Abstract

We review the use of semiparametric mixture models for Bayesian inference in high-throughput genomic data. We discuss three specific approaches for microarray data, for protein mass spectrometry experiments, and for serial analysis of gene expression (SAGE) data. For the microarray data and the protein mass spectrometry we assume group comparison experiments, that is, experiments that seek to identify genes and proteins that are differentially expressed across two biologic conditions of interest. For the SAGE data example we consider inference for a single biologic sample. For all three applications we use flexible mixture models to implement inference. For the microarray data we define a Dirichlet process mixture of normal model. For the mass spectrometry data we introduce a mixture of Beta model. The proposed inference for SAGE data is based on a semiparametric mixture of Poisson distributions.

Introduction

We discuss semiparametric Bayesian data analysis for high-throughput genomic data. We introduce suitable semiparametric mixture models to implement inference for microarray data, mass spectrometry data, and SAGE data. The proposed models include a Dirichlet process mixture of normals for microarray data, a mixture of Beta distributions with a random number of terms for mass spectrometry data, and a Dirichlet process mixture of Poisson model for SAGE data. For the microarray data and the protein mass spectrometry data we consider experiments that compare two biologic conditions of interest. We assume that the aim of the experiment is to find genes and proteins, respectively, that are differentially expressed under the two conditions. For the SAGE example, we propose data analysis for a single biologic sample.