Statistical mechanical modeling of eukaryotic gene regulation

Gene expression patterns are regulated by gene regulatory networks. Central to transcriptional regulation of gene expression is the regulation of the quantities of transcription factors (TFs) bound to genomic regulatory sequences. This thesis work is built on statistical mechanics to study the stochastic interactions of TFs and regulatory sequences. We present a predictive model to learn how TFs interact with cis-regulatory sequences and with each other. By analyzing large scale TF-DNA binding data, the model can discover cooperative interactions among TFs and predict the strength of TF-DNA binding. Less clear is how the genome and the epigenome jointly instruct TFs binding. We present an epigenome-sensitive model to systematically analyze the epigenomic functions in modulating transcription factor-DNA binding. We discovered preferences of TFs for specific combinations of epigenomic modifications, termed as epigenomic motifs. Epigenomic motifs explain why some TFs appear to have different DNA binding motifs derived from in vivo and in vitro experiments. The data suggest that the epigenome can modulate transcriptional noise and boost the cooperativity of weak TF binding sites. We also show that the epigenome might suppress the TF binding differences on SNP-containing binding sites in two people, in theory and in real data. To identify regulatory relationships between TFs and target genes is another major topic in gene regulation. We developed an analytical method to identify a statistical thermodynamic model that best describes the form of TF-TF interaction among a set of TFs for every target gene. Based on this method, we developed a computational framework to infer regulatory relationships from multiple time course gene expression datasets. RNA interference data and large scale TF-DNA binding data independently validated a statistically significant fraction of these regulatory relationships. Moreover, this framework has the flexibility to incorporate other independent datasets to increase prediction accuracy