A0266
Title: A variational empirical Bayes approach to multivariate multiple regression, with applications to polygenic prediction
Authors: Fabio Morgante - Clemson University (United States) [presenting]
Peter Carbonetto - University of Chicago (United States)
Gao Wang - Columbia University (United States)
Yuxin Zou - Regeneron Genetics Center (United States)
Abhishek Sarkar - Vesalius Therapeutics (United States)
Matthew Stephens - University of Chicago (United States)
Abstract: Multivariate (i.e., multi-outcome) multiple regression has been an important tool in different fields of applied statistics. One of these fields is quantitative genetics, where the aim is to accurately predict complex trait phenotypes from genotypes. Multivariate multiple regression can be used to predict multiple correlated phenotypes jointly from genotypes, leveraging the shared genetic effects across such phenotypes and improving accuracy over univariate analyses. However, effects can be shared across phenotypes in a variety of ways, so computationally efficient statistical methods are needed that can accurately and flexibly capture patterns of effect sharing. New Bayesian multivariate multiple regression methods are described as using flexible priors learned from the data, which are able to model many different patterns of effect sharing and specificity across outcomes. The methods are evaluated in their ability to predict multiple phenotypes from genotypes using simulations with different patterns of effect sharing across phenotypes as well as real data applications. The results show that these new methods can provide more accurate predictions than existing univariate and multivariate methods while also being computationally efficient.
