The S100P gene has a relatively favorable ranking in the Pilarsky and Pei datasets and moderate to un-favorable rankings in the other datasets, indicating that analysis of individual datasets may not readily identify the gene. Another example, LAMC2, is ranked favorably in the Ishikawa Ruxolitinib cost and Pei datasets, but relatively higher in the other datasets. Overall, LAMC2 is ranked second in the combined results and is, according the to literature, a purported pancreatic cancer gene [41]. Weighted average ranks for the pancreatic cancer results increase quickly compared to the breast and renal cancer results, indicating increased heterogeneity among the ranks of the individual datasets. One explanation for this is the slight difference in dataset subtype comparisons.

For example, one of the datasets, Ishikawa, extracted RNA samples from pancreatic juice rather than from solid tumors.The degree of differential expression (and consequently, the rank) of a gene can vary significantly from dataset to dataset. Combining DEG detection results by averaging ranks across datasets reduces variability and improves statistical confidence. Analysis of a single microarray dataset may result in errors during DEG detection��for example, false positives and false negatives (genes that should be differentially expressed, but not favorably ranked). In general, these errors can be reduced by increasing sample size. Combining microarray datasets by averaging ranks effectively increases sample size while enabling robust analysis of heterogeneous data.4.

DiscussionIn order to understand the differences in performance among the six meta-analysis-based FS methods, we identify and list the differences and similarities in Table 3. We focus on three properties: (a) basic FS methods forming the basis of meta-analysis, (b) the manner in which these basic FS methods are chosen and applied to individual microarray datasets, and (c) the use of ranks.Table 3Properties of six microarray meta-analysis methods.Among the five meta-analysis methods (not including the naive control method) rank average and mDEDS are the only methods that consider multiple basic FS methods��for example, fold change, t-statistic, SAM, and rank sum��for detecting DEGs (Table 3, row 1). The rank products, Choi and Wang methods use modified forms of basic FS methods.

Moreover, rank average is the only method that chooses one basic FS method for each dataset to maximize prediction performance (Table 3, row 2). In contrast, mDEDS uses all of the available basic FS methods for Dacomitinib each dataset. Finally, rank average and rank products are the only meta-analysis methods that are rank-based (Table 3, row 3).Among the basic FS methods, no method can be considered the best because of the data-dependent nature of microarray analysis.