited number of values. For example,
weight is measured on a continuous scale, and school grades on a discrete scale.
Qualitative variables describe the attributes of the subjects, for example, hair
colour. They are always measured on a discrete scale.
When an experiment is conducted, some variables are manipulated by the ex-
perimenter, and others are measured from subjects. The former are factors or inde-
pendent variables, the latter, dependent variables.

6.2.2 Constants
Measures of subjects, i.e. variables, which can take on only one value during the
experiment are called constants. For example, if only women are included in a
study, then the variable sex is constant during the study.

6.2.3 Distribution
Distribution describes the scores and the number of scores the variable can take.
A continuous variable can be described with a histogram, which is a graphical
representation of the distribution (Figure 6.5). Distributions are more thoroughly
6 Basic statistics 43

covered in the subsequent sections.

6.2.4 Errors

Errors are inaccuracies in measurements. In the laboratory analyses, there are often
an innumerable amount of potential sources of errors, but only a few are usually of
paramount importance.
Errors can be systematic (biases) or random. Systematic errors affect the mea-
surements such a way that you get a wrong estimate of whatever you are measuring.
Systematic errors affect all the subjects similarly. Random errors lack such predic-
tivity. For example, one laboratory technician can be biased, if his/her analysis
results are always consistently higher than the results acquired by other techni-
cians. If the same technician drops the tube so that the liquid is spilled on the floor,
but he/she then slurps the liquid back up and transfers it to the tube with a pipet
(a procedure known as linoleum blot), a random error has been created. If all the
tubes are always cast on the floor, that would systematically bias the results.

6.3 Simple statistics

6.3.1 Number of subjects

Number of subjects (N) or sample size both describe the same thing: How many
subjects there are included in the study. Often the sample size is fixed before the
experiment is conducted. For example, if 1 000 genes are spotted on the DNA
microarray, then the sample size would be 1 000 genes.

6.3.2 Mean (m)

The arithmetic mean of a variable is calculated as the sum of all measurements

divided by the number of measurements. The mean is a good measure of central
tendency for roughly symmetric distributions, but it can be misleading in skewed
distributions due to probable influence of extreme scores. For skewed distributions,
the trimmed mean or median usually gives more meaningful results.

6.3.3 Trimmed mean

A trimmed mean is calculated by discarding a certain percentage of the lowest

and the highest scores and then computing the mean of the remaining scores. A
trimmed mean is obviously less susceptible to the effects of extreme scores than the
arithmetic mean, but it is also a less efficient descriptor than the mean for normal
distributions.

6.3.4 Median

The median is the middle of a distribution: half the scores are above the median and
half are below. The median is less sensitive to extreme scores than the mean and
this makes it a better measure of central tendency for highly skewed distributions.
44 DNA microarray data analysis

6.3.5 Percentile
The percentile is a certain cut-off, under which the specified percentage of the ob-
servations lie. The median is 50th the percentile of the distribution. Similarly, the
80th percentile is a point of the distribution under which 80% of the observations
lie.

6.3.6 Range
The range is the difference between the largest and the smallest value of the dis-
tribution. It is the simplest measure of spread, but it is very sensitive to extreme
scores, because it is based only on two values.

6.3.7 Variance and the standard deviation

The variance (d) and standard deviation (s or sd) are both measures of spread. The
variance is calculated as the averaged squared deviation of observations from their
mean. For example, for the scores 1, 2, and 3, the mean is two, and the variance
is: [(1 − 2)2 + (2 − 2)2 + (3 − 2)2]/3 = 0.667. The standard deviation is the square
√
root of the variance: (0.667) = 0.827. Standard deviation is often more easily
interpreted than variance, because its unit is the same as the unit of measuments.

6.3.8 Coefficient of variation

The coefficient of variation (CV) describes the relative variability of the data. It is
especially good in situations, where distributions of unequal magnitude are com-
pared. For example, the absolute variability measured in kilograms is very small
in a mouse population, but very large in an elephant population. In such cases, the
coefficient of variation gives a better description of the variability than the standard
deviation, because we are usually interested in how the variability is related to the
mean of the species, and because the CV is independent of the absolute values of
the variable. The coefficient of variation is calculated as
s
CV = ,
m
where s is the standard deviation and m is the mean.
Often the CV is reported as a percentage. In biological research, it is a widely-
used measure of reliability for replicated experiments.

6.4 Effect statistics

6.4.1 Scatter plot
The Scatter plot is an important graphical tool for studying the spread and linear-
ity of the data. In its simplest form, two variables are plotted along the axes, and
marks are drawn according to these coordinates. For example, green and red chan-
nel intensities of DNA microarray experiments can be depicted as a scatter plot
(Figure 6.1).
6 Basic statistics 45

6.4.2 Correlation (r)

The correlation of two variables represents the degree to which the variables are
related. When two variables are perfectly linearly related, the points in the scatter
plot fall on a straight line. Correlation measures only linear relationship. Two sum-
mary measures or correlation coefficients, Pearson’s correlation and Spearmans’s
rho, are most commonly used. Both of these measure range from perfectly positive
linear relationship to perfectly negative linear relationship, or from -1 to 1.
It is not wrong to calculate the correlation between variables, which are not
linearly related, but it does not make much sense. If the variables are not linearly
related, the correlation does not describe their relationships effectively, and no con-
clusions can be based on the correlation coefficient only.
Correlation and scatter plot are a good example of how numerical and graphi-
cal tools effectively complement each other.

correlation = 0.962
30000
Red channel intensity

20000
10000
0

0 10000 20000 30000 40000

Green channel intensity

Figure 6.1: An example of a scatter plot. The red and green channel intensities from
a two-color DNA microarray experiment have been depicted as a scatter plot. Variables
seem to have a linear relationship, because a straight line can be drawn through the data
points. In addition, the linear correlation between these variables is 0.962.
46 DNA microarray data analysis

6.4.3 Linear regression

Linear regression is used for describing how much the dependent variable (the pre-
dicted variable) changes if the independent variable (the predictor variable) changes
a certain bit. It would be wrong to apply linear regression to non-linear data. There-
fore, before applying linear regression, the linearity of the data should be checked
from a scatter plot (Figure 6.1). In addition, the normality of the dependent variable
should be checked before running the analysis (see the following sections).
Linear regression fits a straight line through the data points so that the square
sum deviation of the line from the data points is minimized. The result of the
procedure is a regression equation

Y = a X + b,

where b indicates where the regression line cuts the vertical axis. If only two vari-
ables are used in the linear regression model, a is equal to the pearson correlation,
and indicates the slope of the regression line. It also indicates by which number
change in X (independent variable) must be multiplied to give the corresponding
change in Y (dependent variable).

correlation = 0.962
30000
Red channel intensity

20000
10000
0

0 10000 20000 30000 40000

Green channel intensity

Figure 6.2: The two linear regression lines fitted into two-color DNA microarray results.
The results are different if green channel intensity (green line) is used as a dependent
variable instead of red channel intensity (red line).
6 Basic statistics 47

Correlation is completely symmetric. The correlation between A and B is

the same as correlation between B and A. Linear regression does not have this
symmetric property. For linear regression it is absolutely vital to know which is the
dependent variable, in which we want to predict the change. This should be decided
when the experiment is planned. For example, we could try to predict how many
papers we could print with a certain cartridge and machine, but it would not be
possible to predict which cartridge and machine we have, if we know that a certain
number of papers have been printed. For microarray experiments, linear regression
is sometimes used for normalization. This is generally not a good idea, because it
is not possible to decide which labeled sample (red or green) should be used as an
independent and/or dependent variable (Figure 6.2).

Normal density Normal density

0.0 0.2 0.4 0.6 0.8

0.0 0.2 0.4 0.6 0.8

density

density

−4 −2 0 2 4 0 2 4 6 8 10

x x
mean=0, sd=1 mean=5, sd=1

Normal density Normal density

0.0 0.2 0.4 0.6 0.8

0.0 0.2 0.4 0.6 0.8

density

density

−4 −2 0 2 4 −4 −2 0 2 4

x x
mean=0, sd=2.5 mean=0, sd=0.5

Figure 6.3: Normal distributions of different shapes. The mean defines the place of
the distribution along the horizontal axis, and standard deviation (sd) defines the shape
of the curve. Small standard deviation means tight, and large standard deviation a flat
distribution.

6.5 Frequency distributions

6.5.1 Normal distribution
Normal distributions are a family of distributions that have the same general shape.
They are symmetric around their mean with more measurements in the middle than
48 DNA microarray data analysis

in the tails. Normal distributions are sometimes described as bell shaped. The
shape of the normal distribution can be specified mathematically in terms of the
mean and standard deviation (Figure 6.3).
A standard normal distribution is a normal distribution with a mean of 0 and
a standard deviation of 1. Any normal distribution can be transformed to standard
normal distribution by the formula
(X − m)
Z= ,
s
where X is a score from the original normal distribution, m is the mean of the
original normal distribution, and s is the standard deviation of the original normal
distribution. This mathematical procedure is also called standardization.
There are biological and historical reasons for the widespread usage of normal
distribution: Many biologically relevant variables are distributed approximately
normally. Mathematical statisticians also work comfortably with normal distri-
butions, and many kinds of statistical tests can be and are derived from normal
distributions. Some of these tests will be described in the next chapters. We have
already had a peek at one application of normal distribution, linear regression.

Normal density t density

0.0 0.1 0.2 0.3 0.4

0.0 0.1 0.2 0.3 0.4

density

density

−4 −2 0 2 4 −4 −2 0 2 4

x x
mean=0, sd=1 mean=0, sd=1, df=1

t density t density
0.0 0.1 0.2 0.3 0.4

0.0 0.1 0.2 0.3 0.4

density

density

−4 −2 0 2 4 −4 −2 0 2 4

x x
mean=0, sd=1, df=10 mean=0, sd=1, df=100

Figure 6.4: The mean of the t-distribution defines its place on x-axis. The shape of
the distribution is defined by standard deviation (s) and degrees of freedom (df). Stan-
dard deviation defines the tightness of the distribution, whereas df defines the amount of
resemblanse to normal distribution.
6 Basic statistics 49

6.5.2 t-distribution

The t-distribution resembles normal distribution, and in mathematical terms, it is

an approximation of the normal distribution for small samples sizes. T-distribution
differs from the normal distribution, because it has an additional parameter, degrees
of freedom (df), which affects its shape. Degrees of freedom reflect the sample size.
Degrees of freedom can take on any real number greater than zero. A t-
distribution with a smaller df has more area in the tails than the distribution with a
larger df. As the df increases, the t distribution approaches the standard normal dis-
tribution. For practical purposes, the t-distribution approaches the standard normal
distribution relatively quickly, so that when df = 50, the two are virtually identical
(Figure 6.4).

Histogram of gmean
1200
1000
800
Frequency

600
400
200
0

0 10000 20000 30000 40000

gmean

Figure 6.5: A typical histogram of the DNA microarray intensity values of one channel
(gmean). The distribution is highly skewed to the right. Median, mean and 68% standard
deviation are depicted as red, green, and blue vertical lines, respectively.

6.5.3 Skewed distribution

A distribution is skewed if one of its tails is longer than the other. Distributions
having a longer tail on the right are positively skewed, and distributions having a
longer tail on the left are negatively skewed. The best way to identify a skewed
distribution is to draw a histogram of the distribution (Figure 6.5). Numerically,
50 DNA microarray data analysis

a skewed distribution can be identified by comparing the mean and median of the
distribution. If the mean is larger than median, the distribution is skewed to the
right. For left skewed distributions the mean is lower than median.

6.5.4 Checking the distribution of the data

Many statistical tests assume that the data is normally distributed, which means
that the histogram drawn from the values of the variable resembles the normal
distribution. Even the most basic descriptive statistics can be misleading if the
distribution is highly skewed. For example, standard deviation does nor bear a
meaningful interpretation if the distribution significantly deviates from normality
(Figure 6.5).
Normality of the data is most easily checked from an appropriate histogram . If
the distribution is, as judged by eye, approximately symmetric and does not contain
more than one peak, it can be assumed to be normally distributed (Figure 6.5 and
Figure 7.6). Other graphical possibilities include a density plot, which are basically
just a smoothed histogram (Figure 6.6).

Normal density
2.0
1.5
density

1.0
0.5
0.0

−4 −2 0 2 4

intensity
mean=0, sd=1

Figure 6.6: Some real data compared to a normal distribution. Black, normal distribu-
tion; red, non-transformed data; green, log-transformed data. The log-transformation was
performed, because it often makes highly skewed distribution more normally distributed.
The values have been standardized before plotting.
6 Basic statistics 51

A more formal way to test for the normality of the data is the normal prob-
ability plot. The sample values and the theoretical values assuming normality are
plotted against each other in the normal probability plot. If the points fall on the
line, the distribution is normal (Figure 6.7).
Both the histogram and normal probability plot are valid methods for checking
the normality of the data, but the plot can detect much smaller deviations from
normality than histogram.

Normal Q−Q Plot Normal Q−Q Plot

40000

15
Sample Quantiles

Sample Quantiles

13
20000

11
9
0

−4 −2 0 2 4 −4 −2 0 2 4

Theoretical Quantiles Theoretical Quantiles

log−transformed raw data log−transformed raw data, no NAs

Normal Q−Q Plot Normal Q−Q Plot

3
6
Sample Quantiles

Sample Quantiles

2
4

1
2

0
−1
0

−4 −2 0 2 4 −4 −2 0 2 4

Theoretical Quantiles Theoretical Quantiles

standardized raw data standardized log−transformed data

Figure 6.7: The normal probability plots. The raw data has been pushed through different
kinds of transformations, and the results are checked for normality. The standardized log-
transformed data was drawn with green in the Figure 6.6. According to the plots, none of
the data sets is normally distributed, because the datapoints do not fall on a straight line.

6.6 Transformation

Transformation means a mathematical procedure, where a new variable is con-

structed or derived from the original one by applying a certain function or formula.
Data is often transformed, because the original data does not fulfill the distribu-
tional presumptions. For example, many statistical procedures assume that the data
is normally distributed. If this is not the case, transformation can be applied in such
a way that the distribution becomes more normal-like.
52 DNA microarray data analysis

6.6.1 Log 2 -transformation

Log2 -transformation is often used with DNA microarray experiments. Usually, the
intensity ratio is log2 -transformed. The resulting new varible is called log ratio.
The increase of one in the log ratio means that the actual intensity or expression
has doubled.

correlation=0.962
30000
Red channel intensity

20000
10000
0

0 10000 20000 30000 40000

Green channel intensity

Figure 6.8: A scatter plot where one probable outlier has been marked with a red dot.

6.7 Outliers
Outliers are by definition atypical or infrequent observations, data points which do
not appear to follow the characteristic distribution of the rest of the data. These
may reflect the properties of the variable, or be due to measurement errors or other
anomalies which should not be included in the analyses.
Typically, we believe that outliers represent random errors, which we would
like to control or get rid of. Outliers can have many undesirable properties: They
often attract the linear regression line, which might lead to wrong conclusions.
They can also artificially increase the correlation coefficient or decrease the value
of the legitimate correlation. The measure of spread, range, is unreliable if the data
includes outliers.
Outliers are often excluded from the data after the superficial analysis with
quantitative methods. For example, observations that are outside the range of 2
6 Basic statistics 53

standard deviations from the mean, can be discarded. However, definition of an

outlier is subjective, and in principle the decisions should be made on individual
basis for every suspicious observation.
There are two graphical methods with which to identify outliers, scatter plot
and box plot (Figure 6.8 and Figure 6.9).

Typical box plot

Maximum value
40000
30000
Green channel intensity

20000
10000

75% quantile Mean

25% quantile Median

Minimum value
0

Figure 6.9: An example of a box plot. Outliers are often indicated with individual marks
on the top of the boxplot.

6.8 Missing values and imputation

The missing values can sometimes markedly change the results, if only mean or
median values are inspected. Take, for example, the following experiment. We
want to find the genes, which can be used for dividing patients into two cancer
subtypes, those responding and not responding to the treatment. We measure the
expression values (log ratios) in the these subtypes, and plot the results.
Responsive group

No. of obs. 20 3 4 2 1
|------|-------------------------------|
Expression 0.0 mean 5.0 10.0
54 DNA microarray data analysis

Non-responsive group

No. of obs. 0 3 4 2 1
|----------------------|---------------|
Expression 0.0 5.0 mean 10.0

In the example, the missing values influence the mean very much. In a sense,
missing values draw the mean towards them. Even the median would not have
helped us here, because it would have scored 0 for the responsive group!
Missing values are usually either removed from the analysis or estimated using
the rest of the data in a process called imputation. These methods are covered in
more detail in the next chapter.

6.9 Statistical testing

Statistical testing tries to answer the question: “Can the difference between these
observations be explained by chance alone?” or “How significant is this differ-
ence?”. Statistical testing can also be viewed as hypothesis testing, where two
different hypotheses are compared. For example, we can test whether cancer pa-
tients and their healthy controls differ statistically significantly for the expression
of the gene X.

6.9.1 Basics of statistical testing

A suitable statistical test can give us an answer to the questions above. In practise,
statistical testing is divided into several substeps:

1. Select an appropriate statistical test.

2. Select a threshold for p-value.

3. Form the pair of hypotheses you want to compare.

4. Calculate the test statistic.

5. Calculate degrees of freedom.

6. Compare the test statistic to the critical values table of the test statistic distri-
bution.

7. Find out a p-value, which corresponds to the test statistic.

8. Draw conclusions.

These substeps are covered in more detail in the following subsections.

6 Basic statistics 55

6.9.2 Choosing a test

There are at least a couple of hundred of different statistical tests available, but only
a few of those are covered here in more detail. Usually the means of certain groups
are compared in the microarray experiments. For example, we can test whether the
expression of a certain gene is higher in the cancer patients than in their healthy
controls. The tests introduced here compare the means of two or several groups
with each other.
When choosing a test, there are two essential questions, which need to be
answered: Is there more than two groups to compare, and should we assume that
the data is normally distributed?
If two groups are compared, there are two applicable tests, the t-test and Mann-
Whitney U test. If more than two groups are compared, an analysis of variance
(ANOVA) or a Kruskal-Wallis test is used. The ANOVA procedure is covered in
more detail in section 6.10 of this book. Note, that if the ANOVA procedure is
applied to two group means only, it will produce the same results as the t-tests
introduced in the next sections.
If the data is normally distributed (see the tests for this in section 6.5.4), the
t-test for two groups, or the ANOVA for multiple groups can be used for compar-
isons. If the data is not normally distributed, and each group has at least five obser-
vations, the Mann-Whitney U test or Kruskal-Wallis test can be applied. However,
if less than five observations are available per group, it is better to use the t-test or
ANOVA.
Tests, which assume that the data is normally distributed, are called parametric
tests. Non-parametric tests do not make the normality assumption.

6.9.3 Threshold for p-value

The p-value is usually associated with a statistical test, and it is the risk that we
reject the null hypothesis (see section 6.9.4), when it actually is true. Before test-
ing, a threshold for p-value should be decided. This is a cut-off below which the
results are statistically significant, and above which the results are not statistically
significant. Often a threshold of 0.05 is used. This means that every 20th time we
conclude by chance alone that the difference between groups is statistically signif-
icant, when it actually isn’t.
If the compared groups are large enough, even the tiniest difference can get
a significant p-value. In such cases it needs to be carefully weighted whether the
statistical significance is just that, statistical significance, or is there a real biological
phenomenon acting in the background.

6.9.4 Hypothesis pair

Before applying the test to the data, a hypothesis pair should be formed. A hypoth-
esis pair consists of a null hypothesis (H 0) and an alternative hypothesis (H 1). For
the tests described here, the hypotheses are always formulated as follows.
H0 = There is no difference in means between compared groups
H1 = There is a difference in means between compared groups.
56 DNA microarray data analysis

6.9.5 Calculation of test statistic and degrees of freedom

The test statistic is a standardized numerical description of the differences between
the group means. Depending on the test type, the actual mathematical formula for
the calculation of the test statistic is different. Here we will present these formulas
for a couple of t-tests.
The simplest t-test is the one-sample t-test . It is constructed to compare a
sample group mean with a certain hypothesis. The general formula of the test is

M −µ
T= ,
√s
n

where M is the sample mean, µ is the expected sample mean (hypothesis), s is

the standard deviation and n is the number of observations in the sample group.
Degrees of freedom are calculated using the following formula:

df = n−1

The means of two samples are compared with an independent samples t-test
. In order to make the test, we need to know whether the variances of the groups
are equal. As a rule of thumb, the variances of the groups can be assumed to be
equal, if the variance of the first group is not more than three times larger than the
variance of the second group. Assuming that the variances of the two-groups are
not equal, the formula of the test statistic (Welsh’s t-test) is:

Xi − X j
T=
,
si2 s2
ni + n jj

where X i and X j are the means of the compared groups, s i2 and s 2j are the variances
of the compared groups, and n i and n j are the numbers of observations in the
compared groups.
The degrees of freedom are calculated with a formula:
 2
si2 s2
ni + n jj
df = ,
s2 s2
( ni )2 ( n j )2
i
n i −1 + n j j−1

where si2 and s 2j are the variances of the compared groups, and n i and n j are the
numbers of observations in the compared groups.
If the variances of the groups are equal, the test statistic (student’s t-test) is
calculated with the formula:
x̄ 1 − x̄ 2
T=  ,
s × n11 + n12

where
6 Basic statistics 57

  2
n1
i=1 (x 1i − x̄ 1 )2 + ni=1 (x 2i − x̄ 2 )2
s= .
n1 + n2 − 2
Degrees of freedom for the equal variances t-test are calculated as:

d f = ni + n j − 2

Formulas for the test statistics of non-parametric tests are not presented here,
but they can be easily found from any statistical textbook.

6.9.6 Critical values table

A critical values table contains the standardized values of a distribution. The critical
value table for t-tests contains the standardized values of the t-distribution. Recall,
that the shape of the t-distribution was partly defined by degrees of freedom. There-
fore, dfs are also essentially needed when reading the critical values table for the
t-distribution. Below (Table 6.1) is a sample of the ten first rows of a t-distribution
table
Table 6.1: Probability of exceeding the critical value.

df 0.10 0.05 0.025 0.01 0.005 0.001

1. 3.078 6.314 12.706 31.821 63.657 318.313
2. 1.886 2.920 4.303 6.965 9.925 22.327
3. 1.638 2.353 3.182 4.541 5.841 10.215
4. 1.533 2.132 2.776 3.747 4.604 7.173
5. 1.476 2.015 2.571 3.365 4.032 5.893
6. 1.440 1.943 2.447 3.143 3.707 5.208
7. 1.415 1.895 2.365 2.998 3.499 4.782
8. 1.397 1.860 2.306 2.896 3.355 4.499
9. 1.383 1.833 2.262 2.821 3.250 4.296
10. 1.372 1.812 2.228 2.764 3.169 4.143

From the table, read the value at the intersection of the degrees of freedom and
the set p-value threshold divided by two. For example, if the we used p-value of
0.05 and df = 9, the critical value would be 2.262.

6.9.7 Drawing conclusions

Compare the test statistic and the critical value from the table. If the calculated
test statistic is larger, then the null hypothesis is rejected, and we can conclude that
there is a difference between the groups with a certain p-value threshold.

6.9.8 Multiple testing

The more analyses are performed on a data-set, the more the results will meet the
conventional significance level by chance alone. Often the p-value is corrected to
58 DNA microarray data analysis

account for this problem. One commonly used correction is the Bonferroni correc-
tion, where the original p-value is divided by the number of comparisons to create
a new corrected p-value against which those comparisons should be tested.

6.10 Analysis of variance

The purpose of the analysis of variance (ANOVA) is to test for significant differ-
ences between the means of several groups. If we are comparing two means, the
ANOVA will give the same result as the t-test. However, unlike the t-test, ANOVA
does not specify which of the groups are significantly different from each other,
only that there are significant differences.
It may seem odd that a procedure that compares means is called analysis of
variance. The name is derived from the fact that in order to test for a statistical
significance between means, we actually need to compare or analyze variances.

6.10.1 Basics of ANOVA

At the heart of the ANOVA is a procedure where the variances are divided up or
partioned. Variance is computed as the sum of squared deviations from the overall
mean, divided by the corrected sample size. Thus, given a certain sample size, the
variance is a function of sums of squares (squared deviations from the means), or
SS. In the most simple form SS can be divided as SStotal = SStreatment + SSerror.
The variance that is not explained by the treatment or group membership is thought
be due to experimental errors.
There are some assumptions the data should fulfill in order to be eligible for the
ANOVA analysis. First of all, the data should be normally distributed. Furthermore,
variances of the compared groups should be similar. These assumption are robust
to small deviation from these assumptions. The strictest assumption, for which the
test is not robust, is the assumption of independence. The observations should be
independent of each other. Therefore,a time series can not usually be analyzed by
the ANOVA.

