Bi-clustering of Gene Expression Data Using Conditional Entropy

The inherent sparseness of gene expression data and the rare exhibition of similar expression patterns across a wide range of conditions make traditional clustering techniques unsuitable for gene expression analysis. Biclustering methods currently used to identify correlated gene patterns based on a subset of conditions do not effectively mine constant, coherent, or overlapping biclusters, partially because they perform poorly in the presence of noise. In this paper, we present a new methodology (BiEntropy) that combines information entropy and graph theory techniques to identify co-expressed gene patterns that are relevant to a subset of the sample. Our goal is to discover different types of biclusters in the presence of noise and to demonstrate the Superiority of our method over existing methods in terms of discovering functionally enriched biclusters. We demonstrate the effectiveness of our method using both synthetic and real data.