Bayesian Inference on Group Differences in Brain Networks

• Activity Number: 619
• Type: Invited
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: IMS
• Author(s): Daniele Durante* and David Dunson
• Companies: University of Padova and Duke University
• Keywords: Bayesian nonparametrics ; Mixture model ; Multiple testing ; Low-rank factorization ; Network data ; Neuroscience

Abstract

Network data are increasingly available along with other variables of interest. Our motivation is drawn from neurophysiology studies measuring a brain activity network for each subject along with a categorical variable, such as presence or absence of a neuropsychiatric disease, creativity groups or type of ability. We develop a Bayesian approach for inferences on group differences in the network structure, allowing global and local hypothesis testing adjusting for multiplicity. Our approach allows the probability mass function for network-valued data to shift nonparametrically between groups, via a dependent mixture of low-rank factorizations. An efficient Gibbs sampler is defined for posterior computation. We provide theoretical results on the flexibility of the model and assess testing performance in simulations. The approach is applied to provide novel results showing relationships between human brain networks and creativity.