6.10.2 Completely randomized experiment

A completely randomized experiment is the simplest ANOVA design. This proce-
dure allows us to compare the means of different treatments, and is also known as
the one-way ANOVA. For microarray experiments, this would mean, for example,
comparison of the mean expression of different cell lines or different subarrays.
Before the experiment, subjects (here, chips) are randomly selected for the treat-
ment groups. If randomization is not used, the effect of treatment might get masked
by the effect of variation of the subjects.
Let us consider the following experiment (Table 6.2). We want to compare the
effect of three different bacterial treatments on the same cell line. The nine avail-
able cell culture flasks are randomized into treatment groups. The experiment is
conducted, and the mean expression of certain immunorelated genes is quantified.
The partition of variance is done in two phases. First, the SS within groups or
SSerror (how much individual observations deviate from the group mean) is calcu-
6 Basic statistics 59

lated using the groupwise means (Table 6.3). This reflects the various errors in the
experiment. Then the SS between groups or SStreatment (how much the individual
observations deviate from the overall mean) is calculated using the overall mean
(Table 6.4). It reflects the effect of treatment on different groups.
SStreatment has k-1 degrees of freedom, where k is the number of groups.
SSerror has N-k dfs, where N is the total number of subjects, and k is the number of
groups. Mean squares (MS) are calculated by dividing the SSs by the concommitant
dfs.
The results are summarized in an ANOVA table (Table 6.5), which reports
sum of squares, and the F-statistic with an associated p-value. The F-test statistic
is calculated as
M Streatment
F=
M Serror
The p-value reported in the table below has been read from the F-distribution
table of critical values with the appropriate dfs and F-statistic.

Table 6.2: Primary data.

Bacteria 1 Bacteria 2 Bacteria 3

Observation 1 2 6 4
Observation 2 3 7 5
Observation 3 1 5 3
Mean 2 6 4
Overall mean 4

Table 6.3: Calculation of error terms.

Bacteria 1 Bacteria 2 Bacteria 3

Observation 1 2-2=0 6-6=0 4-4=0
Observation 2 2-3=-1 6-7=-1 4-5=-1
Observation 3 2-1=1 6-5=1 4-3=1
Sum of squares 0+1+1 0+1+1 0+1+1
SS within groups 2+2+2=6

Table 6.4: Calculation of treatment effects.

Bacteria 1 Bacteria 2 Bacteria 3

Observation 1 4-2=2 4-6=-2 4-4=0
Observation 2 4-3=1 4-7=-3 4-5=-1
Observation 3 4-1=3 4-5=-1 4-3=1
Sum of squares 4+1+9=14 4+9+1=14 0+1+1=2
SS total 14+14+2=30
60 DNA microarray data analysis

Table 6.5: ANOVA table presents a summary of the results.

SS df MS F p-value
Effect 24 2 12 12 <0.01
Error 6 6 1

The interpretation of the ANOVA p-value is similar to the t-test. Because

the calculated p-value is lower than 0.05, we conclude that the null hypothesis
is rejected, and the alternative hypothesis comes to power. More specifically, we
know that there is a difference between the treatment groups, but we do not know
which treatments differ from each other. T-test would give a definite answer, if we
need to know specifically which groups are different.
The completely randomized ANOVA design presented above is also known
as the one-way ANOVA, because the states of only one variable are studied. If
there had been two variables, say the treatment and the cell-line, then the procedure
would have been called a two-way ANOVA. There are many more experimental
designs for the ANOVA, which are more thoroughly covered elsewhere (Sokal,
Biometry).

6.11 Statistics using GeneSpring

A few examples of statistical data manipulation, simple statistics calculation and
statistical testing using the DNA microarray data analysis software GeneSpring are
given in this section.

6.11.1 Simple statistics

Genewise Simple statistics, like the average, minimum, maximum, standard error
of the mean, and standard error can be produced through Edit->Copy->Copy An-
notated Genelist in GeneSpring. After pasting the information in Excel or other
spreadsheet program, the values will become apparent. These simple statistics can
be produced for intensities of either channel (raw and control data in GeneSpring),
intensity ratio and log-transformed intensity ratio (normalized data in GeneSpring).

6.11.2 Tranformations
In GeneSpring, data transformations are linked to Experiment Interpretation. There
are three options to choose from: Non-transformed data (ratio) log 2 -transformed
data (log of ratio) and fold change . When one transformation is chosen, Gene-
Spring will automatically recalculate the data values, and use the new values for
any subsequent statistical analyses (statistical group comparison, k-means cluster-
ing, etc.).

6.11.3 Scatter plot and histogram

A scatter plot can be produced in GeneSpring through View->Scatter plot. The
shown axes can be modified from View->Display Options. A scatter plot can easily
6 Basic statistics 61

be used for testing the linearity of the data.

A histogram is displayed, if you have selected the View->View Graph, and
additionally have set up parameters so that the quantification results are shown sep-
arately for every hybridized chip. For example, in a simple time series experiment,
setting time as a non-continuous parameter would produce a histogram of expres-
sion values. You can use histograms for assessing the distribution of the data. After
log-transformation the distribution of expression values should be symmetric and
one-peaked.

6.11.4 Correlation
The Pearson correlation between chips is automatically calculated. The values of
correlation coefficients can be viewed through Condition Inspector. Condition In-
spector is invoked when the right-hand mouse button is clicked over one chip in the
navigator bar, and Inspect is selected. From the opening window, select the Similar
Conditions tab. The correlation coefficients between the selected chip and all the
other chips are reported in the column Correlation (Figure 6.10).

Figure 6.10: The Pearson correlation in GeneSpring is found under the Similar Condition-
tab in Condition Inspector.

6.11.5 Linear regression

GeneSpring can calculate a linear regression model producing a line of best fit for
a 2D scatter plot view. The line of best fit is produced from View->Display options.
62 DNA microarray data analysis

Select the Lines to Graph tab, and tick Line of Best Fit box. The linear regression
line is overlaid with the scatter plot, and the regression equation of form y = a X +b
is displayed at the bottom of the scatter plot view. Recall that a in the regression
equation equals the correlation coefficient between the two variables plotted along
the axes.

6.11.6 One-sample t-test

The one sample t-test in GeneSpring is automatically calculated for all the genes,
whenever replicates are available. The t-test p-values can be found from the Gene
Inspector, which is opened by double-clicking the left mouse button over a gene
(Figure 6.11). The same p-values are also reported in the Spreadsheet, which is
invoked from File->View as spreadsheet.

Figure 6.11: In GeneSpring the p-values for the one-sample t-test are found from Gene
Inspector

6.11.7 Independent samples t-test and ANOVA

The independent samples t-test and ANOVA are located in Tools->Filtering and
statistical analysis. Clicking the right hand mouse button on an experiment in
the opening window enables one to Add Statistical Group Comparison. You can
specify the parameters by which the compared groups are defined, select the groups
to compare, select the appropriate statistical test and adjust p-value and multiple
testing correction (Figure 6.12).
The result of the statistical group comparison is a list of genes, which are
statistically differentially expressed between the specified groups. The actual p-
values for these genes can be found from the Gene List Inspector, which can be
opened by clicking a gene list with the right hand mouse button, and selecting
6 Basic statistics 63

Inspect (Figure 6.13).

Figure 6.12: In GeneSpring statistical group comparison is a tool of its own.

Figure 6.13: The result of the statistical group comparison is stored into a genelist, which
can be viewed with a Genelist Inspector.
64 DNA microarray data analysis

6.12 Suggested reading

1. Ranta, E., Rita, H., Kouki, J. (1997) Biometria, Yliopistopaino, Helsinki.

2. Sokal, R. R., Rohlf, F. J. (1992) Biometry, Freeman and co, New York.

3. Hopkins, W. G. (2003) A new View of Statistics, http://www.sportsci.

org/resource/stats/index.html.

4. Pezzullo, J. C. (1999) Interactive Statistical Calculation Pages, http://members.

aol.com/johnp71/javastat.html.

5. Statsoft Inc. (2002) Electronic Statistics Textbook, http://www.statsoft.

com/textbook/stathome.html.

This chapter was written by Jarno Tuimala.

 Part II

Analysis
66 DNA microarray data analysis

7 Preprocessing of data
7.1 Rationale for preprocessing
Preprocessing includes analytical or transformational procedures that need to be
applied to the data before it is suitable for a detailed analysis.
The black-box thinking, where data is fed into a program, and the result pops
out, is gaining ground rapidly. This kind of approach for statistical analysis in sim-
ply erraneous, because the results coming out from the program can be statistically
erraneously derived. In such cases, also the biological conclusions can be wrong.
Statistical tests have often strick assumptions, which need to be fulfilled. Violation
of assumptions can lead to grossly wrong results.
We strongly recommend that the researcher, even if he/she is not himself per-
forming the data analysis, gets basic knowledge of the data, because the results
presented by the bioinformatician or statistician are more easily interpretable, if
one is at least somewhat familiar with the data.
Here we will introduce some methods for checking the data for violation of
statistical test assumptions. We also present some things to consider before and
during the data analysis. The methods are introduced in the order of applicability,
although some methods are needed in several steps.

7.2 Missing values

There are often many missing values in microarray data. As you might recall, miss-
ing values are observations (intensities of spots), where the quantification results
are missing. In the context of microarrays, we define missing values as

• Missing because the spot is empty (intensity = 0).

• Missing because background intensity is higher than the spot intensity (back-
ground corrected intensity < 0).

Missing values can lead to problems in the data analysis, because they easily
interfere with computation of statistical tests and clustering. Therefore, it is worth
giving a thought to the treatment of missing values. There are a couple of options
for the treatment of missing values: They may be replaced with estimated values in
a process called imputation, or they can be deleted from the further analyses.
The default way of deleting missing data (in most of the software packages),
for example while calculating a correlation matrix, is to exclude all cases that have
missing data in at least one of the selected variables; that is, by casewise deletion of
7 Preprocessing of data 67

missing data. However, if missing data are randomly distributed across cases, you
could easily end up with no "valid" cases in the data set, because each of the genes
will highly likely have at least one missing observation on some chip. The most
common solution used in such instances is to use the so-called pairwise deletion,
where a statistic between each pair of variables is calculated from all cases that
have valid data on those two variables.
Another common method is the so-called mean substitution of missing data
(mean imputation, replacing all missing data in a variable by the mean of that vari-
able). Its main advantage is that it produces internally consistent sets of results.
Mean substitution artificially decreases the variation of scores, and this decrease
in individual variables is proportional to the number of missing data. Because it
substitutes missing data with artificially created average data points, mean substi-
tution may considerably change the values of correlations. Imputation is commonly
carried out for intensity ratios, but can also be done for raw data.
Different computer programs manipulate missing values very differently, and
drawing any consensus would be futile. At least statistical programs often offer
a possibility to define whether to use imputation, pairwise deletion or casewise
deletion.

raw data log−transformed data

14
log2(rmean)
20000
rmean

12
10
0

0 500 1500 2500 10.0 10.5 11.0 11.5

mouse1$morphR log2(mouse1$morphR)

raw data vs. corrected data log2−data vs. corr. data

14.0
15000
rmean

rmean

12.5
5000

11.0

0 500 1000 1500 2 4 6 8 10

r.ba r.ba

Figure 7.1: Raw and log-transformed intensity values of the red channel plotted against
its background. Upper row contains uncorrected scatter plots, lower row background cor-
rected scatter plots.
68 DNA microarray data analysis

green.background ratio vs. green background ratio vs. red background

2500
red.background
500 1000

1000
0

0
0.5 1.0 1.5 2.0 0.5 1.0 1.5 2.0

ratio ratio

ratio vs. average background log2(average.background) logratio vs. log−background

average.background

10.5
1500
0 500

9.5

0.5 1.0 1.5 2.0 −2 −1 0 1

ratio log2(ratio)

Figure 7.2: Intensity ratios (expression) plotted against the background of the channels.

7.3 Checking the background reading

There is some dispute on whether the background intensities should be subtracted

from the spot intensities. At the moment, the background adjustion seems to be
commonly used, but strictly speaking, it’s applicability should be checked.
The background intensities should not correlate highly with the spot intensi-
ties, if the spot intensities are truly independent of the background intensities. This
can be assessed by a scatter plot of background intensities against spot intensities.
If the spot intensities are dependent on the background intensities, it is possible ei-
ther not to apply any background correction to the data (Figure 7.1 and Figure 7.2)
or discard the deviating observations from further analyses.
Another common problem with the background correction is that it might pro-
duce “pheasant tail” images on the scatter plot (Figure 7.3). Pheasant tails are
formed by observations, which have exactly the same intensity value on one chan-
nel, but the other channels’ intensities can vary. In such cases long vertical or
horizontal straights of observations are produced, and the resulting data cloud re-
sembles a pheasant tail in a scatter plot. This is especially common in the lower end
of the intensity distribution, and can be a sign that the scanner is not very reliable
below a certain cut-off intensity or that the image analysis software calculates back-
ground intensities in a peculiar way. When pheasant tails are observed, background
uncorrected intensity values can be used the analyses, or the deviating observation
can be excluded from any further analyses.
7 Preprocessing of data 69

Figure 7.3: An A versus M plot, where the pheasant tail is visible in the lower intensity
values.

The background corrected intensity values are calculated spotwise by subtract-

ing the background intensity from the spot (foreground) intensity. The formulas for
Affymetrix and two-color data are

C y3 = C y3spot − C y3background

C y5 = C y5spot − C y5background

Note, that Affymetrix arrays do not quantify background around the spots, but
instead use a certain mathematical formula (check the Affymetrix manual).

7.4 Calculation of expression change

There are three commonly used measures of expression change . Intensity ratio
is the raw expression value, and log ratio and fold change are transformationally
derived from it.

7.4.1 Intensity ratio

After background correction, expression change is calculated. The simplest ap-
proach is to divide the intensity of a gene in the sample by the intensity level of the
same gene in the control. Intensity ratio can be calculated from background cor-
rected or uncorrected data (here we use corrected data) The formula for two-color
data is
70 DNA microarray data analysis

C y3
Intensity ratio =
C y5
For Affymetrix data, substitute Cy3’ and Cy5’ with the appropriate intensities
from the sample and control chips.
This intensity ratio is one for an unchanged expression, less than one for down-
regulated genes and larger than one for up-regulated genes. The problem with
intensity ratio is that its distribution is highly asymmetric or skewed. Up-regulated
genes can take any values from one to infinity whereas all downregulated genes are
squeezed between zero and one. Such distributions are not very useful for statistical
testing.

7.4.2 Log ratio

To make the distribution more symmetric (normal-like) log-transformation can be
applied. The most commonly used log-transformation is 2-based (log 2 ), but it does
not matter whether you use natural logarithm (log e ) or base 10 logarithm (log 10 ) as
long as it is applied consistently to all the samples.
The formula for the calculation of log 2 -transformation is:
 
C y3
Log ratio = log 2 (Intensity ratio) = log 2
C y5
After the log-transformation, unchanged expression is zero, and both up-regulated
and down-regulated genes can take values from zero to infinity. Logratio has some
nice properties compared with other measures of expression. It makes skewed dis-
tributions more symmetrical, so that the picture of variation becomes more realistic.
In addition, normalization procedures become additive. For example, we have
the following intensity ratio results for two replicates (A and B) after normalization:
120
Gene A = = 2.0
60
30
GeneB = = 0.5
60
The mean of these replicates is 1.25 instead of 1, which would have been
expected. If the 2-based logarithmic transformation is applied, the log ratios are:
 
120
Gene A = log2 = 1.0
60
 
30
GeneB = log2 = −1.0
60
The mean of these log ratios is 0, which corresponds to the mean intensity ratio
of 1. Although log-transformation is not always the best choice for microarray data,
it is used because the other transformations lack this handy additive property. One
downside of the log-transformation is that it introduces systematic errors in the
lower end of the expression change distribution.
7 Preprocessing of data 71

7.4.3 Fold change

Another means to make the distribution of intensity ratios more symmetrical is to

calculate the fold change. The fold change is equal to the intensity ratio, when the
expression is higher then one. Below one, the fold change is equal to the inversed
intensity ratio.

C y3
For values >1, fold change = C y5
1
For values <1, fold change = (C y3 /C y5 )

The fold change makes the distribution of the expression values more sym-
metric, and both under and over-expressed genes can take values between zero and
infinity. Note, that the fold change makes the expression values additive in a similar
fashion as the log-transformation.

7.5 Handling of replicates

Replicates are a very powerful way to reduce noisiness of the data. Noisiness (ran-
dom errors) is an undesirable feature of the data, because it potentially abolishes
interesting information. It can results from many sources, some of which are hard
to deal with.

7.5.1 Types of replicates

There are at least three different kinds of replicates that are potentially useful in the
context of DNA microarray analysis. For example, if we treat a certain cell line
with a cancer drug, we can set up several culture flasks, and then harvest them as
biological replicates. For every culture flask, the isolated and labeled mRNA popu-
lation can be hybridized to several chips making a number of technical replicates .
In addition, every hybridized chip can be quantified several times or using different
image analysis software (software replicates) .
These are all valuable sources of information about the variability in the data.
In practise, it might be a good idea to make some biological replicates and some
technical replicates. Then the biological and technical variation can be taken into
account in the same experiment.

7.5.2 Time series

In a time series experiment expression changes are monitored with samples taken
between certain time intervals. Although several replicates can be made per every
time point, it should be considered that these replicate chips can possible be made
a better use of, if they are added to the time series as sampling points. That is, it
should be weighted whether a high precision in every time point is more valuable
than the additional information of expression changes new sampling points (time
points) produce.
72 DNA microarray data analysis

7.5.3 Case-control studies

For case-control studies, where for instance cancer patients are compared with their
non-diseased controls, the individuals in both groups can be considered replicates
of that disease state. This is assuming that we are interested in the differencies of
those two disease states instead of inter-individual variation.

7.5.4 Power analysis

Using power analysis we can easily compute the number of replicates that are
needed to reliably discover the genes that are expressed. For example, we would
need 11 replicates to reliably pick up the genes, which are at least 1.41-fold over-
expressed at a p-value of 0.05. Similarly, 6 replicates are needed if genes with
2-fold over-expression need to be quantified using a p-value cutoff of 0.01.

7.5.5 Averaging replicates

The goal of DNA microarray experiments is to semi-quantitatively analyze the gene

expression level in a certain material. Often the precision of the expression level
estimate is important. The estimate of the true expression necessarily becomes
more precise when more replicates are used for that purpose.
Replicates also produce data on variability between the measured states, for
example cancer patients and healthy controls. More importantly, replicates po-
tentially reduce noise in the data. However, if replicates are averaged, we lose
information about the inter-individual variability in the studied groups.
Assume that we have an experiment, where there are two biological replicates
and three technical replicates of both biological replicates, making a total of 6 repli-
cates altogether. The correlation between any of the technical replicates from the
two biological replicates is, at first, quite low (<0.5). Taking average of the tech-
nical replicates makes the correlation between biological replicates much higher
(>0.8). This also hints that the noisiness of the data has been reduced. Of course,
bad replicates (chips or spots) should ideally be removed before calculation of av-
erages, even though averaging diminishes the influence of outliers on the results.
In practise, replicates are often averaged after normalization, but sometimes
even the raw intensity values are averaged. We feel that it would be better to average
replicates after normalization, because then it is possible to take chipwise variation
into account in the normalizations.

7.6 Checking the quality of replicates

If the experiment includes replicates, their quality can be checked with simple
methods, using scatter plots and pairwise correlations or hierarchical clustering
techniques, which are explained in the cluster analysis chapter in more detail. Of-
ten the quality of replicates is checked before and after the normalization. If the
correlation between replicates drops dramatically after certain normalization, the
applicability of the normalization method should be reconsidered.
7 Preprocessing of data 73

7.6.1 Quality check of replicate chips

The first task is often to find out, how well two replicate chips resemble each other.
In such cases the intensity or log ratios can be plotted along the axes of a scatter
plot. Three replicates can be plotted using a three dimentional scatter plot. If
the replicates are completely similar, the data points in the scatter plot fall onto a
perfectly straight line running through the origin of the plot. If the replicates are
not exactly similar, the observations form a data cloud rather than a straight line.
The absolute deviation from a line can be better visualized, if a linear regression
line is fitted to the data.
Correlation measures the linear similarity quite well. It might give misleading
information, if the data cloud is not linear. Correlation coupled with a scatter plot
gives much more information. In practise, the Pearson’s correlation coefficients
between two replicate cDNA chips produced from the same mRNA pool (technical
replicates) are often in the range of 0.6–0.9, and the correlation between similarly
hybridized Affymetrix chips is typically over 0.95. When biological replicates are
performed, the correlation between replicate chips usually drops from these values.
For cDNA chips, a correlation of 0.7 between replicates may be considered good,
whereas for Affymetrix chips a correlation over 0.9 may be considered a good
result.
In addition, the scaling factor should be checked for the Affymetrix chips. If
the factor is larger than 50, the hybridization is probably bad.
Quality of several replicate chips is most easily checked by hierarchical clus-
tering. For example, if the two replicate chips are placed closest to each other
in the dendrogram, they can be expected to be good replicates. Depending on the
similarity measure, the hierarchical clustering can be applied either after (Euclidian
distance) or before the normalization (Pearson’s correlation).

7.6.2 Quality check of replicate spots

The quality of replicate measurements of one gene (one spot) can be assessed in
a similar fashion to replicate chips, using correlation measures. However, this is
more laborous than with replicate chips, but it can sometimes be essential that bad
replicate spots are found and removed from the data.

7.6.3 Excluding bad replicates

It is quite common to exclude bad replicates from further analyses. For example,
if there are four replicate chips available for the same treatment, and one seems to
deviate from every other replicate, then this most deviating one can be removed
from any further analyses. This often results in a massive loss of data, because
whole chips are removed from the analysis. Thus, it would be better to study the
quality of replicates on a gene-wise basis, after initially studying the quality of
whole chips using the tools described above.
In practise, there often are replicates which deviate from the others. These can
be results of modified experimental conditions (different mRNA batch, etc.) or just
important biological variation, if different animals of human subjects are studied.
74 DNA microarray data analysis

Because of this, some caution should be exercised when removing bad replicates.

7.7 Outliers
Outliers in chip experiments can occur at several levels. There can be entire chips,
which deviate from all the other replicates. Or there can be an individual gene,
which deviates from the other replicates of the same gene. These deviations can
be caused by chip artifacts like hairs, scratches, and precipitation. Precipitation,
among other causes, can also perturb one single spot or probe.
Outliers should consist mainly of quantification errors. In practise, it is often
not very easy to distinguish quantification errors from true data, especially if there
are no replicate measurements. If the expression ratio is very low (quantification
errors) or very high (spot intensity saturation), the result can be assumed to be an ar-
tifact, and should be removed. Most of the actual outliers should be removed at the
filtering step (those that have too low intensity values), and some ways to identify
deviating observations were presented in the section on checking replicates.
In the absence of replicate, the highest and lowest 0.3% of the data (gene ex-
pression values on one chip) is often removed, because assuming normality, such
data resemble outliers. These values are outside the range of three standard de-
viations from the distribution mean. This is often equivalent to a filtering, where
observations with too low or high intensity values are excluded from further analy-
ses.
Formally, a statistical model of the data is needed for the reliable removal
of outliers. The simplest model is equality between replicates. If one replicate
deviates several standard deviations from the mean of the other replicates, it can
be considered an outlier and removed. The t-test measures standard deviation and
gives genes, where outliers are present among replicates, a low significance.
In practise, it is easiest and best to couple the removal of outliers with filtering ,
for example, using the following procedure: First, genes outside the range of three
standard deviations from the chip mean are removed. In the succeeding filtering
steps, the genes whose intensity is too low are removed, if needed. Next, genes
which do not show any expression changes during the experiment are eliminated
(this can be based on either log ratio or standard deviation of log ratios). What is
left in the end, is the good quality data. There are also some advanced statistical
methods developed that also allow for outlier detection and removal, but they are
outside the scope of this book.

