Nonparametric Bayesian Data Analysis: Inference for Differential Gene Expression

• Activity Number: 258
• Type: Invited
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: WNAR
• Author(s): Peter Mueller*+
• Companies: University of Texas MD Anderson Cancer Center
• Keywords: density estimation ; microarray ; mixture models

Abstract

We review methods and models used in nonparametric Bayesian inference, using as a motivating example the analysis of microarray data for differential gene expression. The discussion will highlight advantages and limitations of inference based on full (posterior) probability models for the relevant unknown distributions. We will compare resulting inference with ad-hoc exploratory data analysis, with a similar parametric empirical Bayes approach proposed in Efron et al. (2001), and with fully parametric finite mixture models. We will discuss problems of the empirical Bayes inference that are mitigated by a nonparametric Bayesian approach. As a specific example of nonparametric Bayesian modeling we will focus on an extension of Dirichlet process mixtures of normal distributions. As in most nonparametric Bayesian inference, efficient computation is key to a successful implementation. We will review computational problems and strategies that are relevant for a wide class of nonparametric Bayesian models.