A broad goal in genome research is to better understand transcriptional regulation. We propose to complement current experimental efforts towards this goal through computational approaches. Gene transcription is regulated by a network of transcription factors (TF); to accomplish their task, TFs bind to specific DNA elements in the relative vicinity of the gene, interact with each other and with polymerase. The overall task of transcriptional control is divided among smaller groups of closely functioning TFs, or transcriptional modules, such that each module regulates transcription in response to specific stimuli or environmental condition. This division of control provides a modular mechanism to co-regulate groups of functionally related genes. The combinatorial interactions among TFs and the DNA elements that facilitate these interactions motivate the algorithmic approaches to study transcriptional regulation. We will (1) develop an EM approach to simultaneously detect transcription factor binding motifs and their interacting partners from genome-wide ChIP experiments; we will apply this to genome-wide yeast ChlP-chip data and to genome-wide CREB binding data in rat, (2) develop novel graph-theoretic approaches to detect transcriptional modules, and apply the methods to detect modules driving transcription in two biological processes - long term memory storage and heart failure - in collaboration with experimentalists, (3) develop novel Gibbs sampling approach to detect dense sub-graphs in a multi-partite graph as a means to identify modules, and apply this to detect modules driving tissue-specificity in human. ? ? ?
