Discriminative Learning of Composite Transcriptional Regulatory Modules

A central goal of molecular biology is to uncover transcription regulation mechanisms that govern gene expression. Transcription factors play an important role in those mechanisms, as they affect the transcription rates of genes. Often, such regulatory circuits involve not only one transcription factor but rather several factors that act in concert to modulate the transcription of genes. The recent advances in high-throughput assays, such as microarray experiments and Chromatin Immunoprecipitation, allow us to infer groups of genes that are co-expressed or coregulated. The challenge is to use this wealth of information to gain insights about transcriptional regulation. In this dissertation, we present a procedure for locating regulatory complexes in promoter regions. A regulatory complex represents the binding sites of a pair of transcription factors that act in cooperation. Our procedure takes a discriminative approach, searching for regulatory complexes that are overabundant in the promoter regions of the target group of co-expressed genes and are infrequent in the control group of genes outside the target group. By doing this, we filter out phenomena that are shared among both groups, ideally leaving us with the core motifs. We demonstrate the applicability of our method for finding regulatory complexes in a genome-wide analysis of the yeast genome.