Incorporate Biological Information into a Hierarchical Model for Bayesian Variable Selection in Biomarker Identification

• Activity Number: 341
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Xiaowei Yang*+ and Bin Peng
• Companies: University of California at Davis and University of California at Davis
• Keywords: Bayesian Variable Selection ; Hierarchical Model ; Microarray Data ; Multiple Comparisons

Abstract

A challenge of statistical analysis in biomarker identification is introduced by the problem of high dimensionality. For example, a microarray analysis in identifying genes for differentiating subtypes of ischemic stroke has to deal with thousands of genes, with peripheral blood samples from only 33 patients (21 with cardioembolic stroke; 12 with large-vessel atherosclerotic strokes). In this talk, we introduce a hierarchical modeling strategy for Bayesian variable selection for the purpose of biomarker identification. This strategy not only allows modeling associational or causal relationships between biomarkers (e.g., gene co-expression and gene co-regulation) via Bayesian networks, but also enables integrating various sources of biological information (e.g., pathway and Gene Oncology information) into prior distributions using graphical tools such as Markov random fields. Using the stroke microarray data and simulations, we illustrate that our method could identify genes with meaningful biological interpretation and better control of false discover rate. The method provides a powerful solution to identify biomarkers from other forms of proteomic and genomics data.