Dissecting the gene-regulatory circuitry of disease-associated genetic variants

Abstract

Disease-associated nucleotides lie primarily in non-coding regions, increasing the urgency of understanding how gene-regulatory circuitry impacts human disease. Here, we use the increasing availability of functional genomics datasets and models elucidating how regulatory proteins control genes, to evaluate the impact of genetic variants on the activity of diverse regulators. First, we generate a comprehensive compendium of predicted binding intensities across the entire genome for over 500 transcription factors. Second, we create a novel dataset to connect how these binding intensities change in the context of disease datasets. Third, we develop a statistical framework to integrate these two datasets using dimensionality reduction, latent cluster discovery, and topic modeling. We use these techniques to show that regulatory proteins with analogous biological functions share similar global changes in binding due to genome-wide genetic variation. We also use our framework to discover a latent set of topics behind all genomic locations in chromosome 1, to link the locations in each of the topic clusters with a class of related diseases, and to show that relevant biological processes are statistically enriched in the genomic locations most related to each cluster.