Mining Gene Expression Patterns for the Discovery of Overlapping Clusters

Abstract

Many clustering algorithms have been used to identify co-expressed genes in gene expression data. Since proteins typically interact with different groups of proteins in order to serve different biological roles, when responding to different external stimulants, the genes that produce these proteins are expected to co-express with more than one group of genes and therefore belong to more than one cluster. This poses a challenge to existing clustering algorithms as there is a need for overlapping clusters to be discovered in a noisy environment. For this reason, we propose an effective clustering approach, which consists of an initial clustering phase and a second re-clustering phase, in this paper. The proposed approach has several desirable features as follows. It makes use of both local and global information inherent in gene expression data to discover overlapping clusters by computing both a local pairwise similarity measure between gene expression profiles and a global probabilistic measure of interestingness of hidden patterns. When performing re-clustering, the proposed approach is able to distinguish between relevant and irrelevant expression data. In addition, it is able to make explicit the patterns discovered in each cluster for easy interpretation. For performance evaluation, the proposed approach has been tested with both simulated and real expression data sets. Experimental results show that it is able to effectively uncover interesting patterns in noisy gene expression data so that, based on these patterns, overlapping clusters can be discovered and also the expression levels at which each cluster of genes co-expresses under different conditions can be better understood.