Empirical Bayesian Analysis of Variance Component Models for Microarray Data

• Activity Number: 274
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Sheng Feng*+ and Russell D. Wolfinger and Tzi-Ming Chu and Greg Gibson and Lisa McGraw
• Companies: North Carolina State University and SAS Institute Inc. and SAS Institute Inc. and North Carolina State University and Cornell University
• Keywords: empirical Bayes ; variance component model ; mixed model ; microarray data analysis ; shrinkage estimators

Abstract

A gene by gene mixed model analysis (called "single gene analysis") is a useful statistical method for assessing significance for microarray gene differential expression. While a large amount of data are collected in a typical microarray experiment, the sample size for each gene is usually relatively small, which could limit the statistical power of this analysis. We introduce an empirical Bayesian approach for general variance component models applied to microarray data. The power lost because of small sample size is regained by integrating prior information on variance components estimated from all genes. The approach starts with the single gene analysis. The estimated variance components from each gene are transformed to the "ANOVA components." This transformation makes it possible that the prior density of each "ANOVA component" is estimated independently. For every gene, the posterior density of each "ANOVA component" can be easily derived based on previous work. The modes of the posterior distribution are inversely transformed to compute the posterior estimates of the variance components.