Empirical Bayes Analysis for Detecting Differential Expression in Microarrays

Abstract

<p>The purpose of gene expression microarray studies is to identify differentially expressed genes. Due to the very large number of genes compared to the very small sample size, and the possibility of high level of non-normal random noise, traditional hypotheses tests cannot be used directly. In this thesis, we applied parametric and nonparametric empirical Bayes methodologies to test the hypotheses of differential expression in a real microarray data set from a study of Type 1 Diabetes and some other simulated data sets. In our real data, we saw some problems of applying parametric empirical Bayes (in terms of R software called EBarrays; nonparametric empirical Bayes method implemented in the R packaged Siggenes also has problems in detecting differential expression in real data when some extreme patterns show up in the permutation matrix. We implemented a new function called EBayes based on Efron's idea of nonparametric empirical Bayes method. EBayes performs much better than other empirical Bayes methodologies in dealing with real data. Furthermore, the results of simulated data show that the new function EBayes are comparable to Siggenes EBAM function.</p>