Exploring transcriptional modules through integrative gene clustering guided by transcription factor binding information

Cluster analysis aims to infer regulatory modules or biological functions by grouping the genes with similar patterns. However, clustering results often fail to show strong biological relevance to underlying transcriptional modules. In this paper, we present a computational approach, namely integrative gene clustering guided by motif information, to identify condition-specific transcriptional modules. Specifically, the proposed approach is designed to discover the co-regulated genes from co-expressed genes by integrating ChIP-on-chip binding site (or motif) information and gene expression data. A statistical significance analysis procedure is performed to associate a set of significant motifs with each co-expressed gene cluster. An information-theoretic technique based on the entropy of the motif information is further developed to select an optimal balance point between expression patterns and motif patterns. The experimental results from simulated Saccharomyces cerevisiae data demonstrated that our approach can successfully uncover specific biological processes that are evidently regulated by one or more transcription factors.