A comparison between the results of HighEdgeS, the classical approach, and GAGE as they would be used in practice.

Three pathway analysis methods (ORA, SPIA and LEGO) were applied on the genes yielded by HighEdgeS and the classical approach. GAGE uses the entire set of genes to directly identify pathways so ORA, SPIA and LEGO are not applicable for GAGE. The green background indicates the best results obtained for each dataset (each row). For the classical approach we selected the DE genes using an absolute fold change greater than 2 and FDR-corrected p-value less than 0.05. “No DE” means no genes met those thresholds. The performance was measured using the true positive rate (TPR) and false positive rate (FPR) for each data set. The results show that HighEdgeS yields the best TPR for all datasets. HighEdgeS also yield the best FPR for 2 out of the 3 datasets. Furthermore, for the Pdx1 dataset, even though the classical approach has a lower FPR values, its sensitivity is very low (not being able to identify any true positive (TP) pathway in 2 out of 3 cases, and identifying only one of the 3 true positive pathway in the remaining case). In contrast, HighEdgeS identified all true positive pathways in 2 out of 3 cases (when combined with ORA and SPIA) and 2 out of the 3 TP pathways when combined with LEGO.