An FDR-Controlling Procedure for Analyzing Replicated Microarray Time Course Data with the Multivariate Empirical Bayes Statistic

• Activity Number: 146
• Type: Contributed
• Date/Time: Monday, August 4, 2008 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Nancy N. Wang*+ and Terence P. Speed
• Companies: University of California, Berkeley and University of California, Berkeley
• Keywords: gene expression ; longitudinal data ; Hotelling T^2 ; moderation ; shrinkage

Abstract

A multivariate empirical Bayes statistic proposed by Tai and Speed has been shown useful for ranking genes based on differential expression in replicated microarray time course experiments. This is known as the MB-statistic, or the moderated Hotelling T^2 statistic when all genes share a common sample size. We propose an FDR-controlling procedure for this statistic. A null distribution is obtained using the parametric bootstrap. Critical values are determined according to the empirical Bayes FDR procedure. We compared this method, through simulations, to the frequentist FDR procedure, which requires a theoretical null distribution for calculating the nominal p-values. Although our method is slightly anti-conservative; it is more robust to the variability in the estimates of the hyper-parameters, when the degree of moderation is small.