International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 8

ISSN: 2321-8169
2480 2484

_______________________________________________________________________________________________

Microarray gene expression ranking with Z-score for Cancer Classification

M .Yasodha ,

Dr P Ponmuthuramalingam

Research Scholar
Government Arts College ,
Coimbatore, Tamil Nadu, India

Head and Associate Professor

Government Arts College ,
Coimbatore, Tamil Nadu, India

Abstract--Over the past few decade there has been explosion in the amount of genomic data available to biomedical engineer due to the
advantage of biotechnology. For example using microarray it is possible to find out a persons gene expressions profile more than 30000
genomes. Among this one of the most important gene selection problem is gene ranking. Here we will describe Z-score ranking for microarray
gene expression selection. In that technique it choose the gene and then applied the Z-Score Ranking technique and then divides the genes into
subsets with Successive Feature selection and then finally LDA Applied for the result. With this Z-score ranking technique we will get the
accurate results and less effort. The Lymphoma and Leukemia dataset genes are utilized. The proposed technique shows capable classification
accuracy for the whole test data sets.

__________________________________________________*****_________________________________________________
1.

INTRODUCTION

Cancer has become the leading cause of death

worldwide. The most effective way to reduce cancer deaths
is to detect it earlier. Though cancer research is generally
clinical and biological in nature, data driven statistical
research has become a common complement. Predicting the
outcome of a disease is one of the most interesting and
challenging tasks where data mining techniques have to be
applied. Prevention of tobacco-related and cervical cancers
and earlier detection of treatable cancers would reduce
cancer deaths in India. The absolute number of cancer
deaths in India is projected to increase due to population
growth and increasing life expectancy.
Previously, cancer classification has always been
morphological, and clinical based. This work ranked the
whole set of 7129 genes according to their Z-scores (ZSs) in
the training data set. Then, take out the top 100 genes with
the highest ZSs. The rest of this work ordered as follow:
Section 2 offers the environment material about microarray
gene expression outline. Section 3 demonstrate and classify
the major advances that have been utilized newly for cancer
microarray gene expression profile. Section 4, offers
discussion and study about the large amount of capable
approaches that are offered through out the work. Lastly,
Section 5 concludes the work.
2. PROPOSED METHODOLOGY

z-scores. Instead of translating data to a fixed range as with

percentile rescaling, z-scores are anchored by the mean and
standard deviation of the original values, and rescaled such
that the new mean is 0 and the new standard deviation is 1.
The resulting z-scores correspond to points on the standard
normal curve, with a theoretical range of approximately -3
to +3. The actual range for each indicator, however, will be
different.
Z-scores can be calculated for individual level or
aggregate level data. In either case, each value is treated as
an "individual" member of a sample. The sample size, then,
is equal to the total number of "individuals". When z-scores
are calculated for aggregate data, such as county rates and
percents, the county is the "individual" (unit of analysis), the
group of counties is the sample, and the sample size is
therefore the total number of counties. The formula for
calculating z-scores is as follows:

The proposed method is containing of two steps. A

rank of every genes in the training data set using a scoring
method. Then, retain the genes with high scores. The
performance of t-test can be done to minimize the size of the
gene and to select the top 100 genes.
2.1 Gene Importance Ranking
In this work, compute the significance ranking of
every gene using a feature ranking measure, two of which
are described below.
A) Z-Score Methodology
The classic method for "standardizing" a set of
values (finding a common metric or scale) is calculation of
2480
IJRITCC | August 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 8

ISSN: 2321-8169
2480 2484

_______________________________________________________________________________________________

There are K classes. max

{ , = 1,2, } is the
maximum of all yk. Ck refers to class k that contains nk
samples. xij is the expression rate of gene i in sample . is
the mean expression rate in class k for gene i. n is the whole
amount of samples. is the common mean expression value
for gene i. si is the joint within-class standard deviation for
gene i. In fact, the TS utilized here is a t-statistic among the
centroid of an exact class and the overall centroid of all the

classes. An additional probable model for TS might be a tstatistic between the centroid of an exact class and the
centroid of all the additional classes. To find the minimum
gene subset when after selecting some top genes in the
importance ranking list, this attempt to classify the data set
with only one gene. This work inputs each selected gene
into LDA classifier.
3.2 Successive Feature Selection
Successive Feature Selection SFS method (SFS) a
set of 10 features is procedure single at a time that the
rate of x is taken due to memory constraint and it is
experimentally establish that the fitting values of x is equal
to or lower than 10. The output is the grade of features. In
the successive stage that the feature is reduced once at a
time and a subset of features is achieved. That the
classification accuracy using classifiers calculate, and the
top subset of features is processed to the next level. There
might be further than one top subset of features in a given
stage. A feature is dropped in level 1 that offers four
dissimilar subsets of features. The top set in level 1 {x, x2,
x4} is which is chosen for level 2. In a related way a feature
is dropped from the best set of features of level 1 into level
2, which provides three different subsets of features. The
best sets in level 2 are {x2, x4} and {x1, x2} supposing that
their classification accuracies are the similar and are
elevated than those of other subsets and the best set in level
3 is {x2} .
This procedure is finished when all the features are ranked.
Two ranked sets are achieved in SFS: that is R1= {x2, x4,
x1, x3} and R2= {x2, x1, x4, x3}

