Bayesian Variable Selection for Identifying Genetic Effects on Functional Connectivity

• 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): Brian Reich*+ and Michele Guindani and Abel Rodriguez and Vince Calhoun
• Companies: North Carolina State University and The University of Texas MD Anderson Cancer Center and University of California at Santa Cruz and University of New Mexico
• Keywords: fMRI data ; Bayesian variable selection ; functional connectivity ; Graphical models

Abstract

Functional magnetic resonance imaging (fMRI) data are often used to identify regions of the brain that are functionally connected while performing a cognitive task. Connectivity patterns vary across subjects according to subject-specific characteristics, e.g., the subject having been diagnosed with a form of schizophrenia. Our objective is to identify genetic pathways that affect a subject's functional connectivity in response to a series of external stimuli. We model each subject's connectivity using a graphical model, with potentially a different set of edges for each subject. We assume that the probability of each pair of regions being connected depends on a set of subject-specific genetic covariates. This gives a high-dimensional model, as the number potential region pairs and the number of genetic variables are both large. Therefore, we propose a Bayesian variable selection technique to identify a sparse model for functional connectivity. The approach is illustrated on a a set of genetic and fMRI data from a population of healthy and schizophrenic patients.