7.8 Filtering bad data

Filtering is a process where observations which do not fulfill a preformulated pre-
sumption are excluded from the data. For example, there is a certain limit of the
scanner below which the intensity values can not be trusted anymore. Typically,
the lowest intensity value of the reliable data is about 200 for Affymetrix data and
1000 for cDNA microarray data. These cut-offs are likely to change as the scanners
get more precise. The values below the cut-off point are usually removed (filtered)
from the data, because they are likely to be artifacts.
7 Preprocessing of data 75

The idea of filtering is simple. We want to remove all the data we do not
have confidence in, and proceed with the rest of the data. The results based on the
trustworthy data are often biologically more meaningful than the results based on
very confusing or noisy data.
The first step of filtering is flagging. Flagging is performed at the image anal-
ysis step. It’s idea is to mark the spots which are, by eye, judged to be “bad”.
Then, in the data analysis phase, the spots, which were flagged as “bad” are re-
moved from the data. For example, the spots overlapped with a sizable dust particle
should be flagged as bad and removed at the data filtering step.
There are a couple of statistical measures that can be used for filtering. Some
image analysis programs give signal-to-noise or signal-to-background measure-
ments for every spot on the array. These quality measurements can be used for
filtering out bad data. Often a cut-off point of 90–95% for signal-to-noise ratio is
used, at least on control channel.
Signal-to-background can be calculated after the image analysis, from the in-
tensity values:
spot signal on green channel
background signal on green channel
The spots can then be filtered on the basis of this quality statistic. Signal to
background plots of both channels appear often very much the same as the original
data (Figure 7.4).

green channel stb red channal stb

150

30
100
g.stb

20
r.stb
50

10
0

0 10000 30000 0 10000 30000

gmean rmean

intensity>2*sd, both channels original data

30000

20000
rmean
r.qc

0 10000

0 10000 30000 0 10000 30000

g.qc gmean

Figure 7.4: Raw intensities and log-transformed intensities of different channels plotted
against their signal-to-background ratios (stb).
76 DNA microarray data analysis

It would also be expected that the signal to background ratio would increase as
the intensity of the channel increases, if the background is approximately the same
in all areas of the chip (Figure 7.5). This is one hallmark of nice-quality data.
150
100
g.stb

50
0

9 10 11 12 13 14 15

log2(gmean)

Figure 7.5: The signal-to-background ratios (stb) increase as the intensity of the spot
increases.

What is also often used, is the filtering by the standard deviation of background
intensities. The observations, which have an intensity value lower than background
+ 2–6 times standard deviation of the background intensities, are often removed
from the data.

7.9 Filtering uninteresting data

After the removal of bad data, we are left with the good quality data, of which
most is probably uninteresting for us. If the goal of the study is to find a couple
of dozens of genes for further studies of the biologically interesting phenomenon,
it is a good idea to remove the uninteresting part of the data before clustering or
classification analyses. Uninteresting data comprises of the genes that do not show
any expression changes during the experiment.
For example, assume that we have a time series, and we want to find the genes
that have a most upregulated expression at the end of the sampling. In such cases,
genes that remain unchanging during the study, are removed from further analyses.
This is because the clustering methods generally work better, if the uninteresting
data is not swamping the changes in the interesting part of the data.
7 Preprocessing of data 77

Similarly, if we are interested in finding the genes that differentiate between

two groups of patient, the genes that do not show any change between these groups,
can be removed from the analysis.
Most often the intensity ratio cut-offs for uninteresting data (not-changing
genes) are set at 0.5 and 2.0. These cut-offs also roughly correspond to the thresh-
olds of reliable data: All the variation between these cut-offs can be assumed to
be due to random quantification errors of the non-changing genes. The modern
chip technology is able to produce better chips, and in practise these cut-offs can
often be set on a lower level, for example to 0.7 and 1.5. The cut-offs can be better
characterized using a Volcano plot (see chapter on genelists).

7.10 Simple statistics

There are some simple statistical measures of the data, which can be performed
before or after normalization. They will give hints as to what statistical tests to
apply or what is the distribution of the observations. These statistics should be
checked from the distribution of log-transformed data.

7.10.1 Mean and median

When the distribution of the intensity values is skewed, the median characterizes
the central tendency better than the mean. The median and mean can also be used
to check the skewness of the distribution. For symmetrical distributions mean and
median are approximately equal.

7.10.2 Standard deviation

The standard deviation gives you an idea of how spread your data are. It is also
good to check the coefficient of variation (standard deviation / mean). This should
give you a better idea of the real spread of the data, because the value of CV is
independent of the absolute values of the variable. For example, replicates should
have a small CV, although the value of standard deviation can vary a bit.

7.10.3 Variance
Sometimes you need to make a decision of the appropriate statistical test. For
example, there are different kinds of t-tests for variables with equal or unequal
variances. As a rule of thumb, the variances can be assumed to be equal if the
variance of the first variable if less than three times the variance of the second
variable.

7.11 Skewness and normality

Skewness and normality should be checked after normalization, but it is usually
worthwhile, if they are checked before normalization, too. Then the distribution of
the data (log-transformed intensity ratio) can be compared before and after normal-
ization. This can be partly used for checking the success of the normalization.
78 DNA microarray data analysis

Unacceptable histogram Acceptable histogram

1500
4000
Frequency

Frequency
2000

500
0

0
0 20000 40000 0 10000 30000

gmean gmean
Not normally distributed Not normally distributed

Unacceptable histogram 300

Acceptable histogram
2000
Frequency

Frequency

200
1000

100
0

3.0 3.2 3.4 3.6 3.8 4.0 3.2 3.4 3.6 3.8

log2(na.gmean) log2(na.gmean)
Approximately normally distributed Approximately normally distributed

Figure 7.6: Four histograms of the same data. The two upper histograms contain the raw
data, and the two lower histograms the log-tranformed data. The log-transformed data is
clearly more normal-like than the non-transformed data.

Normality means how well your data fits to the normal distribution. This
should be checked, because most of the statistical procedures assume that the data
is normally distributed. Even the most basic descriptive statistics can be misleading
if the distribution is highly skewed; standard deviation does not bear a meaningful
interpretation if the distribution significantly deviates from normality.
The easiest method for checking normality and skewness of the distribution
is to draw a histogram of the intensities (Figure 7.6). For checking the skewness,
comparison of the mean and median is complementary to this graphical method.
Note that there should be enough columns in the histogram in order to make the
results reliable.
There is also a more informative way to check for the normality, quantile-
quantile plot. This should be used with the histograms as a more diagnostic method
for observing the deviations from normality.

7.11.1 Linearity
It is also important to check the linearity of the data (log ratio). Linearity means
that in the scatter plot of channel 1 (red colour) versus channel 2 (green colour), the
relationship between the channels is linear. It is often more informative to produce
7 Preprocessing of data 79

a scatter plot of the log-transformed intensities, because then the lowest intensities
are better represented in the plot (Figure 7.7). In this kind of a plot, the data points
fit a straight line, if the data is linear.

14
log2(rmean)
20000
rmean

12
10
0

0 10000 30000 9 10 12 14

gmean log2(gmean)
0.5

0.5
logratio

logratio
−0.5

−0.5
−1.5

−1.5

0 10000 30000 0 10000 30000

rmean gmean

Figure 7.7: Checking the linearity of the data. Top row, red channel intensities versus
green channel intensities; bottom row, log ratio versus different channel intensities.

Another way to test linearity is to plot the log ratio versus the intensity of
one channel (Figure 7.7). This should be done for both channels independently.
In this plot the data cloud has been tilted 45% to right, and it should be easier to
identify non-linearity. Again, the data points should fit an approximately straight
line, which in this case is horizontal.
Checking the linearity of the data helps to pick the right normalization method.
It also provides information about the reliability of the data, especially in the lower
intensity range.

7.12 Spatial effects

Often the intensity values (especially background) vary a lot on the same chip de-
pending on the location of the spot on the chip. This effect is known as spatial
effect (Figure 7.8).
80 DNA microarray data analysis

green spot intensities red spot intensities

green background intensities red background intensities

Figure 7.8: Intensities of spots and their backgrounds spotted in the form of the original
chip in order to check for spatial biases. Background intensities seem to be highly variable
in different areas of the chip. Especially the corners have an abnormally high background.
These areas might be removed from further analyses.

Spatial effects are often most pronounced in the corners of the chip, where the
hybridization solution has possibly been squeezed thinner than in the other areas of
the chip. Furthermore, corners of the chip can also dry up more easily than other
areas of the chip.
Another source of spatial effects is the regional clustering of genes with a
similar expression profile. This is often caused by bad chip design, where the
probes have not been randomized onto the chip. For example, if genes acting in a
cell cycle regulation are spotted into one corner of the chip, the expression values in
the corner are probably dependent on the cell cycle stage of the sample material. In
such cases, the spatial effect should not be removed by normalization procedures,
because it contains real biological information instead of random noise.
The similar kind of spatial effect can also be checked for from the calculated
expression values (Figure 7.9). There should not be any suspicious areas on the
subarrays (assuming good chip design), where the intensity values are much higher
than everywhere else on the same chip.
7 Preprocessing of data 81

Figure 7.9: The measured gene expression spotted in the form of the original chip in order
to check for spatial biases.

If the spatial effect is identified as random noise, bad hybridization conditions

or other undesirable sources of errors, it can be compensated for by a suitable nor-
malization method (subarraywise methods).

7.13 Normalization

Normalization can be regarded as a preprocessing step, but it is so central a method

for the microarray experimentation that it warrants a chapter of its own. Thus we
will not discuss normalization in more detail here, but instead chapter 8 is dedicated
for the issue.

7.14 Similarity of dynamic range, mean and variance

Many statistical procedures assume that across the experiment, the dynamic range
(minimum and maximum), mean and variance of the chips are equal. These should
be checked after normalization. If they are not, the results of the statistical tests
should be interpreted with caucion.
82 DNA microarray data analysis

7.15 Examples using GeneSpring

This section describes some preprocessing examples using the GeneSpring pro-
gram.

7.15.1 Importing data

Before any preprocessing can be done for the data using GeneSpring, it should
be imported into the program. GeneSpring read virtually any file format if it is a
tab-delimited text file. In GeneSpring you need to specify what kind of informa-
tion some key columns contain. Basically, the minimal datafile contains just two
columns: one for the gene name (it can also ne Genbank accession number) and
another one for the intensity (Affymetrix) or intensityratio (cDNA chips). Because
importing data is covered very thoroughly in the online manual accessible from
Help->Online documentation, we do not go into details here.

7.15.2 Background subtraction

GeneSpring subtracts the background from the foreground or spot intensities auto-
matically, if the background intensities are present in the datafile.

7.15.3 Calculation of expression change

For cDNA microarrays this is one normalization option, Per Spot: Divide by con-
trol channel, but the option is not used with lowess-normalization. There are two
possible transformations, log-transformation (log of ratio) and fold change. The
transformation options can be accessed from Experiments -> Experiment interpre-
tation.
For the Affymetrix chips, a calculational control channel is created. The ex-
pression is then calculated as with cDNA microarrays. It is also possible to calcu-
late the expression change using a certain control chip. This is one of the normal-
ization options, Per Spot: Normalize to control samples.
Calculation of the expression change is coupled to normalization. Normaliza-
tion only affects the control channel, which is adjusted so that the median intensity
ratio of the chip is one when using Per chip: Normalize to percentile normalization.
Therefore, the control channel is created for the Affymetrix chips, also. Note, that
the original raw intensity values are always retained.

7.15.4 Replicates
When importing the data from a chip where the same spot is present in multiple
copies, GeneSpring automatically calculates an average of those replicates. This
is assuming that the replicate spots have exactly the same Gene identifier in the
data file. Replicate chips are also averaged, after defining the replicates in the
Experiment -> Experiment parameters window and setting up the parameters in
Experiment -> Experiment interpretation.
For example, if we have a time series experiment (Table 7.1) with three time
points and two replicates per every time point, the parameters should be set up
7 Preprocessing of data 83

Table 7.1: Time series experiment

Time point Replicate

chip 1 1 hour 1
chip 2 1 hour 2
chip 3 2 hours 1
chip 4 2 hours 2
chip 5 3 hours 1
chip 6 3 hours 2

as follows. Two parameters, a time point and a replicate are created. In the time
point, mark the time points suitably. The replicates are then set up using the other
parameter. Inside one time point, the replicates are marked with a running number.
Last, set up the value order for the time points. This tells GeneSpring, in which
order the time points should be displayed on the screen.
If there is missing data for some genes either when importing the data or when
setting up the replicate chips, GeneSpring only uses the existing data for the calcu-
lation of means.

7.15.5 Checking linearity

Linearity is easily checked in GeneSpring using the M versus A plot. Go to scatter
plot (View -> Scatter plot). Change the display options (View -> Display options)
so that the horizontal axis is the average of raw and control (A), and the vertical
axis is normalized (M). To help plot the non-linearity, from the Lines to Graph tab
tick the Line of best fit box, which draws the linear regression line to the plot. Note,
that the M versus A plot should be initially produced for unnormalized data.

7.15.6 Normality
Normality can be checked using the histogram. A histogram can be investigated
in (View -> Graph) window by checking All Samples - interpretation mode, that is
automatically produced for each experiment.

7.15.7 Filtering
A filtering tool can be invoked from the menu Tools -> Filtering and statistical
analysis. A new window opens, where one genelist and one experiment should
be selected. The actual filtering tool is accessed by right-clicking on the selected
experiment and selecting Add expression percentage restriction from the opening
list. Often the bad quality data is first filtered out. After that, the not-changing
genes are removed from the dataset.
Using the scatter plot tool , define the intensity value, under which you can’t
trust your data anymore on either raw or signal channel. For cDNA chips, this is
often around 200-1000 and for Affymetrix chips around 200. In the expression
percentage filtering, use this signal value as a minimum cut-off. Also select that it
84 DNA microarray data analysis

applies to all the conditions. Create one such filter for both channels. Create a new
genelist of the results (Make list button).
In the next filtering phase we will try to find the genes that are changing and we
can trust in. Using the genelist created above, set up a new filter, where you select
genes, which have expression values between 0.5 (minimum) and 2 (maximum).
These genes are not showing any expression changes and are uninteresting to us.
This filtering should also apply to the whole dataset (all conditions). After saving
the new genelist, the filtering tool can be closed.
Go to the navigator bar and right-click on the good quality gene list. From the
list, pick Venn diagram -> Left (red). Similarly, add the not-changing and all genes
genelist to the Venn diagram. Then select the All genes genelist from the navigator.
From the Venn diagram identify the region that contains the genes included in the
reliable genelist, but not in the not-changing genelist. Right-click on that area of
the diagram, and make a list of these genes.
Now you have a list of genes, which are reliable and also changing. You’re
ready to proceed with the analysis, for example to clustering or classification anal-
yses.

7.16 Suggested reading

1. Kohane, I. S., Kho, A., Butte, A. J., (2002) Microarrays for an intergrative
genomics, MIT press, Massachusetts.

2. Baldi, P., Hatfield, G. W. (2002) DNA microarrays and gene expression,

Cambridge University Press, United Kingdom.

3. Knudsen, S. (2002) A biologist guide to analysis of dna microarray data,

John Wiley and Sons Inc., New York.

This chapter was written by Jarno Tuimala.

8 Normalization 85

8 Normalization
8.1 What is normalization?

There are many sources of systematic variation in microarray experiments that af-
fect the measured gene expression levels. Normalization is the term used to de-
scribe the process of removing such variation [12]. Normalization can also be
thought of as an attempt to remove the non-biological influences on biological data.
Sources of systematic variation will affect different microarray experiments
to different extents. Thus, to compare microarrays from one array to another, we
need to try to remove the systematic variation, to bring the data from the different
experiments onto a level playing field, so to speak. One goal of normalization is
also to make possible the comparison from one microarray platform to another, but
this is clearly a much more complicated problem and is not covered in this section.
The biggest problem with the normalization process is the recognition of the
source of systematic bias. There is a strong possibility that some or even most of
the biological information will also be removed when normalizing the data. Thus
it is good to keep in mind that the amount of normalization should be minimized to
avoid losing the real biological information. In the next chapter some of the most
common sources of systematic bias will be introduced and also some methods for
recognizing the sources and dealing with the bias will be discussed.

8.2 Sources of systematic bias

8.2.1 Dye effect

Differences in dye (labeling) efficiencies are the most common source of bias and
also easily identifiable. This can be seen when the intensity of one channel on the
array is much higher than the other (see Figure 8.1, B and C). Dye effect can be
corrected for by balancing the dyes using the assumption that both channels should
be equally “bright”. Extra information for dye balancing can be received from dye-
swap experiments. Problems will occur when dye also has interaction effects; i.e.,
the labeling efficiency may depend on the gene sequence.

8.2.2 Scanner malfunction

Scanners might show many different failures. When laser or PMT intensity values
are wrongly adjusted, the scanner can cause the dye effects to show up. But most
of the scanner malfunctions are hard to deal with in silico and the best solution is to
86 DNA microarray data analysis

fix the scanner and repeat the scanning. For example, when lasers are misaligned
the two channels are slightly out of register. This can cause big problems if image
analysis software does not allow the user to align images manually.

8.2.3 Uneven hybridization

Sometimes patterns are seen on the slide that can be caused by many different
reasons. The most common ones of these are described in the next three sections.
When spatial bias is seen on the slide, the first impression is usually that it is
caused by uneven hybridization. In most cases this is also true. Uneven hybridiza-
tion can be recognized, for example, as lighter areas in the middle of the slide or on
the edges of the slide. This is usually very hard to fix numerically and it is recom-
mended to aim for developing more consistent techniques for hybridization. For
single cases, spots that are not hybridized properly can be excluded from further
analyses. Background difference can also cause bias (see Figure 8.1, E).

8.2.4 Printing tip

Slides are usually printed using more than one pen (2, 4, 8, 16, ..). If any of these
pens work differently from others, for example a pen gets infected by hair or is
defective in some other way, the corresponding subarray could differ from other
subarrays. Quite commonly printing pens also wear out at a different rate. One
way to see if one pen performs differently is to visualize the data by subarrays
using colors or regression lines so that the faulty subarray stands out. In some
cases, printing tip error can be corrected for by applying different normalization
parameters to the subarrays.
Printing can also cause problems that are not categorized as spatial bias. Some-
times some or all spots are irregular in shape or they might express as rings instead
of circles. It is likely that the detection of these spots will be inaccurate. This
should be taken into account during image analysis if possible.

8.2.5 Plate and reporter effects

These often ignored but very common effects are not artificial but are caused by
biology itself. Sometimes concentration of reporters (probes) on the slide might
differ and this can cause patterns on a slide. Usually this is easy to notice if the
position of the plates (assuming that concentration is constant on a plate) is known.
Other quality control methods should also be used. Plate effects can be corrected
for by using methods as in the case of different printing tips.
More difficult to recognize and especially to distinguish from other sources of
bias is the bias caused by the biological role of reporters. If reporters are arranged
according to their biological function (as is done in many cases) this can also be
seen as patterns on a slide. It is very important to note that this effect should not
be normalized! The reporter effect needs to be taken into account when slides are
being designed. Related reporters should never be grouped together.
8 Normalization 87

8.2.6 Batch effect and array design

When large amounts of slides have been studied, it has been noted that slides from
the same print run (or batch) often cluster together and also slides from an identical
print design but different print run. Some of these effects might be due to mistakes
in printing. This kind of bias is very hard to notice and the only way to prevent it
is to keep track of printing (LIMS) and use of good pre- and post-printing quality
control methods.

8.2.7 Experimenter issues

Equal to batch effect is the systematic bias caused by the experimenter. Experi-
ments done by the same experimenter often cluster together more tightly than war-
ranted by biology. A survey made at Stanford University showed that the experi-
menter was one of the largest sources of systematic bias. The best but not a very
practical solution for the experimenter issue would be to let the same experimenter
do everything. Because this naturally isn’t possible, consistent hybridization tech-
niques are needed as well as methods to recognize bias caused by the experimenter.

8.2.8 What might help to track the sources of bias?

If you have the possibility to hybridize one or several arrays where the same sam-
ple has been used both as target and control i.e., the same sample labeled with both
dyes on the same array), or to study two replicate arrays as sample and reference,
the results of both channels should be equal, a scatter plot should show a straight
line, and optimally, your data should follow natural distribution. Studying self-self
hybridized arrays shows you clearly the sources of error, such as uneven incorpo-
ration of the dyes, that are not dependent on your sample. It can also help you in
deciding which normalization steps would minimize these errors.
Scatter plot (MA plot, see 8.5.5.) may be used to illustrate the different types
of effects due to intensity-dependent variation as shown in Figure 8.1.

8.3 Normalization terminology

There are three broad groups of normalization techniques, each of which can con-
tain multiple methods. Classically, normalization means to make the data more
normally distributed. In microarray analyses, this is still the goal, but the method-
ology has concealed the original idea.
88 DNA microarray data analysis

Figure 8.1: Truncation at the high intensity end (A) is caused by scanner saturation.
High variation at the low intensity end (B) is from larger channel- specific additive er-
ror. High variation at the high intensity end (C) is from larger channel- specific multi-
plicative error. The curvature in (D) is from channel mean background difference. The
curvature in (E) is from slope difference. The split RI plots in (F) come from hetero-
geneity. Figure adopted and modified from a paper entitled "Data transformation for
cDNA Microarray Data" by Cui, X., Kerr, M. K., and Churchill, G. A. (Submitted to
Statistical Applications in Genetics and Molecular Biology. The manuscript web site is at
http://www.jax.org/staff/churchill/labsite/pubs/index.html)

8.3.1 Normalization, standardization and centralization

In general, normalization means the process of transforming a statistic (e.g., inten-
sity ratio) so that it will approximate normal distribution, or be more normal-like
(see 7.9). In the microarray field, normalization means also the standardization and
centralization methods.
Usually a log-transformation works quite well as a normalization method for
microarray data. Log-transformation of intensity-ratios is good for several rea-
sons. The simple ratio flattens all the underexpressed genes between 1 and 0. Log-
8 Normalization 89

transformation removes this bias. In statistical terms, log-transformed data gives a

more realistic sense of variation, and makes the variation of intensities and ratios of
intensities more independent of absolute magnitudes. Log-transformation also sta-
bilizes the variance of high intensity spots. It evens out highly skewed distributions
(e.g., makes more normal-like), and makes normalization (centralization) additive.
After log-transformation the expected mean of the dyes is 0 as opposed to 1 in the
case of plain intensity ratios.
Standardization is the process of expanding or contracting the distribution of
a statistic so that the experimental values can be compared with those from another
experiment. In statistical terms standardization means converting the intensity ra-
tios to Z-scores (see 6.5.1.), which are distributed as a standard normal distribution.
Most of the methods meant by normalization in the microarray field are cov-
ered by the term centralization. It is a process of moving a distribution so that
it is centered over the expected mean (balancing the two channels). For the log-
transformed intensity ratios, an intensity dependent centralization (e.g., lowess)
might help to correct the dye bias.

8.3.2 Per-chip and per-gene normalization

Per-chip normalization compensates for the experimental biases introduced on the
individual chips. It assumes that the median intensity stays relatively constant dur-
ing the experiment. If there are large differences due to the biological phenomenon,
you can accidentally remove the relevant variation in your data by normalization.
Per-gene normalization accounts for the difference in the detection efficiency
between different spots. It also enables us compare the relative gene expression
levels, as already pointed out. For two-color data, the per-gene normalization is
usually not used by default, because the calculation of intensity ratios corrects for
the same bias.

8.3.3 Global and local normalization

When normalizing the data, we estimate some descriptors of the data (e.g., the
mean). If only one descriptor is used for normalizing the whole data for one chip,
then a global normalization is performed.
Sometimes we have different estimates of means for different subarrays or for
different intensity ranges on the same chip. If these different estimates or a function
(nonlinear regression, i.e. lowess) is used for the normalization of the concomitant
subarrays or intensity ranges, the local normalization method is applied.

8.4 Performing normalization

8.4.1 Choice of the method
There are several normalization methods and little consensus about which one to
use, but here we will try to clarify the choice of methods and their applicability
with a few simple rules. The choice of the normalization method is coupled to
experimental design. Optimally, the method should be chosen before the actual
90 DNA microarray data analysis

experiment is conducted. However, the selection of the best normalization method

depends also very much on how the results turn out, e.g., the outcome of quality,
success of different steps in printing, hybridization, and scanning can be evaluated
only after the experiment has actually been done. Therefore, after inspecting the
quality of your data in general (see previous Chapter), it is also advisable to try
several different methods and see which one gives the most realible results for your
experiment. In any case, inside one experiment the same normalization scheme
should be applied uniformly to all chips.

8.4.2 Basic idea

The basic idea behind all the normalization methods is that the expected mean
intensity ratio between the two channels (two-color data) or two chips (one-color
data) is one, because only about 10–20% of all the genes in a cell are expressed at
any one time. If the observed mean intensity ratio deviates from one, the data is
mathematically processed in such a way that the final observed mean intensity ratio
becomes one. When the mean intensity ratio is adjusted to one, the distribution of
the gene expression is centered so that different chips (arrays) can be compared.
Often, when a big enough number of genes are plotted on a chip, it can be
assumed that most of them actually aren’t changing, and the mean intensity ratio
can be assumed to be one. If only a couple of dozen genes are plotted on the slide,
this assumption does not necessarily hold, and some other means of normalization
should be undertaken. This is especially true, if all the genes plotted on the slide
are relevant to the studied hypothesis.

