A comparison of meta-analysis methods for detecting differentially expressed genes in microarray experiments: An application to malignant pleural mesothelioma data.

The proliferation of microarray experiments and the increasing availability of relevant amount of data in public repositories have created a need for meta-analysis methods to efficiently integrate and validate microarray results from independent but related studies.
Despite its increasing popularity, meta-analysis of microarray data is not without problems. In fact, although it shares many features with traditional meta-analysis, most classical meta-analysis methods cannot be directly applied to microarray experiments because of their unique issues.
Several meta-analysis techniques have been proposed in the context of microarrays. However, only recently a comprehensive framework to carry out microarray data meta-analysis has been proposed. Moreover very few software packages for microarray meta-analysis implementation exist and most of them either have unclear manuals or are not easy to apply.
We applied four meta-analysis methods, the Stouffer’s method, the moderated effect size combination approach, the t-based hierarchical modeling and the rank product method, to a set of three microarray studies on malignant pleural mesothelioma. We focused on differential expression analysis between normal and malignant mesothelioma pleural tissues. Both unfiltered and filtered data were analyzed. The lists of differentially expressed genes provided by each method for either kind of data were compared, also by pathway analysis. These comparisons highlighted a poor overlap between the lists of differentially expressed genes and the related pathways obtained using the unfiltered data. Conversely, a higher concordance of the results, both at the gene and the pathway level, was observed when filtered data were considered. The fact that a significant number of genes were identified by only one of the tested methods shows that the gene ranking is based on different perspectives. In fact, the analyzed methods are based on different assumptions and focus on diverse aspects in selecting significant genes. Since so far there is no consensus on what is (are) the ‘best’ meta-analysis method(s), it may be useful to select candidate genes for further analysis using a combination of different meta-analysis methods. In particular, differentially expressed genes detected by more than one method may be considered as the most reliable ones while genes identified by only a single method may be further explored to expand the knowledge of the biological phenomenon of interest.