Modeling Functional Regulation of Gene Expression from Chromatin Profiles

Chromatin regulation of gene expression plays a critical role in many biological processes including cancer formation and progression. Several experimental techniques including ChIP-seq and DNase/ATAC-seq have been developed to identify genome-wide chromatin profiles. Using chromatin profiles to predict functional DNA elements and transcription factors (TFs) regulating gene expression is an important problem. Here we present Binding Analysis for Regulation of Transcription (BART), a computational method for predicting functional TFs that regulate any given gene set. Using a genomic cis-regulatory profile predicted by MARGE, a regression and semi-supervised learning-based approach for predicting cis-regulatory profiles from a given gene set, BART predicts TFs whose genomic binding profiles are best associated with the cis-regulatory profile, leveraging thousands of publicly available TF ChIP-seq datasets. We show that BART can accurately predict functional TFs from their target genes in several cancer cell systems. Our work demonstrates the power of computational modeling and utilization of public data for quantitative studies in biological systems.