Poisson Mixture Models for Next-Generation RNA Sequencing Data

• Activity Number: 694
• Type: Contributed
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Qiwei Li*+ and Marina Vannucci and Heng Lian
• Companies: Rice University and Rice University and Nanyang Technological University
• Keywords: Bayesian clustering ; Bayesian variable selection ; Dirichlet process ; Markov chain Monte Carlo ; Poisson mixture model

Abstract

Second generation RNA sequencing, a powerful alternative to microarrays for measuring gene expression, has emerged in recent years. This data involve nonnegative counts and it is more appropriately modeled using a mixture Poisson or negative binomial distribution. A main goal in the analysis of such data with substantially smaller sample size relative to the number of variables is to uncover the group structure of the observations and identify the discriminating variables. We propose a Bayesian method for addressing these problems simultaneously. We formulate the clustering problem in terms of a hierarchical mixture model with an unknown number of components and introduce binary latent indicators for variable selection. We perform inference via MCMC methods and demonstrate performances on simulated and real data.