8.4.3 Control genes

If the whole genome of the organism is printed on the microarray, we can assume
that most of the genes are not expressed, or are expressed in very low quantities. In
such cases all the genes on the chip can be used for normalization. If only a few
dozen or hundred of genes are present on the chip, some normalization standards
need to be added on the chip. Suitable controls are housekeeping genes and spiked
controls.
When using spiked controls, a known labeled DNA target is added to the la-
beled cDNA samples. This spiked DNA control hybridizes to its probe on the DNA
microarray chip. The spiked control should optimally be RNA or DNA from an-
other species than the studied one that do not cross-hybridize, but behaves similarly
to your sample RNA or DNA. The expected intensity ratio of the spiked controls
is one. If the observed ratio deviates from expected, the observed ratio is used for
normalization.
Housekeeping genes are now known not to be very reliable controls, because
their expression often changes during the experiment. However, it is often possible
to find a set of genes that do not change during a certain experiment. Using such a
set of non-changing genes may mean that you need to modify your arrays and re-do
your hybridizations after you have found the working set of genes you wish to use
as controls.
Controls are needed in microarrays as in any quality-checked laboratory exper-
8 Normalization 91

iment. Ideally, you would have several types of controls in different concentrations
scattered throughout the printed array. If you use Affymetrix products, chips con-
tain controls for the experiment and for the hybridization of the correct target (per-
fect match and mismatch, Chapter 3). More about controls can be read in Chapters
2 (Experimental design) and 7 (Preprocessing of data).

8.4.4 Linearity of data matters

There are several mathematical procedures and different algorithms (median cen-
tering, standardization, lowess smoothing) for normalization from which to choose
from. Before deciding on the method, the linearity of data should be checked (see
Chapter 7). If the data is linear, such procedures as median centering can be ap-
plied. And the other way around, median centering can only be applied to linear
data. If the data is non-linear, lowess smoothing or an other local method should
rather be applied. Moreover, if the data is normally distributed, it can be stan-
dardized. Also for certain purposes, for example clustering, the spread of the data
should be standardized.

8.4.5 Basic normalization schemes for linear data

In order to make comparisons between chips of the same experiment possible, every
single chip has to be normalized. In it’s simplest form this per-chip normalization
means median centering as described in section 8.5.2 and Table 8.1. In other words,
the median intensity (or expression) of every chip is brought to the same level.
Median centering does not change the spread of the data, which also means that the
original information content of the data is not altered. Median centering is often a
suitable choice for one-color datasets.
In addition to per-chip normalization, per-gene normalization can be applied.
This controls the hybridization efficiency differences between the individual probes.
For two-color experiments, the calculation of the intensity ratio corrects for differ-
ences between spots, and other normalization may not be needed. However, for
one-color experiments, per-gene normalization is usually applied.
One-color experiment can also be normalized against another one-color chip
(a reference chip), and handled thereafter as two-color experiment. With regard to
Affymetrix data, it is good to remember that the data has already been normalized
and scaled to some extent using specific algorithms (to calculate results for perfect
match and mismatches) (see Chapter 3).
If per-chip and per-gene normalizations are applied, both individual chips and
genes can directly be compared with each other. This is often the desired situation.

8.4.6 Special situations

Sometimes basic normalization schemes don’t perform well enough. If most of the
genes on a chip are likely to change, or there are spatial biases on the chip, more
sophisticated methods should be used. Again, global methods should not be used,
if the data is nonlinear, if there are spatial bias on the chip, or if the number of
expressed genes varies a lot between individual chips. In such cases, local methods
92 DNA microarray data analysis

should be used. As already mentioned, lowess smoothing is a a recipe for the

normalization of nonlinear data.
If the expression of most of the genes on the chip are expected to change during
the experiment, a median centering using all the genes is not a viable option. In such
cases spiked controls or housekeeping genes can be used for the normalization (see
8.5.8). These methods correct effectively for the differences between the channels,
but if the amount of mRNA hybridized to individual chips varies a lot, errors are
likely to occur.
Spatial biases can be corrected in multiple ways. Probably the easiest method
is to use median centering subarraywise. In other words, a new median is calculated
for every subarray (or a block of about 500 genes on a chip). This subarray specific
median is then used for the normalization of that subarray. The same principle can
be applied if the printing tip groups deviate from each other. Sometimes the left
side of the chip has lower expression values than the right side of the chip. Such
spatial biases can also be corrected using the median centering inside one probe
column or row.

8.5 Mathematical calculations

There are several ways to calculate the normalized intensity ratios. Here, we present
a few of the most commonly used ones. If you want to play around with these,
please check the logarithmic rules of calculation before proceeding. For example,
the mean centering (subtract the mean or median to get a mean or median of zero)
with log-transformed data is equivalent to mean scaling (divide with a certain num-
ber to get a mean or median of one) with untransformed data. The next formulas
are presented for the log ratios. Usually the normalization is calculated using only
a control channel or a reference chip.

8.5.1 Mean centering

Calculate the mean of the log ratios for one microarray. Produce the centered data
by subtracting this mean from the log ratio of every gene.

8.5.2 Median centering

Calculate the median of the log ratios for one microarray. Produce the centered
data by subtracting this median from the log ratio of every gene.

8.5.3 Trimmed mean centering

Remove the most deviant observations (5%) from the data for one microarray. Cal-
culate the median of the log ratios of the remaining genes. For centered data, sub-
tract this median from the log ratios of every gene.

8.5.4 Standardization

Equivalent to the Z-score calculation. See section 6.5.1.

8 Normalization 93

8.5.5 Lowess smoothing

Lowess-regression is often performed in a space of A versus M. This MA plot, also
called RI plot, is equivalent to the usual scatter plot, but the data cloud has been
tilted 45 degrees to the right. In other words, M is the log-transformed intensity
ratio, and A is the average intensity of the two channels or chips. Under and over-
expressed genes are much easier to find from this scatter plot, and the linearity of
the data is also easier to check. Thus, MA plots can reveal the intensity dependency
of log ratios, which appears as curvature, as well as extra variation at low intensity
spots (see Fig. 8.1.).

M = log2 (R) − log2(G)

log2 (G) + log2(R)
A=
2
After normalization the M and A can be transformed back to the dye intensities as
 1
2
R = 22A+M

 1
2
G = 22A−M
Lowess-normalization works as follows. A lowess-function smoothing is ap-
plied to the data and a curve is estimated by the sliding window method. The nor-
malized log-transformed intensity ratio is calculated by subtracting the estimated
curve from the original values. (Note: If a linear function is used for the local re-
gression, it is called lowess with a “w”. If a quadratic function is used for the local
regression, it is called loess without the “w”.)
Figure 8.2 gives an example of how the lowess-regression treats real life data.

Figure 8.2: Centralization with a lowess-regression. On the left, the log2 -normalized data
(centered around zero) in an A versus M plot. On the right, lowess-normalized data.

Linlog transformation can also be used to remove intensity-dependent varia-

tion. The linlog transformation combines linear and logarithmic transformations
through a smooth transition to take advantage of both methods, because the lin-
ear transformation of raw data is more appropriate for low intensity spots and the
94 DNA microarray data analysis

logarithmic transformation is more appropriate for high intensity spots (Cui et al.,
submitted). The linlog transformation can precede lowess smoothing and in that
way stabilize the variation due to the additive error dominant at low intensity, and
the multiplicative error dominant at high intensity in microarray data.

8.5.6 Ratio statistics

Data is assumed to follow a certain distribution. This distribution can then be used
to normalize the current data. The ratio statistics method is closely related to esti-
mating an error model from the data.

8.5.7 Analysis of variance

If there are data from several replications of the same gene, the ANOVA method
can be applied to centralize the data. With the ANOVA, many sources of error can
be taken into account at the same time.

8.5.8 Spiked controls

We assume that the intensity ratio from the spiked controls is one, if there is no
dye-incorporation bias. If the ratio deviates from one, this information can be used
for the calculation of the centered intensity values for the genes:

red intensity
green intensity =
spiked control intensity ratio

The housekeeping genes are used for normalization in a similar manner.

8.5.9 Dye-swap experiments

Dye-swap experiments can be normalized with the lowess-regression, when the
average of M and the average of A are plotted against each other as in the usual
lowess procedure. New labeling methods should remove the dye-incorporation
bias, but meanwhile this normalization method can be handy, especially in the cases
that indicate that the mRNA sequence may influence the labeling efficiency.
If we assume that the expression of any one gene in the original and the dye-
swap experiment is of equal magnitude but opposite signs, the normalization of
dye-swap experiments can be performed in a similar way as the normalization of
non-dye-swap experiments.
The normalized values for the dye-swap experiment can be then calculated as:
 
1 RG 
log2 − c,
2 G R

where R and G are the original and R’ and G’ are the dye-swap experiment intensi-
ties of the red and green channels, respectively. The normalization constant can be
estimated as  
1 R R
c= log2 + log2 
2 G G
8 Normalization 95

In other words, the average of the original and dye-swap chip is calculated
for every gene. Normalized values are calculated by subtracting the average of the
chips from the averaged genewise intensities.

8.6 Some caution is needed

Most of the aforementioned methods are used for per-chip normalization. Per-gene
normalization is not worthwhile if you only have a few chips, because then you
can potentially introduce errors to your data. This is because the mean, standard
deviation, and regression curves can not be effectively estimated from a very small
number of observations. As a guideline, the experiment should consist of at least
five chips, if you want to perform the per-gene normalization using mean or me-
dian centering. Even more observations (at least 10-25) are needed in order to use
standardization, regression or other advanced normalization method.
Furthermore, using very sophisticated normalization methods can lead to a
phenomenon called overfitting, in which case the model (e.g., a linear regression)
describes the variability of the data too well. This effectively removes biologically
relevant variation from the data, but it also introduces biases to the analysis. It is
common that the correlation of two chips (especially replicates) slightly decreases
after normalization. Whether this means that the biologically relevant information
has been removed, and some noise added, is currently unknown.
The mean and median centering do not usually influence the standard deviation
very much, but the regression and other more sophisticated methods do. Usually
they move towards a higher standard deviation, which might mean more noisy data.
Therefore, the normalization procedure should be as simple as possible, yet taking
the systematic errors into account.

8.7 Graphical example

A couple of examples how the normalization procedure affects the graphical rep-
resentation of the data are presented in Figure 8.3. These examples showes, how
the number of genes affects the results. In the image a diminishing series of house-
keeping genes is used for normalization.

8.8 Example of calculations

A short example of mean centering procedures is presented in Table 8.1. Mean

centering can be applied either as per-chip normalization or per-gene normalization.
With Affymetrix data, both procedures are often applied. If both procedures are
applied, the per-chip centering should be performed before the per gene centering.
Note that the standard deviation is not affected by mean centering.
96 DNA microarray data analysis

Figure 8.3: The four images represent the same data processed with per-chip mean cen-
tering. The number of genes used for the mean calculation has been varied. From top left
to bottom right: 5776 genes, 57 genes, 577 genes, and 6 genes. Note the varying slope of
the linear regression line fitted to the normalized data.

8.9 Using GeneSpring for normalization

GeneSpring offers the basic repertoire of normalization methods, including both
per-chip and per-gene methods. GeneSpring includes calculation of intensity ratios
and dye-swap calculations as normalization methods. Dye-swap normalization in
GeneSpring simply calculates the intensity ratio as (Cy5 / Cy3) instead of the usual
(Cy3 / Cy5) for the specified samples.
Normalization methods in GeneSpring 5.x are

• lowess-smoothing (per-chip and per-gene)

• median polishing (per-chip and per-gene)

• median or percentile scaling (per-chip or per-gene)

Also housekeeping genes or spiked controls can be used for normalization

(Per-chip: normalize to positive control genes) . If chips have been scaled in the
Affymetrix MAS program to a certain mean intensity value, the transformational
effect can be removed in GeneSpring using Per-chip: Normalize to a constant value
option. If one-color data is analyzed, and the experiment includes one control chip
that is used for the calculation of intensity ratio, this can be done in GeneSpring
8 Normalization 97

Table 8.1: An example of per-chip and per-gene mean centering. Because of rounding
errors, the results presented in the table are not necessarily products of individual observa-
tions and their means.

Uncentered data
chip1 chip2 chip3 chip4 mean standard dev.
gene1 2.12 2.01 4.37 2.01 2.63 1.16
gene2 2.20 2.06 4.32 2.03 2.65 1.11
gene3 2.18 1.90 4.37 1.90 2.59 1.20
gene4 2.15 1.92 4.38 1.89 2.59 1.20
gene5 2.14 2.00 4.52 1.99 2.66 1.24
gene6 1.93 2.02 4.18 2.01 2.54 1.09
gene7 2.26 1.96 4.19 1.98 2.60 1.07
gene8 2.07 2.00 4.39 2.01 2.62 1.18
gene9 2.25 2.06 4.34 2.04 2.67 1.12
gene10 1.95 1.76 3.97 1.82 2.37 1.07
mean 2.13 1.97 4.30 1.97
standard dev. 0.11 0.09 0.15 0.07

Per-chip mean centering

chip1 chip2 chip3 chip4 mean standard dev.
gene1 0.00 0.05 0.07 0.04 0.04 0.03
gene2 0.08 0.09 0.02 0.07 0.06 0.03
gene3 0.06 -0.07 0.07 -0.07 -0.01 0.08
gene4 0.03 -0.05 0.08 -0.08 0.00 0.07
gene5 0.01 0.03 0.21 0.02 0.07 0.10
gene6 -0.19 0.05 -0.13 0.04 -0.06 0.12
gene7 0.13 -0.01 -0.11 0.01 0.00 0.10
gene8 -0.05 0.03 0.09 0.05 0.03 0.06
gene9 0.13 0.09 0.04 0.07 0.08 0.04
gene10 -0.18 -0.21 -0.33 -0.15 -0.22 0.08
mean 0.00 0.00 0.00 0.00
standard dev. 0.11 0.09 0.15 0.07

Per-gene mean centering

chip1 chip2 chip3 chip4 mean standard dev.
gene1 -0.51 -0.61 1.74 -0.62 0.00 1.16
gene2 -0.45 -0.59 1.66 -0.62 0.00 1.11
gene3 -0.40 -0.69 1.78 -0.69 0.00 1.20
gene4 -0.43 -0.67 1.80 -0.70 0.00 1.20
gene5 -0.52 -0.66 1.86 -0.67 0.00 1.24
gene6 -0.60 -0.52 1.64 -0.52 0.00 1.09
gene7 -0.34 -0.64 1.60 -0.62 0.00 1.07
gene8 -0.54 -0.62 1.77 -0.61 0.00 1.18
gene9 -0.42 -0.61 1.67 -0.63 0.00 1.12
gene10 -0.43 -0.61 1.60 -0.56 0.00 1.07
mean -0.47 -0.62 1.71 -0.62
standard dev. 0.08 0.05 0.09 0.06
98 DNA microarray data analysis

using Per-gene: Normalize to specific samples option. Furthermore, normalization

can be applied only to certain samples, if needed.
GeneSpring automatically detects the imported data type (one- or two-color),
and suggests a normalization scheme for the experiment. These schemes are fol-
lowing
For one-color data (Affymetrix or nylon filter):

• Data transformation: Set measurements less than 0.0 to 0.0

• Per-chip: Normalize to 50th percentile

• Per-gene: Normalize to median

For two-color data (cDNA microarrays) either:

• Per-chip and per-gene: Intensity dependent (lowess) normalization

or

• Per spot: divide by control channel

• Per-chip: Normalize to 50th percentile

Normalization methods are accessed through Experiments -> Experiment nor-

malizations in GeneSpring. GeneSpring gives a warning if the normalization you
are trying to perform does not make any sense (red text), or that the calculation
can be performed, but the results might be unreliable (orange text). Note that the
normalization methods are applied in the same order as they appear in the list.

8.10 Suggested reading

1. Anonymous. Presentations from Normalization working group, MGED4
meeting, Boston, February 2002. See latest information from the MGED
Data Transformation and Normalization Working Group’s home page at http:
//www.dnachip.org/mged/normalization.html.

2. Beißbarth, T., Fellenberg, K., Brors, B., Arribas-Prat, R., Boer, J., Hauser,
N. C., Scheideler, M., Hoheisel, J. D., Schutz, G., Poustka, A., and Vingron,
M. (2000) Processing and quality control of DNA array hybridization data.
Bioinformatics 16, 1014-22.

3. Bolstad, B. M., Irizarry R. A., Astrand, M., and Speed, T. P. (2003), A Com-
parison of Normalization Methods for High Density Oligonucleotide Array
Data Based on Bias and Variance. Bioinformatics 19, 185-193.

4. Brazma, A., and Vilo, J. (2000) Gene expression data analysis. FEBS Lett.
480, 17-24.

5. Cui, X, Kerr, M. K., and Churchill, G. A.. Data Transformations for cDNA
Microarray Data. Statistical Applications in Genetics and Molecular Biol-
ogy, submitted.
8 Normalization 99

6. Dudoit, S., Y. H. Yang, M. J. Callow, and Speed, T. P.. Statistical methods

for identifying differentially expressed genes in replicated cDNA microarray
experiments (Statistics, UC Berkeley, Tech Report # 578).

7. Eickhoff, B., Korn, B., Schick, M., Poustka, A., and van der Bosch, J. (1999)
Normalization of array hybridization experiments in differential gene expres-
sion analysis. Nucleic Acids Res. 27, e33.

8. Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis,

K. J., Scherf, U., Speed, T.P. (2002) Exploration, Normalization, and Sum-
maries of High Density Oligonucleotide Array Probe Level Data, submitted.
http://www.stat.berkeley.edu/users/terry/zarray/Affy/affy_
index.html

9. Quackenbush J. (2001) Computational analysis of microarray data. Nat Rev

Genet. 2, 418-427.

10. Schuchhardt, J., Beule, D., Malik, A., Wolski, E., Eickhoff, H., Lehrach, H.,
and Hanspeter, H. (2000) Normalization strategies for cDNA microarrays.
Nucleic Acids Research 28, e47.

11. Sherlock, G. (2001) Analysis of large-scale gene expression data. Briefings

in Bioinformatics 2, 350-362.

12. Yang, Y. H., Dudoit, S., Luu, P. and Speed, T. P. (2001) Normalization for
cDNA Microarray Data. Proceedings of Photonics West 2001: Biomedi-
cal Optics Symposium (BiOS),The International Society of Optical Engi-
neering (SPIE), San Jose, California. http://www.stat.berkeley.edu/
users/terry/zarray/Html/normspie.html

13. Yang, Y. H., Dudoit, P., Luu, P., and Speed, T.P. (2001) Normalization for
cDNA microarray data, submitted.

This chapter was written by Jarno Tuimala, Ilana Saarikko, and M. Minna
Laine.
100 DNA microarray data analysis

9 Finding differentially
expressed genes
After removing the bad quality data we are left with reliable data. As we have
already seen, good quality data can be further filtered so that only the genes that
show some changes in the expression during the experiment are preserved in the
dataset. Often the differentially expressed or otherwise interesting genes are stored
as simple lists of gene names, i.e., genelists. Most typically genelists are created
by driving the data through a certain filter, but they can also include lists of genes,
which have been produced by clustering analyses or other more complex methods.

9.1 Identifying over- and underexpressed genes

Sometimes it is sufficient to get information about genes that are either under- or
overexpressed during the experiment. For example, we might be interested in genes
that have an elevated expression because of a drug treatment. Such genes are most
easily found by simple filtering. Simple filtering (by absolute expression change)
can even be used for experiments, where there are no replicates.

9.1.1 Filtering by absolute expression change

If the log-transformed data is used for analysis, the overexpressed genes have an
expression above zero and underexpressed genes have an expression below zero.
However, often the experimental errors are of the order of two, and for finding the
over- and underexpressed genes the cutoffs of -1 and 1 are used. All the genes
which fulfill the filter criteria are saved in a new genelist.
The method can also be used for filtering by expression duration. If the ex-
pression change has been at least x for a duration of y, the gene names are saved
into a new genelist.
This kind of genelist can be created in any spreadsheet program, and the actual
data-analysis can be very simple to perform.

9.1.2 Statistical single chip methods

The crude filtering by absolute expression change is not always the optimal method
for finding the differentially expressed genes, because the information about the
reliablity of the expression change is lacking. In the absence of replicates, there
are still a few statistical methods that can be used, if we want to have statistical
9 Finding differentially expressed genes 101

significance values for expression changes. However, using such methods carries
the risk that the most unstable probes or mRNAs are identified as differentially
expressed.
In the next four sections we will go through some of these methods.

9.1.3 Noise envelope

Noise envelope means simply a method where the standard deviation (calculated
with sliding window) is used as a cut-off for under and over-expressed genes.
The variance in the reported gene expression level is a function of the expres-
sion level so that the variance is higher in the lower expression values and lower
when the expression is high. Thus, the identification of the differentially expressed
genes calls for a statistical model of the variance or noise. The simplest model
could be constructed on the basis of the standard deviation of expression values.
Such a model can be constructed relatively easily using statistical programs.
The method relies on a segmental (sliding window) calculation of standard devia-
tion. A datapoint refers to an (x, y) paring, where x is the absolute intensity value
of a gene from the hybridization, and y is the corresponding intensity ratio value.
Using all data points in a given sliding window of expression values, the standard
deviation of the intensity ratios is calculated. The average of expression values
within the window are then paired with the average intensity ratio value within the
same window plus the number of standard deviations (usually 2 or 3) specified by
the experimenter. This new pair becomes the candidate upper noise envelope point.
Similarly a candidate lower noise envelope points are determined.
Using these noise envelopes, the insignificant expression changes are removed:
The region between the upper and lower noise envelopes contains genes with in-
significant changes.

9.1.4 Sapir and Churchill’s single slide method

After logarithmic transformation and normalization, Sapir and Churchill’s method

fits a linear regression function to a scatter plot of sample intensity versus control
intensity to yield orthogonal residuals. These residues represent the gene expres-
sion changes in the experiment. The expectation-maximization (EM) algorithm
applying the Bayes’ rule is used to obtain estimates of the proportion of differen-
tially expressed genes. Using the Bayes’ rule and information on the residuals, the
(posterior) probabilities of differential expression are calculated.
The method is implemeted in the R language package sma. The results can be
obtained either as a genelist or in a form of a scatter plot. The scatter plot example
is given here (Figure 9.1). The residuals are modeled as a uniform distribution, so
the cut offs are constant for all intensity values.
102 DNA microarray data analysis

1.5
1.0
0.5
M

0.0
−0.5
−1.0
−1.5

8 10 12 14

Figure 9.1: A scatter plot with Sapir and Churchill’s method with the cut off lines marked.
The vertical lines represent the cut offs for up- and downregulated genes. Note that the axes
of the plot are normalized log ratio (M) and average intensity (A).

9.1.5 Chen’s single slide method

The idea central to Chen’s method is that the gene expression levels are determined
by the intrinsic properties of each gene, which means that the expression levels vary
widely among genes. Therefore it is inappropriate to pool statistics on gene expres-
sion differences across the microarray. Assuming that the green and red channel
intensities are normally distributed and have similar coefficients of variation and a
similar mean, the confidence intervals for actual differential expression are calcu-
lated using the data derived from the maximum likelihood estimate of the coeffi-
cient of variation and a set of genes, which do not show any expression change on
the chip.
The method is implemeted in the R language package sma. The results can be
obtained either as a genelist or in a form of a scatter plot. The scatter plot example
is given here (Figure 9.2).
9 Finding differentially expressed genes 103

1.5
1.0
0.5
M

0.0
−0.5
−1.0
−1.5

8 10 12 14

Figure 9.2: A scatter plot with Chen’s method with the 95% (inner, light blue) and 99%
(outer, dark blue) cut off lines marked. The vertical lines represent the cut offs for up-
and downregulated genes. Note that the axes of the plot are normalized log ratio (M) and
average intensity (A).

9.1.6 Newton’s single slide method

Newton’s method models the measured expression levels using terms that account
for the sources of variation. The first obvious source is measurement error. The
second source of variation is due to the different genes spotted onto the microarray.
Newton’s method can be viewed as a hybrid of Chen’s and Sapir and Churchill’s
methods, because it models the variability on the slide very similarly to Chen’s
method, but the mathematical calculations are done with the EM-algorithm similar
to the one Sapir and Churchill used. Results are presented as log-odds ratios that
the gene is differentially expressed.
The method is implemeted in the R language package sma. The results can
be obtained either as a genelist or in the form of a scatter plot. The scatter plot
example is given here (Figure 9.3).
104 DNA microarray data analysis

1.5
1.0
0.5
M
0.0
−0.5
−1.0
−1.5

8 10 12 14
A

Figure 9.3: A scatter plot with Newton’s method with the log-odds ratios 0, 1 and 2 cut
off lines marked. The vertical lines represent the cut offs for up- and downregulated genes.
Note that the axes of the plot are normalized log ratio (M) and average intensity (A).

