Fig. 3

Rank-product analysis of MA and BER methods. Methods are ranked according to their performances, separately for each combination of data compendium (E. coli and Yeast), correlation measure (Pearson and Spearman) and performance metric (AUC, pAUC, AUPRC, AUFDR), for a total of 16 different ranks. These ranks are then combined using the Rank-Product method, and the statistical significance of the ranks are evaluated with the method reported in [75]. The negative logarithm of the Rank-Product score is reported on the y-axis, while methods are listed on the x-axis. Triangular markers indicate BER methods, round markers MA methods, square markers baseline approaches. The color of each marker is directly proportional to the Coefficient of Variation (CV) of the respective log-transformed rank-product score (lighter color corresponds to higher variability). Methods that tend to be consistently ranked in the top positions are placed on the top-right of the plots, while poorly performing methods remain the in the bottom-left corner. The plot on the top report the global, final rank of both MA and BER methods, while the two plots on the bottom focus on BER and MA methods, respectively