A Bayesian Model for Cross-Study Differential Expression

• Activity Number: 224
• Type: Invited
• Date/Time: Tuesday, August 8, 2006 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Andrew Nobel*+ and Haakon Tjelmeland and Rob Scharpf and Giovanni Parmigiani
• Companies: The University of North Carolina at Chapel Hill and Norwegian University of Science and Technology and Johns Hopkins University and Johns Hopkins University
• Keywords: microarray ; cross-study ; Bayesian ; differential expression ; gene expression

Abstract

We propose and analyze a Bayesian method for cross-study analysis of differential expression in two or more gene expression studies. The method is based on a 'joint' hierarchical model of the data in each of the available studies. For every gene, the model includes a binary parameter indicating whether the gene is differentially expressed. Given data, the probability that a gene is differentially expressed is evaluated through the posterior expectation of its associated indicator, which is estimated via MCMC. We apply the method to multiple lung cancer datasets and simulated data derived from the subset of adenocarcinomas. For the simulated data, we compare the posterior mean estimates of differential expression to other cross-study methods using ROC curves and a related measure based on false and missed discovery rates