conducted a large-scale study of gene expression by examining the variation of genes across multiple microarray datasets, regardless http://www.selleckchem.com/products/mek162.html of the clinical focus [12]. Breitling and Herzyk ranked fold changes between all interclass pairs of samples and computed the product of all ranks for each gene [13]. More recently, Campain and Yang reviewed several meta-analysis methods and assessed their performance using both classification accuracy and synthetic data [14]. Research has shifted towards methods that consider multiple FS methods, reflecting the fact that no single FS method performs well for all datasets [15]. Although several meta-analysis methods exist, except for the study by Campain and Yang, the literature rarely compares these methods in a comprehensive manner.

We develop the rank average method, a simple meta-analysis-based FS method, for identifying DEGs from multiple microarray datasets and design a study (Figure 1) to compare rank average to five other meta-analysis-based FS methods. We focus on the predictive ability of genes emerging from meta-analysis and show that rank average meta-analysis is robust with respect to three factors. These three factors are (1) clinical application (i.e., breast, renal, and pancreatic cancer diagnosis or subtyping), (2) data platform heterogeneity (i.e., combining different microarray platforms), and (3) classifier. Using a comprehensive factorial analysis, we rate each meta-analysis-based FS method relative to its peers.

In terms of identifying genetic features with reproducible predictive performance and in terms of robustness to multiple factors, results indicate that rank average meta-analysis performs consistently well in comparison to five other meta-analysis-based FS methods.Figure 1Study design diagram. We compare the predictive performance of meta-analysis-based feature selection (FS) methods by designing a study that considers five components: Brefeldin_A (1) basic FS methods that are the building blocks of some of the meta-analysis methods, …2. Methods2.1. Microarray DatasetsWe use six breast cancer, five renal cancer, and five pancreatic cancer gene expression datasets (Table 1) to compare meta-analysis-based FS methods. Each renal cancer dataset examines patient samples from several subtypes of tumors: clear cell (CC), oncocytoma (ONC), chromophobe (CHR), and papillary (PAP). We are interested in identifying genes differentially expressed between the CC subtype and all other subtypes, that is, CC versus ONC/CHR/PAP. These renal cancer datasets share a similar clinical focus. However, they are heterogeneous in terms of microarray platform [16�C21]. Similarly, the breast cancer datasets are heterogeneous in both platform and clinical focus [22�C26].