Bayesian Nonparametric Feature Selection Over Large-Scale Gene Networks with Missing Values

• Activity Number: 441
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Zhuxuan Jin* and Zhou Lan and Jian Kang and Tianwei Yu
• Companies: Emory University and North Carolina State University and University of Michigan and Emory University
• Keywords: Dirichlet process mixture ; Swendsen-Wang algorithm ; feature selection ; gene networks

Abstract

Feature selection over large-scale gene networks has become increasingly important. However, most recent works cannot distinguish specific types of effects and lose selection accuracy and introduce bias in estimating gene effects. To address those limitations, we propose a Bayesian nonparametric method for gene and sub-network selection. It can identify important genes with two different behaviors: "down-regulated" and "up-regulated" for which a novel prior model is developed for the selection indicator incorporating the network dependence. In posterior computation, we resort to Bayesian inference algorithm incorporating Swendsen-Wang algorithm for efficiently updating selection indicators. The proposed method can take into account missing data, which improves the selection accuracy and reduces the bias in estimating gene effects. We illustrate our methods on simulation studies and the analysis of the gene expression in primary acute lymphoblastic leukemia (ALL) associated with methotrexate (MTX) treatment.