Table 7.

Summary of the performance of similarity search methods on 40 DUD^a and 102 DUDE^b datasets.

Metric	Mean Max-Sim valuec	Mean%diffd	Diff ≥ 0%e	Diff < 0%f	Diff ≥ 1%g	Diff < −1%h
MaxZ vs. Max-Sim method on 40 DUD datasets
ROC_AUC	0.91	1.0	30	10	13	0
BEDROC	0.79	1.7	30	10	16	1
AS vs. Max-Sim method on 40 DUD datasets
ROC_AUC	0.90	1.2	28	12	15	3
BEDROC	0.76	1.1	25	15	16	10
MaxZ vs. Max-Sim method on 102 DUDE datasets
ROC_AUC	0.96	0.5	90	12	18	0
BEDROC	0.90	0.7	87	15	28	2
AS vs. Max-Sim method on 102 DUD datasets
ROC_AUC	0.96	0.2	55	47	13	3
BEDROC	0.90	0.2	58	44	27	20

^aDUD: Directory of Useful Decoys, http://dud.docking.org/.

^bDUDE: Database of Useful Decoys: Enhanced, http://dude.docking.org/.

^cMean ROC_AUC or BEDROC value calculated from the Max-Sim method over 40 DUD or 102 DUDE datasets.

^dMean percentage difference between ROC_AUC or BEDROC values derived from the maxZ or AS methods and the Max-Sim method.

^eNumber of datasets for which ROC_AUC or BEDROC values calculated from the maxZ or AS methods were higher than or equal to the corresponding values calculated from the Max-Sim method, i.e., the number of datasets on which the maxZ or AS method performed comparable to or better than the Max-Sim method did.

^fNumber of datasets for which ROC_AUC or BEDROC values calculated from the maxZ or AS methods were lower than the corresponding values calculated from the Max-Sim method, i.e., the number of datasets on which the maxZ or AS method performed worse than the Max-Sim method.

^gNumber of datasets for which ROC_AUC or BEDROC values calculated from the maxZ or AS methods were at least 1% higher than the corresponding values calculated from the Max-Sim method.

^hNumber of datasets for which ROC_AUC or BEDROC values calculated from the Max-Sim method was more than 1% higher than the corresponding values calculated from the maxZ or AS methods.