Bayesian Mixture Models for Case-Control, Genome-Wide Association Studies

• Activity Number: 136
• Type: Contributed
• Date/Time: Monday, August 4, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lin Li*+ and Andrew G. Clark and Carlos D. Bustamante
• Companies: Cornell University and Cornell University and Cornell University
• Keywords: genome-wide association study ; QTL mapping ; dimension reduction ; case-control study

Abstract

In this paper, we develop Bayesian mixture models for dimension reduction in case-control genome-wide association studies. The method proposes incorporating the prior information that most genotypic markers have a negligible probability of being quantitative trait loci (QTL). Both the logistic model and the probit model are considered for multiple QTL effects on disease susceptibility, and Markov chain Monte Carlo (MCMC) techniques are developed for inference. A novel method called Score Averaging Method (SAM) is also proposed for accelerating dimension reduction. The approaches are applied to simulated data of case-control association studies from real-life genotype data, and our results suggest that the method is nearly always more powerful than single marker analyses in identifying multiple QTL.