Bayesian Integrative Tensor Models for Genetic Interaction Networks

• Activity Number: 16
• Type: Topic Contributed
• Date/Time: Sunday, August 3, 2014 : 2:00 PM to 3:50 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Chuanhua Xing*+ and Tsuyoshi Kunihama and David Dunson
• Companies: Boston University/AstraZeneca - MedImmune and Duke University and Duke University
• Keywords: Nonparametric Bayes ; Tensor factorization ; High order interactions and dependence ; Conditional mutual information ; Genetic interaction networks ; Gene environment interaction

Abstract

High-order gene-gene and gene-environmental interactions play an important role in the development of complex diseases, while most studies identify only main effects. We propose Bayesian integrative tensor (BIT) models, which can capture such interactions, motivated by a gene-environment interaction study of dietary fat and bone density. Relative to other approaches, such as directly parameterizing interactions in regression models, our proposed approach has considerable advantages in scaling up to analysis simultaneously considering thousands of single nucleotide polymorphisms and environmental factors, while limiting false positives. The proposed model extends a previous probabilistic factorization for categorical data to accommodate mixed scale variables, while developing conditional mutual information-based methods for identifying significant interaction networks. The methods are assessed in simulation studies, and used to analyze our motivating dietary fat and bone mineral density data.