Introduction to R for gene expression

data analysis
Please read this document carefully. If already familiar with R syntax, you may skip the first section
and move on to the two study cases. Throughout the document, blue text represents R commands
(on lines starting with the ‘>’ character) and output from the R console. This is provided as code
examples for you to run during the tutorial, and observe the output.

R basics
A detailed introduction to R can be found at http://cran.r-project.org/doc/manuals/R-intro.html

Variables and data types:

 R is not a strongly typed programming language, so different value types can be stored in
the same variable. Eg:

> a=2

> a

[1] 2

> a='2'

> a

[1] "2"

 Available basic data types: numeric (double, integer), logical, character, vector, matrix.
o Vector examples:

> a=c(1:10)

> a

[1] 1 2 3 4 5 6 7 8 9 10

> a[a>7]

[1] 8 9 10

> a[3]

[1] 3

> a[-1]

[1] 2 3 4 5 6 7 8 9 10

> a[4:8]
[1] 4 5 6 7 8

> length(a)

[1] 10

> length(a)=3

> a

[1] 1 2 3

o Matrix examples:

Create a 3X3 matrix:

> a=matrix(1:9,3,3)

> a

[,1] [,2] [,3]

[1,] 1 4 7

[2,] 2 5 8

[3,] 3 6 9

Access a row in the matrix:

> a[1,]

[1] 1 4 7

Access a column:

> a[,1]

[1] 1 2 3

Access an element:

> a[2,2]

[1] 5

> a[2,2]=5.5

> a

[,1] [,2] [,3]

[1,] 1 4.0 7

[2,] 2 5.5 8

[3,] 3 6.0 9

Create a 2x2 matrix:

> b=array(2:3,1:2,dim=c(2,2))

> b

[,1] [,2]

[1,] 2 2

[2,] 3 3

Access more elements in matrix a:

> a[b]

[1] 5.5 9.0

 More advanced data types: lists (heterogeneous vectors), data frames (heterogeneous
matrices).

Create a list:

> l=list("2",3,7,c(4,2,3))

> l

[[1]]

[1] "2"

[[2]]

[1] 3

[[3]]

[1] 7

[[4]]

[1] 4 2 3

Access an element in the list:

> l[[4]]

[1] 4 2 3

Create a data frame:

> d=data.frame(c(1,2,3),c("d","f","t"))

> d

c.1..2..3. c..d....f....t..

1 1 d

2 2 f
3 3 t

Access and element in the data.frame:

> d$c.1..2..3.

[1] 1 2 3

 Checking for type, type conversion: is.numeric (double, integer), is.vector,

is.matrix, as.numeric, as.vector, as.matrix. Eg:

> a='3.4'

> is.numeric(a)

[1] FALSE

> is.double(a)

[1] FALSE

> a=as.double(a)

> a

[1] 3.4

> is.double(a)

[1] TRUE

> is.integer(a)

[1] FALSE

> a=as.integer(a)

> a

[1] 3

File input/output: we will be using mostly data frames and matrices.

 read.table() function: allows reading from text file containing tabular information.

>data=read.table("fileIn.txt", header=TRUE)

 write.table() allows writing a data frame (can be a matrix) to a file.

>write.table(data, "fileOut.txt");

Plotting:

 Various plotting functions exist, and most Bioconductor packages extend these.
 plot() - high level function for plotting. E.g. scatter plots, line plots:
> x=c(3,5,2,7,4,9,2)

> plot(x)

> plot(x, type='l')

> y=c(4,2,1,7,9,6,2)

> plot(x,y)

 hist() - histogram

>hist(x)

 Labels: xlab, ylab - axis labels, main - axis title.

R functions:

 Used to write reusable code.