9.2 What about the confidence?

Say, we have found a gene that is threefold upregulated after the drug treatment.
How do we know that the result is not just an experimental error? We need to de-
termine the statistical significance of the upregulation of the gene. Significance can
only be assessed, if replicate measures of the same gene were performed during the
experiment. The chips (or expression of invidual genes on them) can be compared
by the t-test (two conditions) or ANOVA (more than two conditions).

9.2.1 Only some treatments have replicates

If the gene expression has been measured for control and treatment, but only treat-
ment has been replicated, the experimental error can be crudely estimated from
the variation between the replicates. The standard deviation is determined for ev-
ery gene, and the expression change is compared with the standard deviation. The
more the change exceeds the standard deviation between replicates, the more sig-
nificant the gene is. Note that this method does not allow for the calculation of
p-values.
9 Finding differentially expressed genes 105

9.2.2 All the treatments have replicates: two-sample t-test

The statistical basis for the t-test has been covered in detail elsewhere (see 6.9), but
briefly, the two-sample t-test looks at the mean and variance of the two distributions
(say, control and treatment chip log ratios), and calculates the probability that they
were sampled from the same distribution. The t-test can be applied successfully in
situations, where both the control and treatment have been repeated.
Here (Figure 9.4), we present an example of a graphical t-test result, which
has been produced using R language supplemented with package sma.

T plots
t vs. average A t denominator vs. average A

*********** ** * * * t denominator
0

0.8
****** * *** ***** * ** *
t

* * ** *
−20

0.0 *************** ***** ****** ***** *

8 9 11 13 15 8 9 11 13 15

average A average A

|t numerator| vs. average A t denominator vs. |t numerator|

t denominator
0.0 1.5 3.0
|t numerator|

*
0.8

*
* * * ** **
* * * *
* ************ **** *** * * **** *
*
************* * *
0.0

******
8 9 11 13 15 0.0 1.0 2.0 3.0

average A |t numerator|

Figure 9.4: Package sma automatically produces these four plots for t-test results. The
genes that have an expression pattern most significantly deviating from zero are high-
lighted with a colored star.

After calculating the t-test p-values for the replicated genes, the ones with the
lowest p-value (marked with a star in the images in Figure 9.4) can be saved into
a new genelist and used in further analyses, for example cluster analysis. These
are the genes that most significantly differ between two conditions, say control and
106 DNA microarray data analysis

treatment mice.

9.2.3 All the treatments have replicates: one-sample t-test

Up- and downregulated genes can be more effectively found, if the chips are repli-
cated. In such cases, the one-sample t-test for the deviation from zero (expected
mean) can be performed. Finding the differentially expressed genes is even more
easier, if the so called Volcano plot is produced (Figure 9.5). This plot of log ratio
versus statistical significance shows also which is the lowest expression change we
can be statistically confident in. In our example, 2-fold down-regulated genes have
a p-value of 0.01. In contrast, 1.4-fold upregulated genes have a similar p-value.
This indicates that new chip technology coupled with an appropriate number of
replicates can produce reliable data even for very small expression changes.
4
3
−log10(p)

2
1
0

−4 −2 0 2 4

log2(ratio)

Figure 9.5: Volcano plot, where the statistical significance by one-sample t-tests
[-log10(p)] is plotted against normalized log ratio [log2(ratio)]. Vertical dashed lines rep-
resent a 2-fold difference between the control and the sample. Horizontal dashed line
represents the t-test p-value of 0.01, and the solid horizontal line represent p=0.001.

9.3 GeneSpring examples

An example on how to generate genelists with GeneSpring filtering tools is pre-
sented in section 7.15.6.
9 Finding differentially expressed genes 107

Examples of statistical testing are presented in sections 6.11.6 and 6.11.7.

9.4 Suggested reading

1. Chen, Y., Dougherty, E. R., and Bittner, M. L. (1997) Ratio-based deci-
sions and the quantitative analysis of cDNA microarray images. Journal of
Biomedical Optics 2, 364-374.

2. Sapir and Churchill (2000), Estimating the posterior probability of differen-

tial gene expression from microarray data, http://www.jax.org/research/
churchill/.

3. Newton, M. N., Kendziorski, C. M., Richmond, C. S., Blattner, F. R., and

Tsui, K. W. (1999) On differential variability of expression ratios: Improv-
ing statistical inference about gene expression changes from microarray data.
Technical Report 139, Department of Biostatistics and Medical Informatics,
UW Madison, http://www.stat.wisc.edu/∼newton/papers/abstracts/
btr139a.html.

4. SMA-package for R, http://www.stat.berkeley.edu/users/terry/zarray/

Software/smacode.html.

This chapter was written by Jarno Tuimala.

108 DNA microarray data analysis

10 Cluster analysis of
microarray information
10.1 Basic concept of clustering

Microarray experiments generate mountains of data, which has to be stored and

analysed. For humans it is difficult to handle very large numeric data sets. There-
fore a general concept is to try to reduce the dimensionality of the data. A number
of clustering techniques have been used to group genes based on their expression
patterns. The user has to choose the appropriate method for each task. The basic
concept in clustering is to try to identify and group together similarly expressed
genes and then try to correlate the observations to biology. The idea is that coregu-
lated and functionally related genes are grouped into clusters. Clustering provides
the framework for this analysis. The hard part is to analyze the biological processes
and consequences. However, clustering can be a very useful tool. The other side of
the coin is the visualization of the information.
Before clustering a large number of requirements have to be met. The data has
to be of good quality, the chip design should be correct, one should have interesting
genes contained in the chip, sample preparation has to be flawless, data has to
be properly treated (outliers, normalization etc). Clustering will not improve the
quality of the data. If poor-quality data is used then also the outcome is useless:
garbage in, garbage out.

10.2 Principles of clustering

Clustering organizes the data into a small number of (relatively) homogeneous

groups. Usually normalization of the expression values is used. At this stage, it
is of interest to look at the changes in expression patterns, not to follow the actual
numeric changes. Thus, the methods are used to find similar expression motifs ir-
respective of the expression level. Therefore, both low and high expression level
genes can end up in the same cluster if the expression profiles are correlated by
shape.
The majority of clustering methods has been available already for long. During
the last few years, some new microarray analysis dedicated methods have been
developed, too. The methods can be described and classified in different ways.
It is not possible to go into the details of all the methods here, so only the most
commonly used methods are described.
10 Cluster analysis of microarray information 109

Clustering methods can be grouped as supervised and unsupervised. Super-

vised methods assign some predefined classes to a data set, whereas in unsuper-
vised methods no prior assumptions are applied.
Hierarchical clustering, K-means, self-organizing maps (SOMs), and principal
component analysis (PCA) have been commonly used. There are also other meth-
ods, such as multidimensional scaling (MDS), minimum description length (MDS),
gene shaving (GS), decision trees, and support vector machines (SVMs).

10.3 Hierarchical clustering

Hierarchical clustering is a statistical method for finding relatively homogeneous
clusters. The hierarchical clustering algorithm either iteratively joins the two clos-
est clusters starting from single clusters (agglomerative, bottom-up approach) or
iteratively partitions clusters starting from the complete set (divisive, top-down ap-
proach). After each step, a new distance matrix between the newly formed clusters
and the other clusters is recalculated. If there are N cases, N-1 clustering steps are
needed. There are several methods of hierarchical cluster analysis including

• single linkage (minimum method, nearest neighbor)

• complete linkage (maximum method, furthest neighbor)

• average linkage (UPGMA).

For a set of N genes to be clustered, and an NxN distance (or similarity) matrix,
the hierarchical clustering is performed as follows:

1. Assign each gene to a cluster of its own.

2. Find the closest pair of clusters and merge them into a single cluster.

3. Compute the distances (similarities) between the new cluster and each of the
old clusters.

4. Repeat steps 2 and 3 until all genes are clustered.

Step 3 can be performed in different ways depending on the chosen approach.

In single linkage clustering , the distance between one cluster and another is con-
sidered to be equal to the shortest distance from any member of one cluster to any
member of the other cluster. In complete linkage clustering, the distance between
one cluster and another cluster is considered to be equal to the longest distance
from any member of one cluster to any member of the other cluster. In average
linkage, the distance between one cluster and another cluster is considered to be
equal to the average distance from any member of one cluster to any member of
the other cluster. The hierarchical clustering can be represented as a tree, or a den-
drogram. Branch lengths represent the degree of similarity between the genes. The
method does not provide clusters as such. Conceptually, different clusters and sizes
of clusters can be obtained by moving along the trunk or branches of the tree and
deciding on which level to put forth branches (cut the tree).
110 DNA microarray data analysis

Hierarchical clustering is often applied in the analysis of patient samples to

organize the data based on the cases. Indeed, for patient samples the hierarchical
clustering method is most often the best option. Usually, in addition to patient-
based organization, genes are also clustered by applying two-way clustering. On
one axis are the samples (patients) and on the other axis the genes (Figure 10.1).

Figure 10.1: An example of hierarchical clustering.

10.4 Self-organizing map

Kohonen’s self-organizing map (SOM) is a neural net that uses unsupervised learn-
ing for which no prior knowledge of classes is required. SOMs are usually used
to visualize and interpret large high-dimensional data sets. In SOM, every input is
connected to every output via connections with variable weights. Also, the output
nodes are highly interconnected. SOM tries to learn to map similar input vec-
tors (gene expression profiles) to similar regions of the output array of nodes (Fig-
ure 10.2). The method maps the multidimensional distances of the feature space to
two-dimensional distances in the output map. The SOM algorithm is iterative.
The map attempts to represent all the available observations with optimal accu-
racy using a restricted set of models. At the same time, the models become ordered
on the grid so that similar models are close to each other and dissimilar models
far from each other. So the order and organization of the nodes (tentative clusters)
contain more information than just the actual partition of genes to clusters.
In SOMs, the number of clusters has to be predetermined. The value is given
by the dimensions of the two-dimensional grid or array. One has to experiment
with the actual number of clusters. Most programs facilitate the analysis of all the
genes within a cluster or clusters. One should find such array dimensions that there
10 Cluster analysis of microarray information 111

is a minimum number of poorly fitting genes in the clusters. The actual number of
clusters is also difficult to assign, but the square root of the number of genes is a
good initial estimate. However, it depends on the data set.

Figure 10.2: An example of self-organizing map.

10.5 K-means clustering

K-means is a least-squares partitioning method for which the number of groups, K,
has to be provided. The algorithm computes cluster centroids and uses them as new
cluster seeds, and assigns each object to the nearest seed (Figure 10.3). However,
it is also possible to estimate K from the data, taking the approach of a mixture
density estimation problem.

1. The genes are arbitrarily divided into K centroids. The reference vector i.e.,
location of each centroid, is calculated.

2. Each gene is examined and assigned to one of the clusters depending on the
minimum distance.

3. The centroid’s position is recalculated.

4. Steps 2 and 3 are repeated until all the genes are grouped into the final re-
quired number of clusters.

During the course of iterations, the program tries to minimize the sum, over
all groups, of the squared within-group residuals, which are the distances of the
objects to the respective group centroids. Convergence is reached when the objec-
tive function (i.e. the residual sum-of-squares) cannot be lowered any more. The
obtained groups are geometrically as compact as possible around their respective
centroids (Figure 10.4).
K-means partitioning is a so-called NP-hard problem (there is no known algo-
rithm that would be able to solve the problem in polynomial time), thus there is no
guarantee that the absolute minimum of the objective function has been reached.
112 DNA microarray data analysis

Therefore, it is good practise to repeat the analysis several times using randomly
selected initial group centroids, and check whether these analyses produce compa-
rable results.

Figure 10.3: K-means algorithm. Genes are initially divided into K clusters, and the final
clusters are iterated from these.

Figure 10.4: An example of k-means clustering with four clusters.

10.6 Principal component analysis

Objectives of principal component analysis (PCA) are to discover or to reduce the
dimensionality of the data set and to identify new meaningful underlying variables.
PCA transforms a number of (possibly) correlated variables into a (smaller) num-
10 Cluster analysis of microarray information 113

ber of uncorrelated variables called principal components (Figure 10.5). The basic
idea in PCA is to find the components that explain the maximum amount of vari-
ance possible by n linearly transformed components. The first principal component
accounts for as much of the variability in the data as possible, and each succeeding
component accounts for as much of the remaining variability as possible.
PCA can be also applied when other information in addition to the actual ex-
pression levels is available (this applies to SOM and K-means methods as well).
The method provides insight into relationships, but no matter how interesting the
results may be, they can be very difficult if not impossible to interpret in biological
terms.

Figure 10.5: An example of principal component analysis. The two most significant
principal components have been selected as the axes of the plot.

10.7 Pros and cons of clustering

There are no objective rules for determining the correct number of gene expression
clusters. Therefore, an extensive manual analysis is required to find out the optimal
number of clusters. The expression patterns can be visually inspected, and the
distribution of every gene from the centroid analyzed along with the variation from
the other genes within clusters. In suboptimal clusters, some genes do not follow
the general expression pattern within a cluster due to clearly different expression
profiles. It is important to find out a working solution for the number of clusters,
because the subsequent analyses and data mining rely on the partition obtained
by clustering. The only exceptions are methods using supervised learning, i.e., in
114 DNA microarray data analysis

addition to clustering information other data is used to organize the genes. The
problem with supervised methods is that they may force the data to behave in a
certain way. After all, we cannot know what is the correct partitioning for a given
data set. The usefulness of the data for interpretation of biological processes is in
fact the true test for the correctness of clustering.
Clustering methods provide a relatively easy way to organize the cluster in-
formation. Together with visualization methods they allow for the user an intuitive
way of looking at, understanding and analyzing the data. Different clustering meth-
ods and the same method with different premises produce different end results, so
the user has to try to find out a useful result. In many articles the analysis is finished
at clustering. This is a mistake, since clustering is actually only a starting point for
more detailed and interesting studies.
For the clustering methods to work properly, the data should be well separated
meaning that clusters are easily separable. However, this is usually not the case in
microarray studies. Diffuse or interpenetrating groups cause problems in the de-
termination of cluster boundaries during the clustering, leading to different results
when using different methods and even the same method with different parameters.
Therefore, the user has to consider clusters as a working partitioning, not as the
final truth.
Single linkage is not the best approach for hierarchical clustering, because the
result can be remarkably skewed. The method is good for picking outliers that are
connected in the very last steps of the process. Complete linkage tends to produce
very tightly packed clusters. The method is very sensitive for the quality of the
data. K-means is one of the simplest and fastest clustering methods. The result is
dependent on the initial location of centroids. This problem can be overcome by
applying a number of intialization approaches.

10.8 Visualization

Gene expression experiments may include millions of individual data items. It is

not possible to comprehend such an overwhelming amount of information without
proper visualization methods. Scatter plots are very useful for the initial analysis
and comparison of data sets (Figure 10.6).
It is customary to present the clustering results by grouping genes of clusters
next to each other (Figure 10.7). In SOM based analysis, also the order of the clus-
ters contains information about the relationships between clusters. For the manual
analysis of the goodness of clustering, one has to look at the expression patterns
within the generated clusters. Genes within a cluster should follow the average ex-
pression pattern of the cluster. Because every gene has to be assigned to a cluster,
genes with unique expression patterns do not fit well in any group. The reason for
this behavior may be truly different expression profile or it may originate from ex-
perimental errors or difficulties. It is impossible to find the reason without further
experimental tests unless the behavior is consistent in independent experiments.
10 Cluster analysis of microarray information 115

Figure 10.6: An example of scatter plot.

Figure 10.7: An example of expression pattern within one cluster.

In microarray studies, usually a red/green color scheme is applied. The data

are shown in a matrix format, with each row representing all the hybridization re-
sults for a single gene of the array, and each column representing the measured
expression level for all genes at a single time point. To visualize the results, the
expression level of each gene is represented by a color, red representing overex-
116 DNA microarray data analysis

pression and green underexpression within each row. The color intensity represents
the magnitude of deviation. It is naturally possible to use other colors, too, and it
has been recommended to use a blue/yellow scheme instead, and avoid the prob-
lems with color blind persons. Most visulization packages allow free choice of the
colors.
Red/green figures can be found from most microarray analysis articles. These
figures provide a general overview of the data. With large data sets, it is not possible
to identify or even mark individual genes into such figures. Often either more de-
tailed figures of certain types or groups of genes are provided or the whole data is in
a (web accessible) appendix. From the beginning of 2003, major journals have re-
quired the data to be submitted to data repositories (see chapter on MIAME/MGED
system).

10.9 Programs for clustering and visualization

Several programs are available for clustering and visualization of gene expression
patterns (Table 10.1). Here we present an incomplete list of programs that we
have found useful. Many programs are freely available on the Internet. Both the
GeneSpring and Kensington package available at CSC contain several options for
clustering. Despite the initial time required for learning the use of these programs,
they provide some benefits, mainly due to allowing several analyses to be done
within a single package. These programs are not as such any better than freely
available programs. They, however, might have a more user-friendly interface.

Table 10.1: Freely available software for cluster analysis and visualization.

Program Description
Cluster Performs hierarchical clustering, self-organizing maps
SAM Significance Analysis of Microarrays: Supervised learning
ScanAlyze Processes fluorescent images of microarrays
TreeView Graphical results of analyses from Cluster
Expression Profiler Analysis and clustering of gene expression data
GeneCluster Self-organizing maps
J-Express Clustering and visualization

Program Author Platform

Cluster Michael Eisen Windows
SAM Rob Tibshirani Excel Add-in
ScanAlyze Michael Eisen Windows
TreeView Michael Eisen Windows
Expression Profiler EBI Web
GeneCluster Whitehead Institute / MIT Java
J-Express MolMine Java
10 Cluster analysis of microarray information 117

10.10 Function prediction

Genes that fall into the same cluster might have a similar transcription response to a
certain treatment. It is likely that some common biological function or role is acting
in the background. If the cluster has a bunch of genes with known and similar func-
tions, the characteristics of unknown genes in the cluster can be inferred. Function
prediction can be assisted with additional data on sequence similarity, phylogenetic
inference based on sequences, posttranslational modifications, cellular destination
signals, and so on. Taken together, a sufficiently large number of common features
can be used to make fairly accurate predictions of gene functions.

10.11 GeneSpring and clustering

GeneSpring offers five different clustering and classification algorithms, one of

which is a supervised method. These are hierarchical clustering (gene and ex-
periment trees), K-means clustering, self-organizing maps, principal component
analysis, and parameter value prediction. Clustering tools are invoked from the
menu Tools->Clustering, the supervised classification method can be find from
Tools->Predict parameter values, and the principal component analysis is located
in Tools->Principal components analysis.

10.11.1 Clustering tool

The clustering tool has four fields, which need to be filled in before running the
analysis (Figure 10.8). From top to down, the first box indicates the gene list to
be clustered. Initially the list is the same that was highlighted when the clustering
tool was invoked, but it can be changed from the navigator bar on the left. Next
box contains the information about the experiment to be clustered. This can also be
easily changed from the navigator bar on the left. The pull-down menu offers the
choice of three clustering algorithms (hierarchical, K-means and SOM). The setting
for the current analysis is in the box below the pull-down menu. For K-means, the
number of clusters needs to specified, as well as the number of iterations and the
desired measure of similarity.
GeneSpring has several different similarity measures, which fall into the fol-
lowing categories: correlation, confidence, and distance. The selection of the sim-
ilarity measure should be given some thought, because it significantly affects the
generated results. Pearson’s correlation emphasizes both over- and underexpressed
genes, and the Standard correlation finds especially overexpressed genes. Spear-
man’s correlation is highly similar to Pearson’s correlation except it uses ranks for
the calculation of the correlation coefficient (and is thus a nonparametric measure
of correlation). Distance measures the euclidian distance between two gene expres-
sion profiles. It is calculated as the square root of averaged squared deviation of the
profiles. Spearman’s confidence measures the probability of getting a correlation
of S or higher by chance alone, if the true correlation is zero.
If there is only one measurement per gene (i.e., one chip), only the distance
measure can be used for the clustering of the genes. If there are two replicates of
118 DNA microarray data analysis

every gene, the Standard correlation can also be used. If there are three replicates,
Pearson’s correlation becomes available, and with five replicates the confidence
measures can be used.
In other words, the decision about the applied similarity measure depends on
the biological question you are interested in, and the amount of replicates in your
dataset.
After specifying the aforementioned settings, the run can be started by click-
ing the Start-button on the bottom of the window. When the run has ended, you
can name and save the clustering result. The result appears on the main screen of
GeneSpring. Thereafter, all the results can be found from the navigator bar un-
der the folder classification. Note that the hierarchical clustering results are stored
under two different folders, Gene Trees and Experiment Trees.
After viewing the clustering results, you can get back to the original view by
selecting View->Unsplit window. Hierarchical clustering results can be dismissed,
for example, by selecting View->blocks.

Figure 10.8: The clustering tool in GeneSpring.

10.11.2 Principal components analysis tool

Principal component analysis can be used for creating a set of the most significant
expression patterns from the data. It can also be used for checking the results
of the clustering methods. After selecting Tools->Principal components method
the analysis is run, and the results are displayed. The opening window contains
the most significant profiles. Double-clicking one profile transfers the view to the
Gene Inspector, where genes with a similar expression profile can be searched.
The results are displayed in a scatter plot format in the main window. If you
10 Cluster analysis of microarray information 119

want to compare PCA with some clustering results, you can right-click on one
clustering result in the navigator bar, and select Set as coloring scheme from the
appearing menu. A good clustering result often creates clusters that do not overlap
with each other in the PCA scatter plot.

10.11.3 Predict parameter value tool

The predict parameter value tool (Figure 10.9) is used in situations, where we have
a certain set of known samples, and based on these, we want to predict in which
group the new, unknown samples fall. For example, if we have information about
the leukemia type of 60 samples, we can find the genes, which differentiate these
leukemia types from each other. After finding the suitable set of genes, the identity
of the unknown samples can be predicted. GeneSpring uses the K-means algorithm
described by Golub et al.
Training and test sets (experiments) need to be specified. GeneSpring also
needs to know which parameter contains the information about the groups to be
compared (parameter to predict). After cross-validating the test set, the test set
can be predicted. The result of the analysis is a prediction of the test set sample
identities and a set of genes differentiating the selected groups.

Figure 10.9: The predict parameter value -classification tool in GeneSpring.

10.12 Suggested reading

1. Golub, T. R., Slonim, D. K., Tamayo, P., Huard, C., Gaasenbeek, M., Mesirov,
J. P., Coller, H., Loh, M. L., Downing, J. R., Caligiuri, M. A., Bloomfield, C.
D., Lander, E. S. (1999) Molecular classification of cancer: class discovery
and class prediction by gene expression monitoring, Science 286, 531-7.
2. Brazma, A., Vilo, J. (2000) Gene expression data analysis, FEBS Lett. 480,
17-24.

This chapter was written by Mauno Vihinen, and Jarno Tuimala.

 Part III

Data mining
11 Gene regulatory networks 121

11 Gene regulatory
networks
11.1 What are gene regulatory networks?

By combining gene expression analysis, perturbations or treatments, and mutations

of genes we can study processes like signal transduction and metabolism yielding
information on molecular effects or functions of specific genes. However, gene
expression data permits us to go beyond the traditional line of research and to study
finer structures of molecular pathways exposing causal regulation relations between
genes. The finer structures do not only describe the nature of interactions between
genes, inhibition or activation, but also exemplify direct and indirect effects of
genes. For example, is gene A regulating gene B or vice versa, is the regulation
direct or indirect where there is a mediating gene C (or maybe many) so that A
regulates C and then C regulates B. Inspecting the finer structures, which are
called regulatory networks, gives us a more intricate view of molecular interactions
offering further possibilities for medical interventions.
Inferring regulatory networks from expression data, a process which is called
reverse-engineering, is not a computationally simple problem because an enormous
amount of time is needed even when a trivial approach is applied. Therefore, we
sketch principles and methods with an example approach applicable for the infer-
ence process in next sections.

