Bi-correlation clustering algorithm for determining a set of co-

regulated genes

BIOINFORMATICSvol. 25 no.21 2009

Anindya Bhattacharya and Rajat K. De

Outline

Introduction Bi-correlation clustering algorithm

(BCCA) Results Conclusion

Introduction

Biclustering Performs simultaneous grouping on genes and

conditions of a dataset to determine subgroups of genes that exhibit similar behavior over a subset of experimental condition.

A new correlation-based biclustering algorithm called bi-correlation clustering algorithm (BCCA) Produce a diverse set of biclusters of co-regulated

genes All the genes in a bicluster have a similar change

of expression pattern over the subset of samples.

Introduction

Cluster analysis Most cluster analysis try to find group of

genes that remains co-expressed through all experimental conditions.

In reality , genes tends to be co-regulated and thus co-expressed under only a few experimental conditions.

Bi-correlation clustering algorithm

Notation A set of n genes Each gene has m expression values For each gene gi there is an m-

dimensional vector , there is the j-th expression value of gi.

A set of m microarry experiments (measurements)

n genes will have to be grouped into K overlapping biclusters

}g,...,g,{gX n21

}e,...,e,{eY m21

},...,,{ 21 KCCC

ix ijx

Bi-correlation clustering algorithm

Bicluster: A bicluster can be defined as a subset of

genes possesing a similar behavior over a subset of experiments

Represented as A bicluster contains a subset of

genes and a subset of experiments where each gene in is correlated with a correlation valued greater than or equal to specified threshold , with all other genes in over the measurements in .

kC kI

kJ

),( kkk JIC )( XII kk

)( YJJ kk kI

kI

kJ)(

Bi-correlation clustering algorithm

BCCA Use person correlation coefficient for

measuring similarity between expression patterns of two genes and .

ig jg

m

l

m

lijjliil

m

lijjliil

ji

xxxx

xxxx

1 1

22

1

)()(

))((),(Corr xx

Bi-correlation clustering algorithm

Step 1: The set of bicluster S is initialized to

NULL and number of bicluster Bicount is initialized to 0

Step 2A BCCA generate a bicluster (C) for each

pair of genes in a dataset under a set of conditions

For each pair of genes .BCCA creates a bicluster , where and .

)(, jigg ji

),( JIC }{ ji ,ggI YJ

Bi-correlation clustering algorithm

In step 2C: For a pair of genes in C, if then a

sample is detected from C, deletion of which caused maximum increase in correlation value between and .

If being a threshold, the sample is deleted from . otherwise, C is discarded.

Deletion of a measurement for which genes differ in expression value the most will result in the highest increase in correlation value.

BCCA deletes one measurement at a time from .

),(Corr ji xx

3,' rrJm

ig

jg

J

J

Bi-correlation clustering algorithm

In step 2D(a): Other genes from , which satisfy

the definition of a bicluster are included in C for its augmentation.

In step 2D(b): Whether present bicluster C has been

found. If it is so then we do not to include C, otherwise, C is considered as a new bicluster.

IX

Bi-correlation clustering algorithm

Bi-correlation clustering algorithm

Results

Datasets We demonstrate the affectiveness of

BCCA in determining a set of co-regulated genes (i.e. the genes having common transcription factors) and functionally enriched clusters (and atributes) on five dataset

Results

Variation with respect to threshold Plot of YCCD dataset :

Average number of functionally enriched attributes (computed using P-values) versus correlation threshold value

Results

Follow a guideline on this value from a previous study by Allocco et al. (2004) which has concluded that if two genes have a correlation between their expression profiles >0.84 then therre is >50% chance of being bounded by a common transcription factor.

Results

By locating common transcription factors At first, we only consider those biclusters

that have less than or equal to 50 genes. Use a software TOUCAN 2 (Aerts et al., 2005)

for performance comparison by extracting information on the number of transcription factors present in proximal promoters of all the genes in a single bicluster.

Presence of common transcription factors in the promoter regions of a set of genes is a good evidence toward co-regulation.

Results

Results

Sequences of all the five genes found in a bicluster generated by BCCA from SPTD dataset.

Any transcription factor may be found present in more than one location in upstream region.

Results

Functional enrichment : P-value

The functional enrichment of each GO category in each of the bicluster

employed the software Funcassociate (Berriz et al., 2003).

P-value represents the probability of observing the number of genes from a specific GO functional category within each cluster.

A low P-value indicates that the genes belonging to the enriched functional categories are biologically significant in the corresponding clusters.

Results

P-value of a functional category Suppose we have total population of N genes ,

in which M has a particular annotation. If we observe x genes with that annotation, in

a sample of n genes, then we can calculate the probability of that observation.

The probability of seeing x or more genes with an annotation, out of n, given that M in the population of N have that annotation

n

N

xn

MN

x

M

P

n

xj

n

N

jn

MN

j

M

valueP

Results Only functional categories with

are reported. Analysis of the 10 biclusters obtained for the

YCCD, the highly enriched category in bicluster Bicluster1 is the ‘ribosome’ with P-value of

7100.5 P

17102.4

Results

Results

Conclusion

BCCA is able to find a group of genes that show similar pattern of variation in their expression profiles over a subset of measurements.

Better than other biclustering algorithm: Find higher number of common

transcription factors of a set of gene in a bicluster

More functionally enriched