 Syntax: name <- function(arg_1, arg_2, ...) expression

> add<-function(x,y) {

+ sum=x+y

+ sum

+ }

 Usage:

> add(2,3)

[1] 5

 If you prefer storing your functions or code in a file, you can run that file using

>source("fileName", echo=TRUE)

R packages:

 The main advantage of R stands in the wide availability of free software packages for
statistical analysis.
 Loading a package: library(affy), displaying contents of a package :
library(help=affy).
 Packages for expression data analysis can be found on the Bioconductor page:
http://www.bioconductor.org/packages/2.6/bioc/ . These can be installed through the R
interface menu (Packages->Install packages), after setting the correct repositories (Packages-
>Select repositories -> select all). Here, we will be using packages affy
http://www.bioconductor.org/packages/2.6/bioc/html/affy.html and limma
 http://www.bioconductor.org/packages/2.6/bioc/html/limma.html . You can find more
details and manuals on these web pages.

Datasets used in this tutorial (Yeast cell cycle):

 Dual-channel: Hasse dataset

 Single-channel: Pramila dataset, Spellman dataset
 Download archive and unpack : http://www.computing.dcu.ie/~asirbu/R.zip

Dual channel microarray data - LIMMA package

Pramila dataset
We will start with dataset Pramila. For this, we need to load the limma package, and change the
working directory to that containing the raw data:

>library(limma)

In the R menu, go to File->Change dir and select the corresponding directory (after unpacking the
downloaded archive, this should be Pramila). Please note that in this directory there is a text file
called targets.txt that contains information on samples hybridised to each chip. This does not come
with the raw data, but needs to be created by you. In our case, the data contain time conditioned
samples hybridised against a reference.

Loading raw data

 Reading in data from .gpr file (GenePix microarray image reader):

> targets=readTargets("targets.txt")

> targets$FileName

[1] "GSM81064.gpr" "GSM81065.gpr" "GSM81066.gpr" "GSM81067.gpr"

"GSM81068.gpr"

[6] "GSM81069.gpr" "GSM81070.gpr" "GSM81071.gpr" "GSM81072.gpr"

"GSM81073.gpr"

[11] "GSM81074.gpr" "GSM81075.gpr" "GSM81076.gpr"

> RG <- read.maimages(targets$FileName, source="genepix")

 Contents of RG:

> names(RG)
[1] "R" "G" "Rb" "Gb" "targets" "genes" "source"
"printer"

R,G are the red and green channel values, Rb and Gb the background noise, genes contains
information on chip probes. The can be displayed at the console:

> head(RG$genes)

Block Row Column ID Name

1 1 1 1 YDR407C TRS120

2 1 1 2 YDR180W SCC2

3 1 1 3 YAR050W FLO1

4 1 1 4 YKL129C MYO3

5 1 1 5 YOR328W PDR10

6 1 1 6 YJR138W

Visualisation

 Visualisation with MA plots allows plotting log ratios (R vs G) against average spot intensity.

> plotMA(RG,"GSM81069")

In the resulting plot, most genes need to cluster around 0 M value, which indicates that they
are not differentially expressed, compared to the reference sample. Also, there should be no
dependence of the M values on the intensity. In case these are not true, normalisation is
required.
 Background intensities can be also visualised for each array (this helps identify arrays with
large intensities):

> boxplot(log2(RG$Rb))

 The plot() function can be used to visualise the time series for the red or green channel for a
specific gene (gene on position 100 in the example):

> plot(RG$G[100,],type='l')

 The plotDensities() function can be used to visualise red and green intensities (as density
plots) on the different arrays:

Normalisation

 Normalisation of microarray data starts with background adjustment, then performs within
array normalisation to remove effects of average intensities on the log ratios. Several
 different approaches for this exist, please see limma manual for details. Here, we will use
normexp background correction and loess normalisation.

> RGbk <- backgroundCorrect(RG, method="normexp", offset=50)

> MA <- normalizeWithinArrays(RGbk, method="loess")

 We can now look at the corresponding MA plots for RGbk and MA, and observe how the
dependence between M and A is reduced:

> plotMA(RGbk,"GSM81069")

> plotMA(MA,"GSM81069")

 Also, if we plot densities, we can see that now the two channels in each array overlap better
than before loess normalisation:

> plotDensities(RGbk)

> plotDensities(MA)

 A further step is between array normalisation, which aims at removing differences between
two arrays. This is useful when comparing different arrays or for modelling approaches. We
can apply quantile normalisation for this.

> MA.b=normalizeBetweenArrays(MA, method="quantile")

 If we plot densities again, we can see that all arrays overlap significantly.

