- This event has passed.
PQG Working Group
April 4 @ 1:00 pm - 2:00 pm
University of Chicago
Integrating functional annotations to estimate variant effect sizes from GWAS
The distribution of causal variants– both their distribution throughout the genome, as well as the magnitude of their effects– is a fundamental quantity that underpins many key questions in genetics. Knowledge of this distribution can be used to regularize estimated effects– providing a principled trade off between noisy experimental observations and an informative prior. While the distribution of causal effects is not known a-priori, there is an opportunity to learn this distribution by aggregating information in a genome-wide analysis. Adaptive shrinkage (ASH) is a flexible Empirical Bayes approach to estimate the distribution of genetic effects. However, it assumes that variants are exchangeable. It is well understood that different regions of the genome have varying regulatory potential, that is, some genomic regions are more likely to contain causal varaints of variable effect size. We propose to extend the adaptive shrinkage framework to leverage functional annotations capturing this regulatory potential. Specifically, we propose to model the distribution of effects as a discrete scale-mixture of normals where the mixture weights vary as a function of annotations. Because the relevant annotations may not be known a-priori, we focus on an approach that can select relevant annotations from a large number of correlated annotations. This framework simultaneously provides an interpretable summary of genome wide association signal by identifying relevant annotations, and produces refined effect estimates.