11.2 Fundamentals

It is customary to visualize regulatory dependencies between genes by a graph

G(V , E) with a vertex set V consisting of the genes examined and an edge set
E, see Figure 11.1. A directed edge E (i, j ) between vertices Vi and V j represents
regulation between genes attached to vertices V i and V j , and the direction of the
edge, denoted by V i → V j , reflects the order of regulation where gene V i regulates
gene V j (up- or downregulation). Node V i is called a parent of V j and node V j is
called a child of Vi . To present logical structures of group regulations, which ap-
pear extensively in nature, hyper edges should be used, where an edge is adjacent to
several genes, but these are used quite rarely because of the convenience of simple
drawings. A graph like G is called gene regulatory network.
By inspecting a putative regulatory graph, like G, build from earlier experi-
ments, we can pose several questions. First of all, is the structure right, or if there
122 DNA microarray data analysis

V2

V1

V3 V4

Figure 11.1: Graph G consisting of four nodes V1 , V2 , V3 , V4 , and tree edges E (1,2) , E (1,3) ,
and E (3,4) .

are several possible structures with respect to the experiments done so far, which
one is right? So a graph represents hypotheses based on current knowledge. By
locating central nodes or nodes of interest we can conduct additional experiments
to confirm the dependencies around nodes or get an idea which of the graphs is
the correct one. Choosing the best node to inspect more carefully is not trivial,
at least not, when we try to make distinctions between putative graphs and we try
to minimize experiment costs at the same time. To this category belongs also the
question: is the order of regulation correct? Manual perturbation, like the deletion
of a gene, might give a clear clue on the order of regulation. Finally, if we stick
to the obtained structure there is still questions on the strength of each regulation,
whether they are they right.
But before all the above questions, we need to know how to construct such a
graph directly from data, because, in theory, any kind of an edge is possible between
any set of vertices, and so the problem is to infer the true connections based on the
data produced by the experiments. This approach is called reverse-engineering
of gene regulatory networks. It is interesting to note that similar to biochemical
reactions, sequences, and three dimensional structures having their own motifs, it
is clear that regulatory networks have to have their own regulatory motifs or circuit
motifs based on the lower level motifs.
To infer a regulatory network, several distinct expression samples of the genes
of discourse are needed. In a time series analysis, the expression levels are recorded
along fixed time points, while in a perturbation analysis, the expression levels are
recorded separately for each manual perturbation (over/under, deletion) of specific
genes. A fundamental difference between perturbation data and time series data is
that perturbation allows a firm inferring order of regulation while time series data
can only reveal the probable regulation direction. The problem can be overcome by
combining a few manual perturbation experiments with time series data. In both
cases, the genes are monitored at fixed points inducing an expression matrix

{X i [ j ] | 1 ≤ i ≤ n, 1 ≤ j ≤ m} ,

where X i [ j ] denotes the expression level X i of gene or node V i at point j , see

Table 11.1. Based on the data, we can try to infer the most prominent regulatory
subnetworks, or, if we know the structure of some subnetwork beforehand, we can
try to model mathematically the regulations between nodes in the subnetwork.
Because the number of possible networks increases super-exponentially on the
number of genes, prominent regulatory networks are usually searched for by using
11 Gene regulatory networks 123

Table 11.1: Top, part of the table for time series expression data of genes SWI5, CLN2,
CLN3, and CLB1 [1]. The mean of whole data was 0. Bottom, the expression levels are
discretized so that values higher than or equal to the mean get a value of 1 and values lower
than the mean get a value of 0.

gene t1 t2 t3 t4 t5 t6 t7 t8 t9
1 (SWI5) -0.41 -0.97 -1.46 0.16 0.74 0.72 1 0.77 0.3
2 (CLN3) 0.49 0.62 0.05 -0.13 0.02 0.04 -0.14 0.24 0.91
3 (CLB1) 0.6 -0.53 -1.37 1.03 1.13 1.27 1.04 1 0.07
4 (CLN2) -1.26 1.6 1.54 0.31 -0.14 -0.88 -1.7 -1.88 -1.7

1 (SWI5) 0 0 0 1 1 1 1 1 1
2 (CLN3) 1 1 1 0 1 1 0 1 1
3 (CLB1) 1 0 0 1 1 1 1 1 1
4 (CLN2) 0 1 1 1 0 0 0 0 0

Table 11.2: Conditional probabilities based on the structure of G and the bottom values
of Table 11.1.
Pr(X 1 = 1) X 1 Pr(X 2 = 1) X 1 Pr(X 3 = 1) X 3 Pr(X 4 = 1)
6/9 1 4/6 1 1 1 1/7
0 1 0 1/3 0 1

supervised approaches, where some intrinsic partial knowledge about regulations

between genes are formulated by implicit or explicit rules. For the mathematical
modeling of regulation inside a network, there are many approaches like Bayesian
network, Boolean network and its generalization, ordinary and partial differential
equations, qualitative differential equations, stochastic master equations, Petri nets,
transform grammars, process algebra, and rule-based formalisms, to mention some.
We will use Bayesian network as a sample tool for several reasons: its frame-
work offers a clear separation of structure and parameter optimization, and adding
predefined rules and information is easy. Moreover, it is widely used for microarray
data.

11.3 Bayesian network

Bayesian network modelling consists of two parts: the qualitative part, where direct
(causal) influences between nodes V are depicted as directed edges E in graph
G, and the quantitative part, where local conditional probability distributions of
expression levels are attached to nodes V of G. Thus, expression levels X i of
nodes Vi are considered as random variables and the edges represent conditional
dependencies between distributions of the random variables.
Let us consider a binary case where a gene can be “off”, denoted by 0, or
“on”, denoted by 1, but not both (see the bottom values of Table 11.1). Having a
fixed structure like graph G in Figure 11.1, the conditional probabilities are easy to
calculate, see Table 11.2.
The benefit of this modeling is that we need to store only O(2 k n) parameters
124 DNA microarray data analysis

when each node has at most k parents, i.e., one “success” probability for each row
in each table just calculated.
This is clear advantage over O(2 n ) parameters when the expression value of a
node depends on the expression values of all other nodes. Thus, parameters pertain
only to local interaction.
Underlying the above discussion there are some basic probability rules that are
examined next. The joint probability for random variables X and Y (not necessarily
independent) is calculated using the conditional probability:
Pr(X ∩ Y ) = Pr(X, Y ) = Pr(Y | X) Pr(X) = Pr(X | Y ) Pr(Y ) ,
where notation Pr(X | Y ) denotes the probability of observation X after we have
seen evidence Y , see Figure 11.2. Reorganizing the equation induces Bayes’ rule
Pr(Y | X) Pr(X)
Pr(X | Y ) =
Pr(Y )
while generalization of it forms the chain rule
Pr(K , X, Y , Z ) = Pr(Z | K , X, Y ) Pr(K , X, Y )
= Pr(Z | K , X, Y ) Pr(Y | K , X) Pr(K , X)
= Pr(Z | K , X, Y ) Pr(Y | K , X) Pr(X | K ) Pr(K ) .
If variables X and Y are independent, then
Pr(X ∩ Y ) = Pr(X) Pr(Y ) ,
and if variables X and Y are independent for given a value of Z , then
Pr(X | Z , Y ) = Pr(X | Z ) .
Based on our dependency graph G, probability Pr(X 1 , X 2 , X 3 , X 4 ) can be simplified
as follows
Pr(X 1 , X 2 , X 3 , X 4 ) = Pr(X 1 ) Pr(X 2 | X 1 ) Pr(X 3 | X 1 ) Pr(X 4 | X 3 ) .
For example, Pr(X 4 | X 1 , X 2 , X 3 ) = Pr(X 4 | X 3 ) because for given the value of X 3 ,
X 4 is independent of the values of X 1 and X 2 . So, it is enough to consider only
direct local dependencies from parents when evaluating the probabilities (for X 4 the
only parent is X 3 ). We see that in Bayesian networks the probability calculations
follow in a natural way the structure of a graph.

11.4 Calculating Bayesian network parameters

Having graph G and the expression matrix D = {X i [ j ]}, our aim is to obtain dis-
tribution dependency parameters θ = {θ 1 , θ2 , . . .} that are fitted best to the structure
of G and data D. This is done by using the maximal likelihood principle where we
maximize the likelihood function L(θ : D):

m
L(θ : D) = Pr(D : θ ) = Pr(X 1 [ j ], X 2[ j ], . . . X n [ j ] : θ ) .
j =1
11 Gene regulatory networks 125

X X ∩Y Y

Figure 11.2: A Venn diagram of the probability space , where the areas denote proba-
bilities, for example, Pr(X), is the left circle area divided by the total area of rectangle .
It is easy to see that Pr(X ∩ Y ) = Pr(Y | X) Pr(X).

By using the structure of G (see again Figure 11.1) we can decompose our opti-
mization problem to independent subproblems:

m
m
L(θ : D) = Pr(X 1 [ j ] : θ1 ) · Pr(X 2 [ j ] | X 1 [ j ] : θ1 )
j =1 j =1

m
m
· Pr(X 3 [ j ] | X 1 [ j ] : θ3 ) · Pr(X 4 [ j ] | X 3 [ j ] : θ4 ) ,
j =1 j =1

or, in general,
Li (θi : D) .
i

The network structure decomposes the parameter set θ to independent parameters

θ1 , θ2 , θ3 and θ4 , which are computationally fast to estimate.
For simplicity, we consider again the binary case, see bottom of Table 11.1.
Since V1 has no parents, i.e., it is independent of other nodes, its likelihood function
L1 (θ1 : D) corresponds to the binomial probability where the probability of getting
value 1 is θ1 while 1 − θ1 is the probability of getting value 0. Thus,

m
L1 (θ1 : D) = Pr(D : θ1 ) = Pr(X i [ j ] : θ1 ) = (θ1 ) N(X 1 =1) (1 − θ1)m−N(X 1 =1) ,
j =1

where N(·) denotes the number of each occurrence in data with respect to the re-
quirements inside parentheses. The estimator θ̂1 for θ1 maximizing the product is
simply the already calculated probability for Pr(X 1 = 1), i.e.,

N(X 1 = 1)
θ̂1 = = 6/9 .
N(X 1 = 1) + N(X 1 = 0)
Similarly, the conditional estimators for parameters θ (2|X 1 ) , θ(3|X 1 ) and θ(4|X 3 ) are
the probabilities already calculated:

θ̂(2|X 1 =1) = 4/6 , θ̂(2|X 1 =0) = 1 ,

θ̂(3|X 1 =1) = 1 , θ̂(3|X 1 =0) = 1/3 ,
θ̂(4|X 3 =1) = 1/7 , and θ̂(4|X 3 =0) = 1 .
126 DNA microarray data analysis

Together these estimators form θ̂ = {θ̂1 , θ̂(2|X 1 ) , θ̂(3|X 1 ) , θ̂(4|X 3 ) }, which is the best
explanation for the observed data D restricted to the structure of G, and we can
compute

L(θ : D) = (θ̂1)6 · (1 − θ̂1)3 · (θ̂(2|X 1=1) )4 · (1 − θ̂(2|X 1=1) )2

· (θ̂(3|X 1=0) )1 · (1 − θ̂(3|X 1=0) )2 · (θ̂(4|X 3=1) )1 · (1 − θ̂(4|X 3=1) )6
= (6/9)6 · (3/9)3 · (4/6)4 · (2/6)2 · (1/3)1 · (2/3)2 · (1/7)1 · (6/7)6

as the maximal likelihood value for graph G. Note that we have dropped out terms
1x = 1 and 00 = 1.
Usage of more states for genes like “low”, “medium”, and “high” follows the
multinomial theory conveniently generalizing the binomial theory. Using the pa-
rameters, we can formulate hypothetical behaviors of genes in different situations
fixing some expression levels and observing others; moreover, updating the param-
eters is easy as new data arrives.

11.5 Searching Bayesian network structure

Given a graph G, we know now how to calculate the parameter set θ G maximizing
the likelihood score L(θ G : D). To restrict the search space of possible networks, it
is practical to have constrains on which kind of networks are allowed, that is, which
dependencies are possible between genes. This helps a lot because of the super-
exponential number of possible networks; for example,
 the number of possible
graphs consisting of four nodes is 64 since there are 42 possible (undirected) edges,
4
each of which can be taken to form a graph making 2 (2) different graph in total. For
evaluation of the resulting candidate networks, we use again the likelihood score
calculation of a network structure using maximum likelihood estimates of θ G when
the structure is searched

L(G, θ G : D) = Pr(X 1 [m], X 2 [m], . . ., X n [m] : G, θ G )
m

Pr(X n [m] | Parents(X n )[m] : G, θnG ) .
m n

Here, Parents(X n ) denotes the expression values of all parents of node V n and θnG
denotes the parameter of node V n in graph G. Instead of direct calculation it is more
convenient to use a logarithm of L(G, θ G : D) for comparing different structures.
After simplifying we have a surprisingly simple formula


n
log(L(G, θ G : D)) = m (I(Vi , Parents(Vi )) − H(Vi )) ,
i=1

where H(Vi ) is the entropy of expression values X i of node Vi

 1
H(Vi ) = Pr(X 1 = x) lg
x=0,1
Pr(X 1 = x)
11 Gene regulatory networks 127

V2

V3 V1

V4

Figure 11.3: A different structure explaining conditional dependencies between the ex-
pression values of nodes V1 , V2 , V3 , V4 .

which can be ignored because it is the same for all network structures. The function
I(Vi , Parents(Vi )) is the mutual expression information between node V i and its
parent nodes Parents(Vi ) = {P1 , P2 , . . .} defined by

 Pr(X i = x i Pi = x i )
Pr(X i = x i Pi = x i ) lg  .
Pr(X i = x) i Pr(Pi = x i )
x=0,1
(∀i):x i =0,1

Function I(Vi , Parents(Vi )) ≥ 0 measures how much information the expression val-
ues of nodes Parents(Vi ) provide about V i . If Vi is independent of parent nodes,
then I(Vi , Parents(Vi )) has the value of zero. On the other hand, when V i is totally
predictable for given values of Parents(V i ), then I(Vi , Parents(Vi )) reduces into the
entropy function H(V i ). It should be noted that in general I(X, Y ) = I(Y , X) so the
direction of edges matters.
Is the structure presented in Figure 11.1 optimal with respect to the data? What
about the structure in Figure 11.3. Because our data set is small we do not resort to
the above calculations but we calculate the parameter set θ̂ :

θ̂2 = 7/9 , θ̂4 = 3/9 ,

θ̂(3|X 2 =0,X 4 =0) = 1, θ̂(3|X 2 =0,X 4 =1) = 1,
θ̂(3|X 2 =1,X 4 =0) = 1, θ̂(3|X 2 =1,X 4 =1) = 0,
θ̂(1|X 3 =0) = 1/3 , and θ̂(1|X 3 =1) = 1.

The comparison of these figures with the previous ones shows that the new graph
is more probable with respect to data.

11.6 Conclusion
Our example data was complete without any missing observations of expression
values, which happens quite rarely in reality. Two of the most familiar conventions
to deal with the problem of missing expression values is to omit the data rows miss-
ing some values or substituting the most common values for the missing values. Of
course, what is the best policy, is an everlasting topic to debate on.
Even a more difficult problem is to decide how the discretization from expres-
sion values to logical “on/off” values should be done. Can we really use a global
step value like in our example or should we use use gene-specific values? On the
other hand, can we use a single limit because then a small variation in measure-
ments can toss a gene between the “on” and “off” state without any real reason?
128 DNA microarray data analysis

Table 11.3: Bayesian network softwares (some are commercial).

Bayesia www.bayesia.com
JavaBayes www-2.cs.cmu.edu/ javabayes
PowerConstructor family www.cs.ualberta.ca/ jcheng/bnsoft.htm
Bayesian Knowledge Discoverer kmi.open.ac.uk/projects/bkd
GeNIe www.sis.pitt.edu/ genie
Prevision www.prevision.com
Hugin www.hugin.com
Norsys www.norsys.com
Knowledge Industries Company www.kic.com
Microsoft MSBN system www.research.microsoft.com/msbn

One cure for the last problem might be to use fuzzy logic, where a gene has an
increasing grade of being “on” and after a fixed point it is completely or fully “on”;
similarly for “off” but reversed.
Today, there are some Internet-based servers set up for inspection of reverse-
engineering results based on researchers’ own data and few software packages, see
Table 11.3. Our better graph was calculated by using a server hosted by Techni-
cal High School (B-course). In future, some of these tools will be integrated to
common gene expression software tools like GeneSpring and Kensington.

11.7 Suggested reading

1. Spellman, P. T., Sherlock, G., Zhang, M. Q., Iyer, V. R., Anders, K., Eisen,
M. B., Brown, P. O., Botstein, D., Futcher, B. (1998) Comprehensive identi-
fication of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae
by microarray hybridization. Mol. Biol. Cell 9, 3273-97.

2. Pe’er, D., Regev, A., Elidan, G., Friedman N. (2001) Inferring subnetworks
from perturbed expression profiles, Bioinformatics 17, 215S-224.

3. D’haeseleer P., Shoudan Liang S., Somogyi R. (2000) Genetic network infer-
ence: from co-expression clustering to reverse engineering, Bioinformatics
16, 707-726.

This chapter was written by Tomi Pasanen, and Mauno Vihinen.

12 Data mining for promoter sequences 129

12 Data mining for

promoter sequences
12.1 Introduction

In all gene expression studies there is an underlying element of gene regulation. In

order to be expressed, the gene has to be transcribed to RNA in a regulated pro-
cess. Therefore, many expression experiments can be used to probe the regulation
patterns. This is most obviously the case when you observe gene expression in dif-
ferent conditions or gene expression over time in some process, but comparisons of
different cells or tissues or diseased v. healthy tissues may benefit from analysis of
regulation, too. Whatever your study, there will be changes in the expression levels
of some genes, either upregulation or downregulation. How the expression levels
can be correlated to mechanisms of regulation is the topic of this chapter.
We start with an introduction about gene regulation and promoters and then
turn to the practical issues of bioinformatics in promoter data mining . We deal
with finding promoter sequences, searching known regulatory sequence patterns
and searching potential regulatory patterns that are not known a priori.

12.2 Introduction

Changes in the expression levels of genes can be due to a number of factors. First,
there are such large-scale factors as chromatin accessibility: some regions of DNA
are packed too tightly to allow strand separation or even binding of various factors.
One well-known mechanism of chromatin packing is found in methylation of cy-
tidines in CpG islands to effect gene silencing via heterochromatin formation, and
another one is related to histone acetylation/methylation. Second, there are specific
transcription factors that may inhibit or activate transcription. The balance between
activating and inhibiting transcription factors can determine the expression level in
loosely packed euchromatin. Third, there are some genes that seem to be on “by
default” without much apparent regulation, the so-called maintenance or household
genes which are needed in all cell types. “Household gene”, however, does not im-
ply constant expression; in any gene the expression level will fluctuate over time
due to the general metabolic activity of the cell, especially the activity and amount
of parts in the transcription machinery itself. If maintenance genes are used for
normalization of array data it is necessary to verify the constant expression, e.g., by
RT-PCR in the system and conditions of the experiment. (See also the chapter on
130 DNA microarray data analysis

normalization.)
Data mining for gene regulation is currently feasible only in the second case
above, i.e., in search and study of transcription factor binding sites.
Most known transcription factor binding sites are located close to the tran-
scription start site (TSS), particularly in the 500 bp directly upstream (5’) of TSS.
This upstream region is often referred to as the promoter region (but sometimes
the binding sites themselves are called promoters). Other regions that activate tran-
scription (the enhancers) may occur almost anywhere downstream (3’) or upstream
from the gene, even 20 kB away from the promoter region, or in the introns. The
methods for the search of regulatory elements that we describe could be used for
potential enhancer regions as well as for the promoter regions, but commonly only
promoters are analyzed. This because a larger search space weakens the signal-to-
noise ratio. Therefore, we limit our discussion to data mining in promoter regions
only.
Coexpressed genes are often grouped by different clustering methods. It is
reasonable to present the hypothesis that some of the similar changes of expression
are due to the effect of the same or similar transcription factors – therefore it makes
sense to search for shared sequences that could be transcription factor binding sites.
Of course, depending on your clustering (cluster sizes and number), you may have
several mechanisms working within one cluster, or you may have the same mech-
anism affecting members of several clusters, but in any case, at least some clusters
may show enrichment of similar transcription binding sites.
In brief, it is presumed that coexpression implies coregulation, and looking for
shared patterns in promoter regions allows you to formulate hypotheses of regula-
tion mechanisms for focused experimental testing. However, before verification,
your results will be only sets of potential promoter sites of potentially coregulated
genes.

12.3 Finding promoter region sequences

Promoter analyses need the promoter region sequences, but getting them may not
be trivial! In this section, we introduce some tools and data sources that we have
found useful for retrieving promoter regions for human genes .
First of all, it is instrumental to know precisely which genes are contained in
the microarray. The data from the manufacturer is not always systematically pre-
sented or complete. There may be a mix of references to GenBank accession num-
bers, UniGene clusters, gene names, protein accession numbers, RefSeq mRNA or
protein accession numbers, and LocusLink ID codes, and if you are lucky, even
actual sequences that are contained on the array.
Even if we skip the concern over the correctness of the data provided by the
manufacturer, there are some issues that you should be aware of before you can
trust your gene assignments:
• gene names have been subject to many changes, so your data may not contain
the current official names,
• the UniGene clusters of ESTs are dynamic entities that undergo fusions and
12 Data mining for promoter sequences 131

fissions in every UniGene build as more data becomes available, so your

ESTs may be assigned to other genes, or may lack a UniGene association in
the current build,

• the data regarding your filter or chip may have been updated after it left the
factory, so the shipped gene lists may not be as complete and up-to-date as
you can find at the manufacturer’s web site. You should definitely work with
the most recent data available as long as it refers to the same manufacturing
batch that you actually used,

• you may find several RefSeq mRNA and protein codes for a single gene
locus, corresponding to alternatively spliced forms,

• the provided EST code is not always the actual sequence on the array, be-
cause the manufacturer may have used a longer and more correct version of
the 5’-end of the corresponding mRNA.

The ultimate source for the promoter regions is in the genome data. The anno-
tations of gene locations are still incomplete, and there are no ready-made tools for
pulling out the regions preceding each gene. Currently (early 2003) we have found
that human promoters are easiest to retrieve in large scale from three data sources.
They are multi-species resources, so when more genomes become available, the
same approaches work.

• University of California in Santa Cruz (UCSC) provides ready-made collec-

tions of upstream sequences of as many genes as they can localize based
on RefSeq mRNA sequences – different sets contain different lengths of 5’-
sequences preceding the TSS, called upstream1000, upsteam2000 etc. The
human data from the most recent freeze of November 2002 is at http:/
/genome.ucsc.edu/goldenPath/14nov2002/bigZips/, but you will al-
ways find most recent data at http://genome.ucsc.edu/downloads.html
under Full data set.

• LocusLink gives pointers to NCBI RefSeq Contigs, allowing retrieval of pro-

moters by a little programming of your own. LocusLink is essential for using
the UCSC data, too, if you do not have RefSeq mRNA codes.

• EnsMART (http://www.ensembl.org/EnsMart/) allows direct retrieval of up-

stream sequences of any length if you can map your genes to Ensembl gene
codes or RefSeq codes (more details follow).

For a smaller number of genes the most reliable data (regarding the place-
ment of TSS) is available in the Eukaryotic Promoter Database (EPD), which con-
tains promoter regions of 500 bp, only from genes with an experimentally verified
TSS. Organism-specific promoter databases may exist to fit your needs, consult the
genome site of your favourite organism or Michael Zhang’s lab at http://rulai.
cshl.org/software/index1.htm.
If these sources fail, or if you feel you cannot trust all of your findings, you can
always resort to using your own similarity searches to place the gene in the genome
132 DNA microarray data analysis

assembly, and then pick the region preceding the gene, but this is far from straight-
forward. It may be advisable to drop data for which you cannot get unambiguous
gene mapping. Currently full genome information is in most cases not available for
all genes contained in the microarrays.
If you aim at finding your promoter sequences from the UCSC upstream data
sets, you need the corresponding RefSeq mRNA accession codes, because these
codes are used as identifiers for the promoter sequences. You can also retrieve
Ensembl genes with RefSeq codes. Figure 12.1 shows you the paths and tools for
arriving at RefSeq codes from other data.

Figure 12.1: Different data items that you may receive from your microarray manufacturer
and how you can find missing data items and the actual sequences.

