Fig. 4

Win-loss statistics to compare AUPRC scores of ECFP diameter 4 to diameters 0, 2, and 6 (for unprocessed fingerprints and 1024 filtered fingerprint fragments). Each bar corresponds to a pairwise comparison of two methods on 76 datasets. Wins of the first/second method are colored in blue/red colors and drawn above/below zero respectively. Intense colors indicate significant wins/losses and are additionally stated in numbers above each bar. Diameter 4 yields in general the most predictive models. Exceptions are naive Bayes and diameter 0 and 2, which is due to the low number of features with low diameter. However, when applying filtering ECFP4 works best for naive Bayes. Moreover, slightly more predictive support vector machines can be build using ECFP6