Computational Identification of Transcription Factor Binding Sites via a Transcription-factor-centric Clustering (TFCC) Algorithm

While microarray-based expression profiling has facilitated the use of computational methods to find potential cis-regulatory promoter elements, few current in silico approaches explicitly link regulatory motifs with the transcription factors that bind them. We have thus developed a TF-centric clustering (TFCC) algorithm that may provide such missing information through incorporation of biological knowledge about TFs. TFCC is a semi-supervised clustering algorithm which relies on the assumption that the expression profiles of some TFs may be related to those of the genes under their control. We examined this premise and found the vicinities of TFs in expression space are often enriched with the genes they regulate. So, instead of clustering genes based on the mutual similarity of their expression profiles to each other, we used TFs as seeds to group together genes whose expression patterns correlate with that of a particular TF. Then a Gibbs sampling algorithm was applied to search for shared cis-regulatory elements in promoters of clustered genes. Our working hypothesis was that if a TF-centric cluster indeed contains many targets of the seeding TF, at least one of the discovered motifs would be the site bound by the very same TF. We tested the TFCC approach on eight cell cycle and sporulation regulating TFs whose binding sites have been previously characterized in Saccharomyces cerevisiae, and correctly identified binding site motifs for half of them. In addition, we also made de novo predictions for some unknown TF binding sites.