The tools featured in Figure 12.1 include:

• similarity searches, such as Blast or BLAT (http://genome.ucsc.edu/
cgi-bin/hgBlat?command=start). For large-scale searches you probably
want to install them locally and use text processing tools such as Perl for ex-
tracting relevant information from the result files. BLAT has the advantage
of giving directly the coordinates in the genome, and on-line BLAT accepts
modest batches of sequence (a few hundred) if you just want to patch missing
data,
• EST annotation machine – a nice service at http://bio.ifom?firc.it/
EST_MACHINE/index.html for obtaining UniGene, RefSeq, and/or LocusLink
codes and functional data starting from accession codes of ESTs or other se-
quences. Batch runs of 1000 codes at once are possible,
• Batch Entrez – mass retrieval of sequences from NCBI (http://www.ncbi.
nlm.nih.gov/entrez/batchentrez.cgi?db=Nucleotide),
12 Data mining for promoter sequences 133

• utilization of LocusLink data (http://www.ncbi.nlm.nih.gov/LocusLink/

– use Download to get the data) – very useful for finding alternative gene
names, or RefSeqs corresponding to UniGene clusters etc. In large scale you
need programming or setting up a local database for lookups of the informa-
tion,

• the HUGO nomenclature service (http://www.gene.ucl.ac.uk/

nomenclature/) can be used for verifying gene names, too, even though
LocusLink files make it easier in large scale, and they aim to conform to
official HUGO names,

• CSC also offers tools for fast BLAST searches (gepardi.csc.fi) and mass re-
trieval of sequencies in FastA format (Seqret in the EMBOSS package).

If you want to be really certain that you know which genes you are dealing
with, you should follow several paths in the diagram to see if they all lead you to
same Locus IDs. Remember that there may be several RefSeq mRNAs per gene,
whereas Locus IDs are unique.
The data files from GeneLynx (http://www.genelynx.org/cgi-bin/
a?page=info) are likely to allow many shortcuts to obtain RefSeq mRNA codes,
but we have not tried this yet. Likewise, querying of various UCSC genome data
tables seems a promising option.
If you work with the Affymetrix Arabidopsis chips, you may want to look at
VIZARD (http://www.anm.f2s.com/research/vizard/), which includes an-
notation and upstream sequence databases for the majority of genes represented on
the Affymetrix Arabidopsis GeneChip
R array. Whitehead Institute at MIT pro-
vides upstream sequence data for Neurospora crassa and several other organisms,
see e.g. http://www?genome.wi.mit.edu/cgi?bin/annotation/neurospora/

download_license.cgi

12.4 Using EnsMart to retrieve promoter regions

Yet another option for retrieving upstream sequences is found at the EnsMart ser-
vice (http://www.ensembl.org/EnsMart/) of the Ensembl project. This is lim-
ited only by the number of Ensembl genes that are annotated (currently almost
25,000), and it is very easy to use, as illustrated in Figure 12.2. The performance of
EnsMart v. UCSC upstream data searches is compared later in this chapter. They
have several species to select from:
134 DNA microarray data analysis

Figure 12.2: EnsMart start screen.

Next, enter your list of gene identifiers in the Filter step (Figure 12.3):

Figure 12.3: EnsMart filtering. In addition to RefSeq, there are other options, too, but you
should be aware of that some mappings are more complete than others. Next to internal
Ensembl references, RefSeq is your best choice. A long list of other filtering options is
omitted from the figure.

Then, in the next phase, you should choose output as Sequences, and select
your options as below to retrieve your upstream sequences in Fasta format (Fig-
12 Data mining for promoter sequences 135

ure 12.4). Additional options for data compression, saving locally etc. are not
shown.
For a few microarrays (currently only two human Affymetrix chips), EnsMart
provides direct mappings, so you can skip all the previous steps for finding a con-
sistent set of gene identifiers. Instead, you can choose the genes in these chips
directly in the Filter step (Figure 12.5).
We can expect such mappings with microarray contents to become more com-
mon in the genome sites, both at Ensembl and at UCSC. Therefore, some day
retrieving the upstream sequences will be less of a problem.

Figure 12.4: Sequence output options in EnsMart.

Figure 12.5: Selecting genes that are represented on AFFY-HG-U133. Compare with Fig.
12.3.

12.5 Comparison of EnsMart and UCSC searches

In order to evaluate how the most extensive two search systems perform, we took
a set of 1187 RefSeq codes (from our own microarray results) and retrieved the
upstream sequences from both services. USCS upstream 1000 data set was from
136 DNA microarray data analysis

Table 12.1: Search of upstream sequences from the precomputed upstream data set of
UCSC (http://genome.ucsc.edu/) and an interactive search from Ensembl data via
EnsMart (http://www.ensembl.org/EnsMart/)

Entries retrieved Unique mRNA codes

UCSC 1145 1087
EnsMart 1171 1131
Found in both 1054
UCSC not Ensmart 33
EnsMart not UCSC 77
Not found in either 23

the November 2002 freeze of the human genome, and Ensembl data was as of 31st
March, 2003. Table 12.1 summarizes our findings.
There seem to be two lessons to be learned. First, there is some duplication,
so finding the real gene and promoter sequence is not unambiguous even if you
know a code in a curated collection and use the most authoritative data sources.
Second, currently it is useful to use both services, even though both data sources
are expected to become more complete in the near future as the human genome
data becomes more fully assembled and annotated. The situation may be different
for your favourite organism.
We did not make any comparison whether the sequences or genome locations
matched between the two data sets, but we would guess there are some differences.
Two final warnings regarding upstream data:

• in some cases the genome assembly may be broken close to the start of your
gene, so that you do not get a full 1000 (or whatever) nucleotides of sequence.
In Ensembl this shows as a long string of Ns in the sequence (this may be
another reason why Ensembl gives more sequences)

• the start of a RefSeq mRNA may not always correspond to a true transcrip-
tion start site. There are mRNAs or ESTs which give evidence of longer
transcript variants than some of the current RefSeqs. UCSC genome browser
only includes entries with a certain TSS in their upstream1000 data set Search-
ing known patterns

Once you have found the genes and their 5’ regions you can proceed to the
analysis of patterns in the regions.
If you want to run a check of known transcription factor binding sites , Tran-
scription Regulatory Regions Database (TRRD, http://wwwmgs.bionet.nsc.ru/
mgs/dbases/trrd4/trrdintro.html) is a site with lots of information, but only
for on-line searching. The complete database is not available. The Transcription
Factor Database (TFD) and some tools that utilize the data are available at http:/
/www.ifti.org/.
Transfac database ver. 5.0 (from 2001) is said to be publicly available for
academic use (http://transfac.gbf.de/TRANSFAC/), but currently most of the
12 Data mining for promoter sequences 137

data and services seem to be closed. The EMBOSS program tfscan (http://www.
csc.fi/molbio/progs/emboss/Apps/tfscan.html) can be used to find Trans-
fac sequence patterns in your sequences once Transfac is set up in the EMBOSS
environment.
The most recent versions of Transfac database and its search tools have moved
into a commercial environment at http://www.gene-regulation.de/. They of-
fer PC software for promoter analysis, Transplorer (Figure 12.6). Their “Transfac
Professional” package contains search tools, too.

Figure 12.6: Screenshot from Transplorer.

Whatever method you choose to search for TF binding sites, you have to bear
in mind that they are short patterns that will be found very frequently by chance
alone, so you need to confirm the statistical significance of your findings. A search
strategy which aims at finding clusters of several sites might work better in this
respect.

12.6 Pattern search without prior knowledge of what you are

searching for
Several programs and different approaches are available for automated pattern dis-
covery in your promoter region sequences. We will mention here a few examples
of different programs and strategies, all of which are capable of finding patterns in
sequences without their previous alignment. A more extensive review can be found
in Vilo and Kivinen, 2001.
Expression Profiler is a software suite for many tasks in microarray data analy-
sis (http://ep.ebi.ac.uk/EP/). It contains the program SPEXS, Sequence Pat-
tern EXhaustive Search, for pattern discovery, which finds regular-expression-like
discrete patterns. As an example, this program was used by Vilo et al. (2000) to
find regulatory patterns from publicly available yeast expression data. To overcome
the fuzziness created by the incertainties in clustering, they generated a large num-
ber of different clusters (52,000, with lots of overlapping) and analyzed enriched
patterns from all of them. After filtering and grouping the patterns, they had 62
groups, 48 of which contained patterns matching some sites in the Saccharomyces
cerevisiae Promoter Database.
MEME (http://meme.sdsc.edu/meme/website/intro.html) is based on
138 DNA microarray data analysis

position-dependent letter-probability matrices, which describe the probability of

each possible letter at each position in the pattern. The WWW interface analyzes
your set of sequences that you believe to contain common patterns, and gives you
the strongest patterns that MEME finds. In addition, you get automatically results
from a companion program MAST, which locates the occurences of the patterns in
your set of sequences.
AlignACE (http://arep.med.harvard.edu/mrnadata/mrnasoft.html)
provides a Gibbs sampling algorithm for finding patterns in DNA sequences. The
program is optimized for finding multiple motifs and it automatically considers
both strands in the sequence. The authors used the program to identify successfully
several known transcription factor binding sites and to propose some previously un-
known regulatory mechanisms based on the earliest public yeast gene expression
data sets [2].
All of the above methods depend on a clustering that you have made before
pattern discovery, and it is presumed that the clustering is correct. Kimono (http:
//www.fruitfly.org/∼ihh/kimono/) takes a unified approach that optimizes
clustering and pattern discovery together. The approaches described above first find
genes with similar expression patterns, and then see if they have similar promoters.
Kimono finds clusters of genes that have similar expression patterns and similar
promoters as described by Holmes and Bruno (2000).

12.7 Summary
In summary, we identified a similar expression profile of certain genes using DNA
microarrays and clustering tools. We hypothesized that the similar expression was
due to common regulatory elements situated in the promoter regions of the genes.
We retrieved the promoter sequencies from the databanks, analyzed the promoter
regions, and successfully identified a common element on their promoter region.

12.8 GeneSpring and promoter analysis

GeneSpring includes a promoter analysis tool, which can be used for finding novel
common regulatory sequencies in a gene list, or to search for a known sequence.
The tool can be invoked from Tools->Find potential regulatory sequencies. In
order to search for potential regulatory sequencies, you need to have a whole
genomic sequence of the organism under study. In principle, if only a partial
genome of the organism is known, it is not possible to search for regulatory el-
ements (GeneSpring forbids the use of the tool), because the statistical support
and frequency values of the elements would be erraneous. However, there is a
trick, which allows the analysis of partial genomes. For more information, see the
tech note at http://www.silicongenetics.com/cgi/TNgen.cgi/GeneSpring/
GSnotes/Notes/how_contig.
The tool opens a new window (Figure 12.7). First, you need to select a genelist
you want to study, but do not use the “all genes” or “all genomic elements” list, be-
cause then you would compare the whole genome against itself, which is not a
viable analysis. From the pull-down menu, select whether you want to search for
12 Data mining for promoter sequences 139

new sequencies or for a specific sequence. You can also select the lenght of the se-
quence to be considered a promoter region, how long a regulatory element is being
searched, and how many unknown bases are allowed. The longer the sequence, and
the larger the number of unknown bases, the longer the analysis time. You have
control over the probability statistics: The p-value cut-off for a significant pattern
can be modified. Whether the sequence is relative to the sequence upstream of
other genes or relative to the whole genomic sequence can also be modified. The
first option is far more common.
After the analysis have completed, or you stop the search, the results are re-
ported. They appear on right side of the toolbox. Potential regulatory sequencies,
the number of genes they were detected in, and the detection p-value are reported.
The best findings are reported first.

Figure 12.7: Find potential regulatory elements -tool in GeneSpring.

12.9 Suggested reading

1. Holmes, I., Bruno, W. J. (2000) Finding regulatory elements using joint like-
lihoods for sequence and expression profile data. Proc. Int. Conf. Intell.
Syst. Mol. Biol. 2000 8, 202-10.

2. Roth, F. P., Hughes, J. D., Estep, P. W., Church, G. M. (1998) Finding DNA
regulatory motifs within unaligned noncoding sequences clustered by whole-
genome mRNA quantitation. Nat. Biotechnol. 16,939-45.

3. Vilo, J., Brazma, A., Jonassen, I., Robinson, A., Ukkonen, E. (2000) Mining
for Putative Regulatory Elements in the Yeast Genome Using Gene Expres-
sion Data. Proc. Int. Conf. Intell. Syst. Mol. Biol. 2000 8,384-94.

4. Vilo, J., Kivinen, K. (2001) Regulatory sequence analysis: application to

interpretation of gene expression Eur. Neuropsychopharmacol. 11, 399-411.

This chapter was written by Martti Tolvanen, Mauno Vihinen and Jarno
Tuimala (GeneSpring examples).
140 DNA microarray data analysis

13 Annotations and article

mining
Annotation is a “comment” attached to a gene. The comments can, for example,
describe the gene’s biological function, it’s interactions between other genes, and
the metabolic pathways the gene is acting in. After the interesting gene has been
identified using statistical tools, it’s annotation should also be acquired, in order to
make inferences about the validity of the findings, and to generate new hypotheses.
Stanford Knowledge Systems lab defines ontology as “a formal and declara-
tive representation which includes the vocabulary (or names) for referring to the
terms in that subject area and the logical statements that describe what the terms
are, how they are related to each other, and how they can or cannot be related to
each other. Ontologies therefore provide a vocabulary for representing and com-
municating knowledge about some topic and a set of relationships that hold among
the terms in that vocabulary”.
Gene Ontology Consortium ( http://www.geneontology.org/) aims to pro-
duce “controlled vocabularies for the description of the molecular function, bio-
logical process and cellular component of gene products”. The Consortium has
produced a hierarchical structure of Gene Ontologies (GO), which is independent
of any organism. Genes of different organisms may be annotated using GO de-
scriptions: Three GO-numbers, which correspond to a gene function, process or
localization, are assigned to every gene.

13.1 Retrieving annotations from public databases

Sequences are usually annotated when they are submitted to databanks, e.g., Gen-
bank or EMBL. Sequences in Genbank and EMBL are computationally annotated
and might contain errors. Some sources of annotations, like SWISS-PROT and Ref-
Seq, contain curated annotations for the sequences, and the information is probably
more reliable than in Genbank. Annotations for individual genes can easily be
retrieved from the databanks using the sequence accession numbers.
Also Locuslink and Unigene, which are available at NCBI (http://www.
ncbi.nlm.nih.gov/), contain valuable information about gene functions. Lo-
cuslink presents information on official nomenclature, aliases, sequence accessions,
phenotypes, EC numbers, MIM numbers, UniGene clusters, homology, map lo-
cations, and related web sites. This information can be retrieved using common
gene names or Locuslink accession numbers. In addition, GeneCards ( http:
13 Annotations and article mining 141

//bioinfo.weizmann.ac.il/cards/) contain information about gene functions

and their associations with certain diseases. GeneCards can be queried using gene
name or accession number.
Genome databases, such as Ensembl, also include detailed and curated in-
formation about the gene functions. Ensembl also has very good links to remote
databases. The Ensembl data can be queried using any sequence database accession
number. The Ensembl server can also be used for converting from one numbering
system to another. Uses of the Ensebml database are described in more detail in
Chapter 12.
The databases can be queried one gene at a time using public www-interfaces.
With a little programming, multiple genes can be retrieved from the databases using
the public www-interfases. It is also possible to install the databases locally, and
make searches using the local query tools.

13.2 Retrieving annotations using BLAST

If sequences, or at least Genbank accession numbers for the genes of interest are
available, their annotations can be updated using BLAST. After performing a (stan-
dard) BLAST search, you can pick three to five best scoring hits, and check what
annotations they have. Likely, your query sequence has functions similar to the
best hits you find from the database.

13.3 Article mining

Articles and abstracts contain masses of valuable information about the functions
of the genes, but the data as such is difficult to approach. Of course, individual
searches with gene names can be made from PubMed (http://www.ncbi.nlm.
nih.gov/Pubmed/), but this is not feasible, if the number of genes or published
articles is high. Article mining makes the retrieval of gene functions easier. Using
a suitable query term, e.g., a gene name, a list of functions or common keywords
identified from the articles and abstracts can be produced.
MedMiner http://discover.nci.nih.gov/textmining/main.jsp is an ar-
ticle mining tool for biomedical information. Currently, MedMiner combines the
information of Whitehead Institute’s GeneCards and Pubmed. Searches about gene,
gene-gene or gene-drug functions and interactions can be made. MedMiner does
not necessarily contain information about all human genes, because the genes in the
current version of the program have been selected from a certain DNA microarray.
National Cancer Institute also offers many other text mining tools, like GO-
Miner (http://discover.nci.nih.gov/gominer/), which helps to interpret the
human -omic-data (including DNA microarrays) by classifying the genes into bio-
logically coherent categories based on GO ontologies. Classifications can then be
assessed using the tool.
Note that article mining is not a substitute for getting familiar with the rele-
vant articles, but it enables one to easily infer probable gene functions from the
established literature.
142 DNA microarray data analysis

13.4 Annotation and gene ontologies using GeneSpring

13.4.1 Annotations

GeneSpring contains a built-in annotation tool (Figure 13.1), which can be used if
Genbank accession numbers for the genes on the chip are known. If you do not
have Genbank accession numbers, don’t panic, because RefSeq and SWISS-PROT
(and many other) accession numbers can be converted to Genbank IDs using the
Ensembl database or other servers available on the Internet.
Moreover, when importing the data to GeneSpring, you have to specify which
column in your datafile contains the Genbank accession numbers. This column is
marked in GeneSpring as GenBankID.
The annotation tool GeneSpider can be accessed from the menu Annotation-
>GeneSpider. GeneSpider has four different options: you can annotate your chip
using either Genbank, Unigene or Locuslink database or Silicon Genetics’ mirror.
It is recommended to use the Silicon Genetics’ mirror, because it contains all the
combined information of three other databases. It is also faster to update annota-
tions using the mirror server.
GeneSpider contains several settings. First of all, you need to select the col-
umn containing Genbank IDs. Additionally you need to select the sources for the
annotations. Annotations from all the selected sources can be combined. Alter-
natively, only the highest priority annotations are used. Of course, the priority of
the databases can be mofified. Usually, the existing annotations are overwritten,
but you can also decide not to do so. In addition to annotations, it is possible to re-
trieve sequences, but it can be very time-consuming, especially if you are analyzing
human data.
It is also possible to request GeneSpring genomes for some commercial chips,
e.g., Affymetrix, directly from Silicon Genetics. This often cuts down the hazzle
with annotations and Genbank IDs tremendously.

13.4.2 Ontologies

After retrieving annotations for your genes, you can build a simplified ontology.
The simplified ontology creates a number of new gene lists, which fall into three
categories; molecular function, biological process and cellular component of gene
products, as specified in the GO ontology. The tool is located in Annotations-
>Build simplified ontology. The simplified ontology tool provided by GeneSpring
is, as the name indicates, a simplified version of the actual GO ontology, and it
is not updated as often as the official GO ontology is, so you should consider the
formed genelist groupings suggestive.
Build simplified ontology tool is handy if you are interested in a certain gene
type. For example, if you are interested in transcription factors, you can use the
automatically created gene list (transcription factors) as a basis for further analyses.
You can create a similar gene list using text filtering tools, but it is much more
convenient to let GeneSpring do the work for you.
13 Annotations and article mining 143

Figure 13.1: GeneSpring annotation tool GeneSpider. Using the current settings, data
from GenBank, Unigene and Locuslink will be searched, but only the Unigene annotations
will be saved. Moreover, old annotations will not be overwritten, and no sequences will be
retrieved.

This chapter was written by Jarno Tuimala.

 Part IV

Tools and data management

14 Reporting results 145

14 Reporting results
14.1 Why the results should be reported

Although microarrays work best as a screening tool, sometimes the final results that
you wish to publish lean largely on the microarray data.
There are several good reasons why you should make your microarray data
publicly available. It will be beneficial for you and others. Analyzing several good-
quality, well-documented experiments together will ease the task of finding actual
correlations between the gene expression. Since the microarray experiments are
expensive, adding publicly available information from other microarrays into your
own findings will save money.
You should think of publishing your microarray results already from the very
beginning of the experimental work. Later on, it may be very difficult to “mine your
notebook” for all the details on how you performed your experiments and how you
handled your data.
Microarray data without detailed information about the experimental condi-
tions and analysis steps is useless, and it should always be accompanied with that
supportive information. To be able to compare your data with other researchers’ ex-
periments, all this information has to be reported in a common way, using a widely
accepted form. Such a form is called Minimum Information About a Microarray
Experiment, MIAME.
Furthermore, two major scientific journals, Nature and Cell, as well as EMBO
Journal, have already set a condition that the microarray data in the submitted paper
have to be MIAME-compliant and publicly available, before they accept the paper
for publication (see instructions for authors at http://www.cell.com and http:/
/www.nature.com, and Nature opinion article “Microarray standards at last”, Na-
ture 419, 323 (26 Sep 2002)). In a similar way, The Lancet has adopted MIAME
quidelines (http://www.mged.org/Workgroups/MIAME/miame_checklist.html).

Probably this demand will become a prerequisite also for other journals, in a same
way as publishing your sequence data in public databases, GenBank and EMBL.

14.2 What details should be reported: the MIAME standard

The MIAME standard outlines the minimum information that should be reported
about a microarray experiment to enable its unambiguous interpretation and repro-
duction. The latest version of the MIAME standard can be found from the MGED
146 DNA microarray data analysis

(Microarray Gene Expression Data) group’s web page, see http://www.mged.

org/miame. Simply, “MIAME is what one scientist should tell another scientist
about his/her microarray experiment so that the other scientist could reproduce it
and verify the results”.

Figure 14.1: The MIAME organization. A schematic representation of the six components
of a microarray experiment (A. Brazma et al. 2001)

The MIAME includes a description of the following six sections (Figure 14.1):

• Experimental design: the set of hybridization experiments as a whole, which

may contain several hybridizations

• Array design: each array used and each element (feature, reporter, compos-
itegroup) on the array and protocols used

• Samples: samples used, extract preparation and labeling, protocols used

• Hybridization: procedures and parameters

• Measurements: image quantification and specifications

• Normalization controls: types, values and specifications

• See link http://www.mged.org/Workgroups/MIAME/miame_glossary.html.

Microarray Gene Expression Data (MGED) Society is an international group

of people from different disciplines who has a common task to standardize the
methods for presenting and sharing microarray data. In addition to MIAME build-
up, MGED Society has several working groups developing microarray standards,
such as The Ontology Working Group (http://mged.sourceforge.net/ontologies/index.php).
The ontology is a defined vocabulary to describe and name different terms (events,
data, methods, material) related to microarray research so that the microarray data
14 Reporting results 147

can be annotated based on this ontology. MGED ontology is used as a basis for
building the MAGE object model and markup language. MGED ontology is an
integral part of the MIAME standard, and they are being developed together.

14.3 How the data should be presented: the MAGE standard

MAGE (MicroArray and Gene Expression) is a standard for the representation of
microarray expression data and it is able to capture information specified by MI-
AME. MAGE has been developed by the European Bioinformatics Institute (EBI)
and Rosetta Biosoftware and recently became an adopted specification of the OMG
(Object Management Group) standards group (http://www.omg.com), but it’s still
an ongoing project. MAGE consists of three parts: An object model (MAGE-
OM), a document exchange format (MAGE-ML), which is derived directly from
the object model, and software toolkits (MAGE-STK), which seek to enable users
to create MAGE-ML.

14.3.1 MAGE-OM
MAGE-OM (microarray gene expression object model) is an Object Model for
the representation of microarray expression data that facilitates the exchange of
microarray information between different data systems and organizations.
MAGE-OM is a data-centric model that contains 132 classes grouped into 17
packages containing, in total, 123 attributes and 223 associations between classes.
Entities defined by the MIAME standard are organized as MAGE-OM components
into packages, such as Experiment, BioMaterial, ArrayDesign, BioSequence, Ar-
ray, and BioAssay packages, and the gene-expression data into a BioAssayData
package. The packages are used to organize classes that share a common purpose,
and the attributes and associations define further the classes and their relations.
A database created using MAGE-OM is able to store experimental data from
different types of DNA technologies such as cDNA, oligonucleotides or Affymetrix.
It is also capable of storing experiment working processes, protocols, array designs
and analyzing results. MAGE-OM has been translated into an XML-based data
format, MAGE-ML, to facilitate the exchange of data.

14.3.2 MAGE-ML; an XML-translation of MAGE-OM

MicroArray Gene Expression Mark-up language – MAGE-ML – is an XML-based
file format able to capture all MIAME required information, and its derived directly
from MAGE-OM. For example, the manufacturer of a chip may deliver to a user a
MAGE-ML file containing information about the chip design and production pro-
cess. Then the user can import the file directly to a local database. Because of
the complexity of the model it’s not quite human readable (though it’s XML), but a
software is needed to support importing and exporting MAGE-ML. Implementation
of MAGE-ML include Rosetta Biosoftware, which supports it, and MIAMExpress
that produces files in that format. The EBI has developed a freely available soft-
ware tool kit (MAGE-STK) that eases the integration of MAGE-ML into end users’
148 DNA microarray data analysis

systems.

