A Bayesian Dimension Reduction Approach for Detection of Multilocus Interaction in Case-Control Studies

• Activity Number: 552
• Type: Topic Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Debashree Ray*+ and Xiang Li and Wei Pan and James S. Pankow and Saonli Basu
• Companies: University of Minnesota and University of Minnesota and University of Minnesota and University of Minnesota and University of Minnesota
• Keywords: Dimension reduction ; Multilocus interaction ; Reversible Jump MCMC

Abstract

Genome-wide association studies (GWASs) have identified hundreds of genetic variants associated with complex diseases, but these variants appear to explain very little disease heritability. The typical single locus association analysis in a GWAS fail to detect variants with small effect sizes and to capture higher order interaction among these variants. Multilocus association analysis provides a powerful alternative by jointly modeling the variants in a gene or a pathway and by reducing the burden of multiple hypothesis testing in a GWAS. We proposed a powerful dimension reduction approach to model multilocus association. We used a Bayesian partitioning model to cluster SNPs as per their direction of association, model higher order interactions using a flexible scoring scheme, and use posterior marginal probabilities to detect association between the SNP-set and the disease. Extensive simulation studies showed that our approach has better power to detect multilocus interactions than several existing methods. When applied to ARIC dataset to study gene based associations for type 2 diabetes, our method identified some novel variants not detected by conventional single locus analyses.