Clust: automatic extraction of optimal co-expressed gene clusters from gene expression data

Abstract

Identification of co-expressed gene clusters can provide evidence for genetic or physical interactions between genes. Thus, co-expression clustering is a routine step in large-scale analyses of gene expression data. We show that commonly used clustering methods produce results that substantially disagree with each other, and do not match the biological expectations of co-expressed gene clusters. Furthermore, these clusters can contain up to 50% unreliably assigned genes. Consequently, downstream analyses of these clusters (e.g. functional term enrichment analysis) suffer from high error rates. We present clust, an automated method that solves these problems by extracting clusters that match the biological expectations of co-expressed genes. Using 100 datasets from five model organisms we demonstrate that clusters generated by clust are better than those produced by other methods, both numerically and for use in functional analysis. Finally, we show that clust can simultaneously cluster multiple datasets, enabling users to leverage the large quantity of public expression data for novel comparative analysis.