14.3.3 MAGE-STK
The MAGE Software Toolkit is a collection of Open Source packages that im-
plement the MAGE Object Model in various programming languages. The suite
currently supports three implementations: MAGE-Perl, MAGE-Java, and MAGE-
C++. The idea is to be able to have an intermediate object layer that can then be
used to export data to MAGE-ML, to store data in a persistent data store such as a
relational database, or as input to software-analysis tools.

14.4 Where and how to submit your data

14.4.1 ArrayExpress and GEO
There are two main public repositories for microarray data, ArrayExpress
(http://www.ebi.ac.uk/arrayexpress) at the EBI, and Gene Expression Om-
nibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) at NCBI. The former sup-
ports MIAME format. Both of them provide you with an accession number that can
be used in the publication analogous to GenBank ID of a sequence. These are ac-
cepted databases to dispose your data by, e.g., Nature and The Lancet. Since GEO
currently isn’t MIAME-compliant we will here concentrate more on ArrayExpress.
ArrayExpress supports all the requirements specified by the MIAME standard,
as developed by the MGED group and aims to store well-annotated data. In addi-
tion to sending your data into ArrayExpress, you can search, browse and retrieve
(in MAGE-ML format) other microarray data from it. Currently (spring 2003) Ar-
rayExpress contains 19 publicly available experiments. The amount of submitted
data is and will be naturally larger, but there’s a lag period before new data have
been curated and published.
In addition to ArrayExpress and GEO, there are several species- or site-specific
public data repositories available, such as the Stanford Microarray Database (SMD,
http://genome-www5.stanford.edu/MicroArray/SMD/). For the review and
a www page listing of these public repositories, please see a paper by Gardiner-
Garden and Littlejohn (2001), and visit at http://ihome.cuhk.edu.hk/ b̃400559/
arraysoft_public.html. However, the way data are presented in many of these
special gene expression databases is quite diverse, or sometimes only the analyzed
data have been made available, which makes it difficult to retrieve and evaluate the
data, and impossible to co-analyze them together with your experiments.
Besides direct submission using MAGE-ML (to ArrayExpress) or SOFT (to
GEO), microarray data can be submitted to the public databases via web interfaces.
Next we will descibe some basic aspects of such submission tools.

14.4.2 MIAMExpress
MIAMExpress (http://www.ebi.ac.uk/miamexpress/) is a MIAME-compliant
microarray data submission tool. It has a web interface that allows you in a step-by-
step manner to submit your microarray data to ArrayExpress database. First you
14 Reporting results 149

will create an account, log in to that account, and then select a submission type:
Protocol, Array design or Experiment. Protocol and Array design can be submitted
individually, but the Experiment always need to be accompanied by details about
Protocol and Array Design to complete the submission.
In addition to MIAME attributes, you can add other qualifiers. If the microar-
ray data is being submitted to a journal, the journal name and status (submitted/in
press/accepted) should be indicated. After filling the information about a certain
protocol and array design, you next link that information to your experiment. The
actual experiment results are submitted as scanned files (raw data), that can be ei-
ther CEL files for Affymetrix data, or .gpr files, which contain two wavelengths
in a same file, for others. If you wish, you can also add a data file corresponding
to the transformation of all the results generated from the raw datafiles in a given
experiment. In that case, the submitted protocol should describe all steps necessary
to allow a third party to recreate the transformed data file.

14.4.3 GEO
GEO web interface works in a similar manner, although the vocabulary differs;
first you log in, identify yourself, and then define your platform (array type and
specifications, annotated gene list as tab-delimited text file, organism used, etc.)
using pop-up menues. Next you bring in sample data (scanned file) in either of
two standard sample data table formats, one for one- and another for two-channel
data. These tables contain columns with standard headings, which can be either re-
quired or optional, and non-standard headings, which are user-defined and always
optional. User-defined columns may contain useful information for the submit-
ter and other users, but queries on these columns are not supported within GEO.
Samples can currently be of four types: single channel, dual channel, comparative
genomic, or SAGE. Samples (yours and others in GEO database) can be grouped
in series such as dose-response or time course series, or as repeat samples group.

14.4.4 Other options and aspects

There are several other MIAME-compliant data management systems or software.
To see a current list, please visit http://www.mged.org/Workgroups/MIAME/
miame_software.html. One of these is an open-source application called BASE
(BioArray Software Environment, http://base.thep.lu.se/). BASE consists
of array production LIMS (see below), a relational database, and a web interface
for management and analysis of microarray data. Additional analysis tools can be
added as plug-ins. Although promising, BASE is still in a fairly early developmen-
tal stage.
In the future, CSC may provide a centralized place and tools to submit your
microarray data to, e.g., ArrayExpress.
Since microarray data is connected to much more information than just that
required by MIAME standards, it may be reasonable to set up a LIMS (Laboratory
Information Management System) for your laboratory for a storage place to all the
information related to the microarrays, especially if you use custom-made arrays
instead of commercial ones.
150 DNA microarray data analysis

14.5 MIAME-compliant sample attributes in GeneSpring

GeneSpring has a sample manager, a small database in XML-format, where you
can, when loading the data in, define numerous MIAME-compliant attributes. Gene-
Spring contains a reduced list of attributes by default, but a complete, up-to-date
list can be downloaded from Silicon Genetics’ web site. It is advisable to use pre-
defined attributes to describe your experiments and samples. A sample information
can be retrieved in XML-format from the genome’s data-folder.

14.6 Suggested reading

1. Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P., Stoeck-
ert, C., Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., Gaasterland,
T., Glenisson, P., Holstege, F. C., Kim, I. F., Markowitz, V., Matese, J. C.,
Parkinson, H., Robinson, A., Sarkans, U., Schulze-Kremer, S., Stewart, J.,
Taylor, R., Vilo, J., and Vingron, M. (2001) Minimum information about a
microarray experiment (MIAME)-toward standards for microarray data. Nat.
Genet. 29, 365-71.

2. Brazma, A., Parkinson, H., Sarkans, U., Shojatalab, M., Vilo, J., Abeygu-
nawardena, N., Holloway, E., Kapushesky, M., Kemmeren, P., Garcia, L. G.,
Oezcimen, A., Rocca-Serra, P., and Sansone, S-A. (2003) ArrayExpress-a
public repository for microarray gene expression data at the EBI, Nucleic
Acids Research 31, 68-71.

3. Edgar, R., Domrachev, M., and Lash, A. E. (2002) Gene Expression Om-
nibus: NCBI gene expression and hybridization array data repository, Nu-
cleic Acids Research 30, 207-10.

4. Gardiner-Garden, M. and Littlejohn, T. G. (2001) A comparison of microar-

ray databases, Briefings in Bioinformatics 2, 143-158.

5. Gollub, J., Ball, C.A., Binkley, G., Demeter, J., Finkelstein, D. B., Hebert, J.
M., Hernandez-Boussard, T., Jin, H., Kaloper, M., Matese, J. C., Schroeder,
M., Brown, P. O., Botstein, P., and Sherlock, G. (2003) The Stanford Mi-
croarray Database: data access and quality assessment tools, Nucleic Acids
Research 31, 94-96. http://genomebiology.com/2002/3/9/research/
0046.1

6. Saal, L. H., Troein, C., Vallon-Christersson, J., Gruvberger, S., Borg, Å., and
Peterson, C. (2002) BioArray Software Environment: A Platform for Com-
prehensive Management and Analysis of Microarray Data, Genome Biology
3, software0003.1-0003.6.

7. Spellman, P. T., Miller, M., Stewart, J., Troup, C., Sarkans, U., Chervitz, S.,
Bernhart, D., Sherlock, G., Ball, C., Lepage, M., Swiatek, M., Marks, W. L.,
Goncalves, J., Markel, S., Iordan, D., Shojatalab, M., Pizarro, A., White, J.,
Hubley, R., Deutsch, D., Senger, M., Aronow, B. J., Robinson, A., Bassett,
14 Reporting results 151

D., Stoeckert, C. J. Jr, and Brazma, A. (2002) Design and implementation of

microarray gene expression markup language (MAGE-ML), Genome Biol-
ogy 3, research0046.1-0046.9.

8. Stoeckert, C. J., Causton, H. C., and Ball, C. A. (2002) Microarray databases:

standards and ontologies, Nature Genetics 32, 469-473.

This chapter was written by M. Minna Laine, and Teemu Toivanen.

152 DNA microarray data analysis

15 Software issues
Traditionally, biological data, including DNA and amino acid sequencies and DNA
microarray results, have been stored in a flat file format. Flat files are basically just
text files, which have been formatted in such a way that programs can read their
contents. Flat files are not the ideal solution for long term storage of large data sets.
They also cause problems when the same data needs to be analyzed using several
programs that use a different kind of file format.
The data standardization and storage in central databases (see chapter 14 for
more details) is one solution to the data format issues. However, this solution has
not yet fully matured, and most of the programs used for the DNA microarray
data analyses are not able to read the data directly from these databases or even
import the XML-format the databases commonly produce. Databases and programs
will surely learn to cooperate better over time. The standardization and storage of
the data is currently under heavy development, and many software developers are
implementing their own solution to the these problems. However, the solutions are
not always MIAME compliant.

15.1 Data format conversions problems

Especially, if freeware tools are used for the analysis, there is often a need to ex-
port the partly analyzed data from one program to another. More often than not, the
datafile formats are not compatible, which means additional and tedious conversion
work. Data can quite easily be converted between formats using Microsoft Excel
or any other spreadsheet program. However, the work load becomes quickly un-
bearable when the amount of data increases. Excel has some macro programming
capacities, but there exists also other, and possibly better tools, if one is interested
in programming (see section on programming). In addition, Excel does not tolerate
huge data files (more than 65 536 rows or 256 columns). Sometimes it becomes
necessary to transpose (switch the rows and columns) the data file, and if there are
more than 256 genes, Excel can’t be used.

15.2 A standard file format

If the data needs to be imported into several programs at the same time, it is often
easier if a “standard file” is first generated from the files written by the chip scanner.
Standard files can be produced in Excel or another spreadsheet program easily.
Briefly, the first column of the standard file should contain the gene identifiers, for
15 Software issues 153

example their common names. The second column should consist of the sequence
accession numbers, Genbank accession numbers are probably the most usable ones.
The next four columns should contain the spot and background intensities of the
used colors. Depending on your analysis needs, it is also possible to use ratios, for
example normalized log ratios, instead of intensity values. The standard file should
be saved in a tab-delimited text format.

15.3 Programming
When the capacities of the spreadsheet program are not sufficient, some program-
ming tools can be used instead for the datafile management purposes or for data
analyses. There are actually several programming languages that are especially
suited for the data file formatting and other text file manipulations.

15.3.1 Perl
One of the most common languages is Perl, which has very powerful and easy to
use text manipulation tools. Perl is available for free for UNIX, Linux http://
www.perl.org and PC machines http://www.activeperl.com. Perl is a pro-
gramming language, which means that getting to know it takes some time and ef-
fort. However, if data conversion tools are needed everyday, it would definitely be
worthwhile to befriend Perl.
To illustrate how easily text can be manipulated using Perl, we present a short
example. The next code produces a complementary DNA sequence from the origi-
nal sequence, which has been stored into the text string $sequence. Both the origi-
nal sequence and its complement are printed on the computer screen.
The program starts with a line that tells the computer where The Perl software
can be found. The next four lines contain the actual commands that manipulate
the DNA sequences. Note that every command line has to end with a semicolon.
Function tr makes a complementary sequence from the original one, and function
print outputs the result to the screen.

#!/usr/bin/perl
$sequence="aaattcgagtaggtcaggcat";
print "Original: $sequence\n";
$sequence=~ tr/acgtACGT/tgcaTGCA/;
print "Complementary: $sequence\n";

You can use pico editor on CSC’s Cedar server to create a similar file and test
the example yourself. Perl programs are started in Cedar with a command perl
filename.

15.3.2 Awk
Awk is a standard UNIX and Linux tool, which is available on CSC’s servers.
With Awk, individual columns can be easily extracted from tab-delimited text files.
Using other standard UNIX tools, these individual columns can be saved into a new
154 DNA microarray data analysis

text file. For example, the next script takes the first column from a specified datafile
and saves it into a new file.

Awk ’{print$1}’ datafile > newfile

Two columns can be “awked” into a new file next to each other separated with
a space:

awk ’{print\$1, \$2}’ datafile > newfile

or one after another:

awk ’{print\$1}{print\$2}’ datafile > newfile

15.3.3 R
R is a free statistical analysis tool and a self-sufficient programming language. It
is available for UNIX, Linux, Macintosh and PC platforms. In R, scripts for anal-
yses and data file manipulations can be easily constructed. R has many add-on
packages for cluster analysis, self-organizing maps, and neural networks, among
others. There are also some packages available that have been specifically tailored
for DNA microarray data analysis.
Here is an example on how to read the tab-delimited datafile to R and how to
process it into a new table, which is then written out to a new file. The function
read.table reads in the specified file with the headers. The table is then saved in a
variable data. Two columns are extracted from the data, and saved into new variable
(x and y). The new variables are used for the creation of a new table (dataout),
which is then written to a new text file (filenameout.txt). Such a script can easily
be automated using R, and the analyses can simultaneously be intergrated with the
datafile conversions.

data<-read.table("filename.txt", header=T)
x<-data$greenintensity
y<-data$redintensity
dataout<-cbind(x,y)
sink("filenameout.txt")
dataout
sink()

Many images included in chapters 5–8 have been produced by R using real
DNA microarray datasets.

15.4 Freeware software packages

There are several software packages that are either totally free or free for academic
researches. We have experience with quite a few software packages, a couple of
which we introduce here.
15 Software issues 155

15.4.1 Cluster and treeview

Cluster and Treeview are among the first free software tools for DNA microarray
data analysis. They are still heavily used for clustering, i.e. production of expres-
sion heatmaps. These are small and easy to use tools for beginners.

More information: http://rana.lbl.gov/EisenSoftware.htm

15.4.2 Expression profiler

A web-based clustering tool Expression Profiler has been developed at the EBI.
Expression profiler performs clustering by various algorithms and distance or dis-
similarity methods. It also has nice links to sequence, metabolome, and pathway
databases. At the moment the system works best with yeast data, because the links
to the outside databases are more extensively implemented for yeast.

More information: http://www.ebi.ac.uk/microarray/ExpressionProfiler/

ep.html

15.4.3 ArrayViewer
The Institute for Genomic Research (TIGR) has released a couple of software pack-
ages for DNA microarray data analysis. ArrayViewer comes in two versions. One
is suited for viewing the results from one chip only, the other can cope with multiple
slides at the same time. Analysis options cover some basic filtering and normaliza-
tion methods and various clustering algorithms.

More information: http://www.tigr.org/software/

15.4.4 MAExplorer
MAExplorer can be used as a web-based tool or a desktop program. It performs
clustering, but also normalization, and has the best statistical tools among the free
programs. The program contains a possibility to use a tailored database. MAEx-
plorer is also under constant development, and new features are frequently added.

More information: http://maexplorer.sourceforge.net/

15.4.5 Bioconductor
Bioconductor is an attempt to build a DNA microarray data analysis environment
on top of R (see section 14.3.3). The broad goals of the projects are to provide
access to a wide range of powerful statistical and graphical methods for the analy-
sis of genomic data; facilitate the integration of biological metadata in the analysis
of experimental data: e.g., literature data from PubMed, annotation data from Lo-
cusLink; allow the rapid development of extensible, scalable, and interoperable
software; and promote high-quality documentation and reproducible research.

More information: http://www.bioconductor.org and www.r-project.org

156 DNA microarray data analysis

15.5 Commercial software packages

Many commercial software packages have been developed for DNA microarray
analysis during the last couple of years. The ones we have experienced with are
briefly presented here.

15.5.1 VisualGene

VisualGene is a powerful data exploration tool. Algorithms are based on self-

organizing maps, Summon’s mapping, and Fuzzy modeling. In addition, some
filtering tools are implemented to pretreat the data before cluster analyses.

More information: http://www.visipoint.fi/

15.5.2 GeneSpring

GeneSpring is a general analysis tool specifically tailored for DNA microarray data
analysis. The intented group of users are the biologists, who actually perform the
experiments. The program contains various data preprocessing (filtering, normal-
ization) and clustering tools. Gene annotation can also be retrieved directly from
web-based databases. GeneSpring can be expanded with user-made scripts or pro-
grams (Java APIs).

More information: http://www.sigenetics.com/

15.5.3 Kensington

Kensington Discovery Edition offers a heavy data mining tool for DNA microar-
ray analyses. Kensington has a visual programming language that describes the
dataflow through different analysis steps. Kensington has more or less the same
analysis tools as GeneSpring, but it implements the database connections in a more
sensible fashion, for example, data can be retrieved from any field of the Genbank
report. Kensington can be expanded with user-made modules containing the new
algorithms.

More information: http://www.inforsense.com/

15.5.4 J-Express

J-Express is another appealing commercial tool. It has some filtering and other
preprocessing tools, but the main emphasis is on various clustering algorithms and
the visual examination of normalization results. It has nice links to web-based
databases. J-Express is also user customizable, because plug-ins, which allow
custom algorithm implementation, can be constructed with Java-programming lan-
guage. There is also a free version of the program available.

More information: http://www.ii.uib.no/∼bjarted/jexpress/

15 Software issues 157

15.5.5 Expression Nti

Expression Nti is a newcomer in the field of DNA microarray data analysis tools.
It has welded together some analysis tools and a tailored database. The analysis
tools contain filtering and normalization, and various clustering algorithms. The
database system can read directly from any user-made GEML-based database.

More information: http://www.informaxinc.com/solutions/xpression/

15.5.6 Rosetta Resolver

Rosetta Resolver is a huge analysis tool, which combines analysis tools and a
database under the same program. The novelties of Rosetta Resolver are the built-
in error models, which are made for each DNA microarray format separately. Error
models for home-made microarrays can also be constructed. Because of error mod-
els, Rosetta Resolver also has advanced tools for the detection of outliers.

More information: http://www.rosettabio.com/products/resolver/

default.htm

15.5.7 Spotfire
Spotfire is a data mining tool that has been modified to suit DNA microarray anal-
yses. Spotfire runs in a web-browser, and different analysis tools are loaded on the
user’s machine on as-needed basis. Spotfire has very nice tools for visualization
of the microarray data. Because it is based on general statistical tools, it also has
some well-implemented functions for outlier detection and the assessment of the
statistical significance of the results.

More information: http://www.spotfire.com/

This chapter was written by Jarno Tuimala.

158 DNA microarray data analysis

Index

A principal component analysis, 112

Affymetrix self organizing map, 110
change call, 29 principles, 108
change p-value, 28 visualization, 114
comparison analysis, 27 coefficient of variation, 44
detection call, 25, 26 constant, 42
detection p-value, 25 correlation, 45
discrimination score, 26 Pearson’s, 45
Gene Array scanner, 25 Spearman’s, 45
Genechip, 25
Microarray Suite, 25 D
normalization, 28 degrees of freedom, 49
robust, 28 distribution, 42
scaling, 28 normal, 47, 77
signal, 25, 26 skewed, 49, 77
signal log ratio, 29 left, 50
single array analysis, 25 right, 50
Tau, 26 standard normal, 48
annotation, 140 t, 49
retrieval dose response, 38
BLAST, 141 dye-swap, 94
public databases, 140 dynamic range, 81
article mining, 141
E
B EM-algorithm, 101
background subtraction, 68 error
Bayes rule, 101 bias, 43
random, 43, 52
C systematic, 43
case-control studies, 72 expectation-maximization, 101
cDNA microarrays, 25 experimental design, 38
centralization, 88, 89 available technology platforms, 39
Chen’s method, 102 controls, 38
clustering cell lines, 39
function prediction, 117 human, 38
methods, 109 mice, 39
hierarchical, 109 gene clustering and classification,
k-means, 111, 114 40
Index 159

replicates, 39 positive control genes, 96

expression change, 69 specific samples, 98
fold change, 71 two-color data, 98
intensity ratio, 69 warnings, 98
log ratio, 70 one sample t-test, 62
ontologies, 142
F Pearson correlation, 61
filtering, 74 principal component analysis, 118
absolute expression change, 100 promoter analysis, 138
bad data, 74 ratio, 60
flagging, 75 replicates, 82
outliers, 74 scatter plot, 60, 83
uninteresting data, 76 standard error, 60
flagging, 75 standard error of the mean, 60
two-sample t-test, 62
G
genotyping, 31
Gene regulatory networks, 121
Bayesian network, 123 H
parameters, 124
histogram, 49, 50
structure search, 126
housekeeping genes, 25, 90
fundamentals, 121
perturbation data, 122 I
time series data, 122 imputation, 54
GeneSpring
annotations, 142 L
ANOVA, 62 linear regression, 46
average, 60 linearity, 78
background correction, 82 log ratio, 52
classification, 119 log_2-transformation, 70
clustering, 117 log-transformation, 52, 88
experimental parameters, 82
expression change, 82 M
filtering, 83 MA plot, 87
fold change, 60 MAGE, 147
genelist, 84 MAGE-ML, 147
histogram, 61, 83 MAGE-OM, 147
importing data, 82 MAGE-STK, 148
linear regression, 61 Mann-Whitney U test, 55
linearity check, 83 mean, 77, 81
log of ratio, 60 arithmetic, 43
maximum, 60 trimmed, 43
MIAME-compliant attributes, 150 median, 43, 77
minimum, 60 MGED, 146, 148
normalization, 96 MIAME, 145
constant value, 96 microarray
dye-swap, 96 dyes
one-color data, 98 Cy3, 19
per chip, 96 Cy5, 19
per gene, 96 hybridization, 20
160 DNA microarray data analysis

obtaining, 17 P
printing, 16 pheasant tail, 68
RNA sample preparation, 19 plot
scanning, 20 normal probability, 51
typical applications, 21 power analysis, 72
Microarray databases preprocessing, 66
ArrayExpress, 148 Promoter data mining, 129
GEO, 149 Promoter datamining
MIAMExpress, 148 pattern databases, 136
missing values, 53, 66 pattern search, 137
casewise deletion, 66 retrieving sequences, 130
mean substitution, 67 EnsMart, 133
pairwise deletion, 67
multiple chip methods, 104 R
one-sample t-test, 106 range, 44
standard deviation, 104 replicates
two-sample t-test, 105 averaging of, 72
biological, 71
N case-control studies, 72
Newton’s method, 103 chips
noise envelope, 101 checking quality, 73
normality, 77 excluding bad replicates, 73
normalization, 81, 85, 88 handling, 71
analysis of variance, 94 power analysis, 72
dye-swap, 94 software, 71
global, 89 spots
housekeeping genes, 90 checking quality, 73
linearity of data, 91 technical, 71
local, 89 time series, 71
lowess, 91, 93
mean centering, 92 S
median centering, 91, 92 sample size, 43
per-chip, 89 Sapir and Churchill’s method, 101
per-gene, 89 scatter plot, 44
ratio statistics, 94 M versus A, 87
spiked controls, 90, 94 signal-to-background, 75
standardization, 92 signal-to-noise, 75
trimmed mean centering, 92 single chip methods, 100
number of subjects, 43 Chen’s method, 102
Newton’s method, 103
O noise envelope, 101
oligonucleotide pairs, 25 Sapir and Churchill’s method, 101
mismatch, 25 single nucleotide polymorphism, 31
perfect match, 25 SNP, 31
ontology, 140 genotype calls, 32
outliers, 52, 74 methods
quantification error, 74 APEX, 31
spot saturation, 74 single base extension, 31
statistical modeling, 74 software
Index 161

ArrayViewer, 155 U
awk, 153 UPGMA, 109
bioconductor, 155
Cluster, 155 V
Expression Nti, 157 variable, 42
Expression Profiler, 155 dependent, 42
GeneSpring, 156 independent, 42
J-Express, 156 qualitative, 42
quantitative, 42
Kensington, 156
variables, 42
MAExplorer, 155
variance, 44, 77, 81
Perl, 153
Volcano plot, 106
R, 154
Rosetta Resolver, 157 Z
Spotfire, 157 Z-score, 89
Treeview, 155
VisualGene, 156
spatial effects, 79
spiked controls, 90
standard deviation, 44, 77
standardization, 48, 88, 89
statistical testing, 54
ANOVA, 55, 58
completely randomized, 58
Bonferroni correction, 58
choosing the test, 55
critical values table, 57
hypothesis pair, 55
Kruskal-Wallis test, 55
one-sample t-test, 56
p-value, 55
t-test, 55
test statistic, 56
two-sample t-test, 56
systematic bias, 85
array design, 87
batch effect, 87
dye effect, 85
experimenter issues, 87
plate effects, 86
printing tip, 86
reporter effects, 86
scanner malfunction, 85
uneven hybridization, 86
systematic variation, 85

T
time series, 38, 71