> plotDensities(MA.b)

Spellman dataset
Several other types of files can be automatically read by function read.maimages(). To use these,
the parameter 'source' needs to be changed into e.g. 'agilent' or 'smd', etc. In case the file format is
not available, we can read from text files containing tabular data, by specifying the labels for the red
and green channel, and for the background. We can do this for the Spellman dataset (change
working directory to Spellman):

> targets=readTargets("targets.txt")

> targets$FileName

[1] "13.txt" "14.txt" "15.txt" "16.txt" "17.txt" "18.txt" "19.txt"

"20.txt" "21.txt" "22.txt" "23.txt"

[12] "24.txt" "25.txt" "26.txt" "27.txt" "28.txt" "29.txt" "30.txt"

> RG <- read.maimages ( targets$FileName, columns = list ( R =
"ScanAlyze:RESULT.CH1I_MEAN", G = "ScanAlyze:RESULT.CH2I_MEAN", Rb =
"ScanAlyze:RESULT.CH1B_MEDIAN", Gb = "ScanAlyze:RESULT.CH2B_MEDIAN"))

 Having the data loaded, we can now proceed to the analysis as above. You should try and
adapt the above commands to this new dataset.

Affymetrix data - affy & limma packages

 Change the working directory to Hasse, and load the affy package:

> library(affy)

 Reading in the raw data:

> data= ReadAffy()

 Data visualisation:

Box plots for each array:

> boxplot(data)

Density graphs for each array:

>hist(data)

MA plots of each of the first 4 arrays against a median reference:

> par(mfrow=c(2,2))

> MAplot(data[,1:4], method=”smoothScatter”)

MA plots of first three arrays (against each other):

> MAplot(data[,1:3], method=”smoothScatter”, pairs=TRUE)

 Data normalisation and computation of expression values:

RMA

>rma=rma(Data)

MAS5

>mas5=mas5(Data)

Custom normalisation, using expresso to control all steps:

> custom = expresso(Data, normalize.method ="qspline",
bgcorrect.method="rma", pmcorrect.method="pmonly",
summary.method="liwong")

For details on available methods for each normalisation step, you can see affy Biocondutor web
page . There are several commands that also display this information:

> bgcorrect.methods()

[1] "bg.correct" "mas" "none" "rma"

> express.summary.stat.methods()

[1] "avgdiff" "liwong" "mas" "medianpolish"

"playerout"

> pmcorrect.methods()

[1] "mas" "methods" "pmonly" "subtractmm"

> normalize.AffyBatch.methods()

[1] "constant" "contrasts" "invariantset" "loess"

"methods" "qspline"

[7] "quantiles" "quantiles.robust"

 Normalised data visualisation:

Same as above for hist and boxplot, just using exprs(rma), exprs(mas5) or
exprs(custom) instead of data. For example, to obtain the boxplot of expression values for each
array after RMA normalisation (which can be compared to that before normalisation):

>boxplot( exprs(rma))

 Writing expression values to file:

>write.exprs(rma,file=”normalisedData.txt”)

Remarks
This document represents a quick introduction to microarray data processing using R, and includes
pre-processing only. However, the limma package offers tools for modelling and differential
expression analysis, which are detailed in the user guide on the limma web page
http://www.bioconductor.org/packages/2.6/bioc/html/limma.html . However, keep in mind that, in
general, in order to perform differential expression analysis, datasets need to contain replicates. A
good source for microarray data is GEO database (http://www.ncbi.nlm.nih.gov/geo/ ), where you
can look for different types of experiments (not necessarily time series, as these examples here), and
try to perform differential expression analysis.
Also, for those interested in Next Generation Sequencing data, GEO does contain these as well, and
a recent R package for analysis is DESeq
http://www.bioconductor.org/packages/release/bioc/html/DESeq.html

A few datasets that may be of interest (time series data):

Drosophila microarray data:

Single channel:

http://www.fruitfly.org/cgi-bin/ex/insitu.pl

Dual-channel :

http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE14086

Drosophila NGS (RNAseq) data:

http://www.ncbi.nlm.nih.gov/sites/entrez?db=sra&term=SRP001065