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

A meta-analysis begins with a well-formulated objective. As highlighted in the introduction, meta-analysis of microarray studies can be used for several purposes, for example to identify DEGs between two or more groups, to identify co-expressed genes, to build gene networks or to evaluate reproducibility and bias across studies. In the following we will focus on meta-analysis for DEGs detection, however most of the considerations apply regardless of the specific topic.
The study selection process is guided by the definition of the inclusion/exclusion criteria. These criteria should be a priori established and should derive immediately from the objective(s) of the study. They can be based on biological (e.g. specific disease, type of outcome, type of tissues, organism) or technical issues (e.g., density of array, minimum number of arrays). A clear, detailed and unambiguous formulation of inclusion/exclusion criteria, possibly in the form of a real protocol, is essential to avoid the most frequent criticism of the meta-analysis, that is "mixing apples and oranges".
Locating the studies is by far the most difficult and the most frustrating aspect of any meta-analysis but it is the most important and critical step. Many meta-analyses begin with a systematic literature search. Keywords concerning the research question and their synonyms are typically used to identify studies for inclusion in the review. In order to retrieve all the relevant studies on a given topic, the search should be as comprehensive as possible, therefore it is recommended to search all the
main electronic databases of abstracts. Reading the latest review articles and contacting specific investigators that are known to be active in the area can help to identify additional studies missed by automated search and ongoing research efforts with unpublished data.

Concerning microarrays, it is appropriate to extend the search to public microarray data repositories, as well as to a few more specialized databases. A quick review of the abstracts and experiments description is essential to eliminate those studies that are clearly not relevant to the meta-analysis or do not meet the specified selection criteria.
After the identification of candidate studies from abstracts, the articles or inherent information from authors, where available, have to be retrieved to confirm their eligibility. To limit the risk of compromising the quality of meta-analysis results, the included studies should undergo a quality assessment, that is an accurate evaluation of the study characteristics in terms of the study design, implementation and analysis. In fact, if a meta-analysis includes many low-quality studies, then the
errors in the primary studies will be carried over to the meta-analysis, where they may be harder to identify, and the obtained result will be biased ("garbage in, garbage out"). Regarding microarray studies, the quality assessment should be performed at the study-level as well as at the data-level, as will be extensively described in the following paragraphs.