Weighted Meta-Analysis for Small Sample Microarray Studies

Abstract

An abundance of data from microarray studies are available in publicly-accessible databases. Most of these studies are conducted by university based research labs. It is not uncommon for such studies to run only three or four replicates for each experimental condition tested. With this low sample size and the high variability and multiple testing problems inherent to microarray technology, it is difficult to draw statistically significant conclusions from any one such study. Meta-analysis could improve this situation by combining evidence from related studies to increase statistical power. In this work we discussed several meta-analysis methods for small sample gene expression studies. We compared the performances of the traditional Fisher's log-sum and Stouffer's-Z meta-analysis methods, as well as three weighted variants of Stouffer's method. Higher false discovery rates were observed for the traditional methods compared to the weighted methods.