Bayesian Bridge Regression for Genetic Association Studies

• Activity Number: 139
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Himel Mallick*+ and Nengjun Yi
• Companies: University of Alabama at Birmingham and University of Alabama at Birmingham
• Keywords: Bridge ; Genetic Association Studies ; Penalized Regression ; Lasso ; Elastic Net ; Adaptive Lasso

Abstract

We propose a novel variable selection tool viz. Bayesian bridge regression (BBR), which is essentially a Bayesian version of the classical bridge regression model. To mimic the properties of bridge regression, we treat the coefficients of the bridge regression model as generalized Gaussian (GG) priors. We obtain the posterior mode estimates of the coefficients by a fast and efficient Expectation Conditional Maximization (ECM) algorithm, which can simultaneously fit and estimate all possible genetic effects and covariates in a genetic association study. In simulations, BBR outperformed classical bridge regression in model selection performance with equivalent empirical power, comparable computational time and significantly better control of Type-I error for both continuous and binary traits. Moreover, BBR showed equivalent performance when compared to lasso, adaptive lasso and elastic net. The method is illustrated by re-analyzing the well-known data from the Dallas Heart Study (DHS). Our finding supports previous results about triglyceride-related variants in the Dallas population.