February 17 2016

2:00 pm Boyer 159

This seminar is sponsored by Human Genetics, Institute for Quantitative and Computational Biosciences, and MIMG

Alexander Gusev
Harvard School of Public Health

Methods for integrating functional and genomic data to understand complex disease


Over the past decade, large-scale genome-wide association studies \(GWAS\) have successfully identified thousands of associations, revealing a fundamental dilemma: the vast majority of these associations lie in non-coding regions and do not clearly implicate a relevant gene. Leveraging recent advances in high-throughput sequencing and functional assays, I will present approaches that connect GWAS data to specific biological mechanisms. First, I will demonstrate how large-scale heritability partitioning can be used to perform in silico biological experiments and assess the total contribution of functional categories to disease even when individually significant associations are scarce. Applying this method to many complex diseases revealed the majority of heritability to lie in or near regulatory regions marked by DNase hypersensitivity sites. Focusing on genomic data from a single disease - prostate cancer - this approach localized heritability from \>\;500 putative regulatory categories to a handful of novel and highly-relevant functional features in binding sites and enhancers from tumor tissue. Next, I will present a novel method for integrating gene expression and GWAS summary data to discover novel genes whose cis-regulated expression is associated with disease and empowering transcriptome-wide association studies \(TWAS\). In real data, TWAS was substantially more effective than current strategies both for fine-mapping specific genes at known loci, and for identifying novel gene-trait associations. These research areas separately infer relationships between disease and either epigenome or transcriptome data. I will conclude by presenting new directions aimed at unifying these two mechanisms and establishing causality.











































































































































































































































































































































































































































































































































































































this is idtest: