Negative Correlation Aided Network Module Extraction

In this paper, we propose a method to construct an unweighted co-expression network that considers both positive and negative correlation among gene expressions. A measure named NCNMRS(Negative Correlation aided Normalized Mean Residue Similarity) is introduced. The measure can detect both of these correlations and it is used to determine whether a pair of genes are highly correlated either in terms of positive correlation or negative correlation. A greedy technique is also proposed to extract modules from unweighted network. The technique picks a pair of genes with next highest NCNMRS score at a time such that none of the genes in the pair has been included in any network module extracted so far and extends this partial module to a complete network module including genes with high connectivity into the partial module. The technique was applied on a number of real life gene expression datasets and the results have high biological relevance.