Meta-analytic approaches for microarray data

It is becoming increasingly common for multiple laboratories to use microarray technology to study the genetic basis of the same disease or condition of interest in the same organism. With this technology, each laboratory seeks to identify genes that are differentially expressed between conditions. Differences in experimental results can arise from chance variation, as well as fundamental differences between laboratories. Furthermore, estimates of each gene\u27s magnitude of differential expression from multiple experiments in the same laboratory may not be independent. To achieve a clearer understanding of each gene\u27s true relationship to the condition of interest, these differences and dependencies need to be accounted for when combining results from different laboratories. A meta-analytic approach for combining results from the Affymetrix platform is developed, focusing on the use of covariate and covariance information. Fixed effects, random effects, and hierarchical Bayes frameworks are presented. The traditional univariate Affymetrix approach for quantifying differential expression via the signal log ratio is extended to the multivariate case to allow for covariance estimation. This novel approach is evaluated using data from a simple simulation model and experimental data from a mouse model for multiple sclerosis. The performance of this approach is compared with alternative methods for estimating magnitudes of differential expression and other previously proposed approaches to combine results from multiple microarray experiments