Imaging Genetics via Bayesian Variable Selection

• Activity Number: 637
• Type: Invited
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: International Society of Bayesian Analysis
• Author(s): Fan Li*+ and Tingting Zhang
• Companies: Duke University and University of Virginia
• Keywords: Bayesian variable selection ; fMRI ; genetics ; hierarchical models ; SNP

Abstract

A central goal in imaging genetics is to infer the population-based relationship between neuroimaging phenotypes such as the functional Magnetic Resonance Imaging (fMRI) time series and human genotypes such as the single-nucleotide polymorphisms (SNP). Despite the enormous research conducted in the separate fields of fMRI data analysis and genome-wide association studies (GWAS), relatively little work has been done on combining these two types of data. Built upon the General Linear Model (GLM) framework, we propose a Bayesian hierarchical model that allows for simultaneous inference of subject-specific brain activities and selection of important genotypes. Model sparsity is achieved by imposing the Bayesian variable selection priors on both phenotypes and genotypes. Spatial information among voxels is also incorporated via the Ising prior. We apply the method to analyze the large scale Duke Neurogenetics Study.