2
3
4
1

x1

x3

3
4

2
4

4
2

1 = 2 , 4 , 1 , 3

4
1

4
1

2 = 2 , 1 , 4 , 3

3
Figure 1 Successive Feature Selection

2481
IJRITCC | August 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 8

ISSN: 2321-8169
2480 2484

_______________________________________________________________________________________________
8.

If more than one set of Fb is obtained, then perform

cross-validation to get one best set (for crossvalidation, decompose training samples randomly n
times2 into training sets and validation sets using
the proportionality ratio p and compute the average
classification accuracy for all sets in Fb; select a set
if Fb for which the average classification accuracy
is the highest)
9. Repeat Steps 2-8 for another random
decomposition of training samples. Let the new
training set and validation set be defined as Tr* and
V*. This will give a best set .
10. Find the best set and its corresponding average
classification accuracy (b) using and ; i.e.,
[Fb,b] find_set &alpha (Fb,Fb*)
11. Repeat Steps 9-10 until b does not show any
improvement.
The dimensionality of the feature space is reduced
either through feature selection or through feature
extraction. Linear Discriminant Analysis (LDA) is a wellknown technique for feature selection-based dimensionality
reduction.

3.3 Block Reduction

A d-dimensional feature vector has been
partitioned into m roughly equal blocks, Sj, for j = 1m of
size h _ 10. Each block has at least r features. All the blocks
have been processed through the SFS procedure one at a
time, which yields top-r feature sets, Fj, for j 1 . . . q.
Then, the unique features of two consecutive feature sets, F1
and F2, are used to find the best top-r feature set, Fb. Next,
the unique features of Fb and F3 are used to obtain the best
set. This process is continued for all the q sets. The obtained
best top-r feature set, Fb, from the block reduction
procedure is stored for further pruning.

3.

The innovative of the work reason to discover the

ranking gene with correct cancer classifications for this
LDA classification is chosen, it is an adequately good
classifiers. The planned methodology was useful to the
openly obtainable cancer datasets namely Lymphoma and
Leukemia cancer dataset and the experimented using
MATLAB.
(i) Lymphoma dataset
Lymphoma data set holds 42 samples resultant
from diffuse large B-cell lymphoma (DLBCL) and 9
samples from Follicular Lymphoma (FL) later than 11
samples from Chronic Lymphocytic Leukaemia (CLL). The
whole data set hold 4026 genes. In this data set, a little part
of data is absent.
(ii) Leukemia dataset
The leukemia data set holds expression levels of
7129 genes taken over 72 samples. Labels point out that
which of two variants of leukemia is present in the sample.
This dataset is of the similar type as the colon cancer dataset
and can consequently be used for the similar kind of
experiments.

Figure 2 Block Reduction

1.

2.

3.
4.

5.
6.
7.

EXPERIMENTAL RESULTS

Select the r number of features to be investigated,

where 1< r < h, and select the block size h, where h
10.
Decompose the training samples randomly into a
training set (Tr) and a validation set (V) using a
proportionality ratio p1.
Partition the features of the sets (Tr and V) into m
roughly equal blocks, Sj, for j =1m.
Apply the Successive Feature Selection (SFS)
procedure on each of Sj to get the Z-Score ranked
feature set, Fj and its corresponding classification
accuracy, j, for j=1q, where q m and Fj Fl
j l,
Initialize i 1 and j 2.
Find the best features set Fb find_bestSet(Fi, Fj)
Terminate the process if j = q, or else update i b
and j j + 1, and go to Step 6.

Table 1 Dataset used in the Experiment

Dataset

Class

Number
of Gene

Training
samples

Test
samples

Lymphoma

4026

43

19

Leukemia

7129

40

19

2482
IJRITCC | August 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 8

ISSN: 2321-8169
2480 2484

_______________________________________________________________________________________________
Table 2 Accuracy and Execution Time for proposed method

Dataset

Methods

Number of
Selected Genes

Accuracy (%)

Execution Time
(Seconds)

Existing Feature
Selection with LDA

250

89

10

Proposed Feature
Selection with LDA

20

97

Existing Feature
Selection with LDA

150

83

14

Proposed Feature
Selection with LDA

25

95

11

Lymphoma

Leukemia

Table 2 shows the accuracy and execution time for feature selection in gene expression data. Figure 3 shows the
comparison of accuracy for the proposed method of feature selection with LDA classification and the Existing method of feature
selection with LDA, from the table obviously noticed that the planned technique provides improved results by their correctness in
percentage.

100
90
80

Existing Feature
Selection with
LDA

Accuracy (%)

70
60

Proposed Feature
Selection with
LDA

50
40
30
20
10
0

Lymphoma

Leukemia
Datasets

Figure 3 The accuracy for Proposed feature selection methods

Figure 4 shows the comparison of execution time in seconds for the existing feature selection with LDA classification
and the proposed feature selection with LDA used by the Lymphoma dataset and Leukemia dataset. By the similarities clearly
noticed that the proposed feature selection with LDA classification construct the improved outcome in the reduced time.

