ABSTRACT Genome-wide association studies (GWASs) provide no mechanism of disease. Moving from GWAS to mechanism requires priori knowledge of cis-regulatory elements (CREs), as most risk variants reside in noncoding regions. Therefore, methods for defining noncoding networks consisting of CREs and their gene targets are essential. Current methods for CRE prediction are based on histone marks, transcription factor binding, and evolutionary sequence conservation. These methods have high sensitivity, but poor specificity. For example, over 6k enhancers were reported in human heart. Subsequent methods to predict CRE targets inherit poor specificity. Here, we propose new strategy to precisely predict risk CREs and target genes from noncoding and coding transcriptomes. Emerging evidence suggests that functional CREs are themselves transcribed. We posit that tissue-specific expression in CRE transcripts can quantitatively define prioritize functional CREs from candidate loci. We posit that a CRE transcript and its gene targets coordinately expressed, providing a quantitative measure of CRE-target gene activity. We will utilize CRE transcription to prioritize CREs harboring candidate functional genetic variation. To this end, we need to systematically mine multi-scale high-throughput data, which requires efficient bioinformatics methods. In this proposal, we will develop a non-coding transcriptome model. First, we will integrate haplotype blocks of GWAS findings, disease-dependent noncoding expression, and distal chromatin interaction into a computational prioritization of CRE-promoter pairs. We will introduce our ?soft threshold?-algorithm to assess chromatin accessible CREs together with their transcribed ncRNAs. Second, building on our early success of PGnet algorithm, we will build a tripartite network (disease traits, CRE variants, and target genes) with a nonparametric model to infer risk CREs and gene targets. We will develop novel gene-based association and compared to PrediXscan. We will compare our prediction with TargetFinder. Collaborated with an expert in cardiac conduction system, I select atrial fibrillation (AF) as a research platform. AF is the most common human arrhythmias, affecting over 33 million people worldwide. The deliverables of this project will include critical cardiac rhythm CREs, CRE variants, and target genes. The ultimate goal of this work is an evaluated computational model to set the stage for future functional evaluations. This new analytic suit to define AF-associated noncoding regulatory pathway will have border implication on other diseases.

R21LM012619

2017-09-07

2019-08-31