Functional characterization of genetic variation with in silico predictions of cell-type-specific regulatory elements

Abstract

Genome-wide association studies (GWAS) have implicated thousands of complex trait-variant associations, an estimated 90% of which reside in the noncoding genome. While noncoding variants generally have poorly understood regulatory function, previous work has shown that disease-driving genetic variation often affects cell-type-specific gene regulation, such as transcription factor (TF) binding. However, maps of TF-mediated cell-type-specific regulation are currently incomplete due to limited amounts of experimental data. In this thesis, I introduce a novel strategy to annotate the noncoding genome with cell-type-specific regulatory element probabilities via integration and modeling of thousands of publicly available epigenetic datasets. I show that these functional annotations in the disease-driving cell type are more highly enriched for disease heritability than experimentally derived functional annotations. Next, I use these functional annotations to prioritize disease-relevant variants in the context of polygenic risk score (PRS) models. I show that this approach improves the trans-ethnic portability of PRS by reducing the confounding effects of population-specific linkage disequilibrium. Lastly, I introduce a novel strategy to leverage the unprecedented resolution of single cell data to elucidate cell-state-specific activity of trait-driving variants identified by polygenic fine-mapping data from GWAS. This strategy consists of calculating cell-specific enrichments of genome-wide genetic variation in functional regions and then associating these enrichments with polygenic regulatory programs. I show that this approach identifies heterogeneity of risk variant accessibility, nominating putatively causal cell states and regulatory mechanisms. Altogether, this work demonstrates the importance of comprehensive functional annotations to better understand disease and trait etiology.