Novel gene sets improve set-level classification of prokaryotic gene expression data

BMC Bioinformatics

Table 4 Summary of the main experimental findings

Both the selected types of the newly proposed gene sets (i.e., COPR and REG) perform significantly better than their randomized and gene-level versions. On the contrary, the state-of-the-art gene set type (KEGG+GO) performs indistinguishably from its randomized version and significantly worse than its gene-level version. As detailed in main text, the p-values correspond to the one-sided paired Wilcoxon test applied on the win/tie/loss counts determined by leave-one-out cross-validation of predictive accuracies

ISSN: 1471-2105