Title: Empirical Bayes for Microarrays
Speaker: Dr. Christina Kendziorski, Department of Biostatistics and Medical Informatics, University of Wisconsin - Madison
Place: LAEB 2280; Tuesday, 4:30pm

Abstract

Empirical Bayes hierarchical models provide a flexible approach to the analysis of gene expression data and can be used to address questions common to microarray experiments. Such questions include the identification of expressed genes and estimation of true intensity, the estimation of differential expression, the identification of genes which exhibit significant differential expression, and the classification of genes based on one or more features of interest in a given experiment. A modeling framework to address each of these questions will be presented, although I will focus in detail on the identification and classification of differentially expressed genes. Two parametric formulations have proven effective in this effort: the Gamma-Gamma-Multinomial and the LogNormal-Normal-Multinomial. Each characterize fluctuations in both gene-specific expected expression and fluctuations in measured expression given the underlying means. Significant gene expression changes are identified in each formulation by deriving the posterior odds of change. The utility of this approach, the relationship between the posterior odds and other measures to assess differential expression, and the effect of data pre-processing will be discussed.