2483
IJRITCC | August 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 8

ISSN: 2321-8169
2480 2484

_______________________________________________________________________________________________

Execution Time(Seconds)

60
50

Existing Feature
Selection with
LDA
Proposed Feature
Selection with
LDA

40

30
20
10
0

Lymphoma

Leukemia

Datasets
Fig4 shows the execution time for both Existing and Proposed feature selection methods
4.

CONCLUSION

Cancer is one of the significant characteristic in the

biomedicine field. Exact calculation of numerous tumor
kinds has higher value in offering improved treatment and
toxicity decrease on the patients. In the history, cancer
classification is usually depends on morphological and
clinical analysis. These preceding cancer classification
methods are confirmed to have numerous drawbacks in their
analytical potential. To overcome those disadvantages in
cancer classification, capable technique in agreement with
the global gene expression assessment have been evolved.
This gene data must be preprocessed for classification with
good correctness using the classifier. The gene ranking
technique is utilized to preserve that task. This effort uses
enhancement score for ranking the gene. Then the classifier
is educated with that data. Finally, the classification of gene
for identifying the cancer is carried out. This proposed
technique is used to choose the top genes.

[6]

[7]

[8]

[9]

[10]

[11]

REFERENCES
[1]

[2]

[3]

[4]

[5]

Chuang, Li-Yeh, Cheng-Huei Yang, Kuo-Chuan Wu, and

Cheng-Hong Yang (2011), "A hybrid feature selection
method for DNA microarray data." Computers in biology
and medicine 41, no. 4: 228-237.
Nam, Dougu, and Seon-Young Kim(2008), "Gene-set
approach for expression pattern analysis." Briefings in
bioinformatics 9, no. 3: 189-197.
Guyon, Isabelle, and Andr Elisseeff (2003), "An
introduction to variable and feature selection." The Journal
of Machine Learning Research 3: 1157-1182.
Deutsch, J. M. "Evolutionary algorithms for finding
optimal
gene
sets
in
microarray
prediction." Bioinformatics 19, no. 1 (2003): 45-52.
Xiong, Momiao, Wuju Li, Jinying Zhao, Li Jin, and Eric
Boerwinkle (2001), "Feature (gene) selection in gene
expression-based
tumor
classification." Molecular
Genetics and Metabolism 73, no. 3: 239-247.

[12]

[13]

[14]

[15]

Yu, Lei, and Huan Liu (2004), "Redundancy based feature

selection for microarray data." In Proceedings of the tenth
ACM SIGKDD international conference on Knowledge
discovery and data mining, pp. 737-742. ACM.
Shevade, Shirish Krishnaj, and S. Sathiya Keerthi (2003),
"A simple and efficient algorithm for gene selection using
sparse logistic regression." Bioinformatics19, no. 17:
2246-2253.
Tibshirani, Robert, Trevor Hastie, Balasubramanian
Narasimhan, and Gilbert Chu (2002), "Diagnosis of
multiple cancer types by shrunken centroids of gene
expression." Proceedings of the National Academy of
Sciences 99, no. 10: 6567-6572.
Chuang, Li-Yeh, Hsueh-Wei Chang, Chung-Jui Tu, and
Cheng-Hong Yang. "Improved binary PSO for feature
selection using gene expression data."Computational
Biology and Chemistry 32, no. 1 (2008): 29-38.
Zhang, Hao Helen, Jeongyoun Ahn, Xiaodong Lin, and
Cheolwoo Park. "Gene selection using support vector
machines with non-convex penalty."Bioinformatics 22, no.
1 (2006): 88-95.
Setiono, Rudy. "Generating concise and accurate
classification rules for breast cancer diagnosis." Artificial
Intelligence in medicine 18, no. 3 (2000): 205-219.
Chu, Feng, Wei Xie, and Lipo Wang. "Gene selection and
cancer classification using a fuzzy neural network."
In Fuzzy Information, 2004. Processing NAFIPS'04. IEEE
Annual Meeting of the, vol. 2, pp. 555-559. IEEE, 2004.
Kim, Hyunsoo, Gene H. Golub, and Haesun Park.
"Missing value estimation for DNA microarray gene
expression
data:
local
least
squares
imputation."Bioinformatics 21, no. 2 (2005): 187-198.
Boln-Canedo, Vernica, Noelia Snchez-Maroo, and
Amparo Alonso-Betanzos. "A review of feature selection
methods on synthetic data."Knowledge and information
systems 34, no. 3 (2013): 483-519.
Sharma, Alok, Seiya Imoto, and Satoru Miyano. "A top-r
feature selection algorithm for microarray gene expression
data." IEEE/ACM Transactions on Computational Biology
and Bioinformatics (TCBB) 9, no. 3 (2012): 754-764.

2484
IJRITCC | August 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________