Fig. 2

The performance in terms of A inter-AUROC and intra-AUROC, B inter-Rs and intra-Rs, and C top1 and top3 success rates based on the threefold clustered cross-validation (CCV) of the PDBbind-ReDocked dataset. A more intuitive comparison of different validation methods can be found in D–F taking NNscore_XGB as an example. The error bars represent the standard deviation of the 3 repetitions (10 repetitions for 4:1 random splitting). The dotted lines in C indicate the ceiling of the success rate, and those in F for 4:1 random splitting, threefold random CV, and threefold CCV are colored in black, purple, and blue, respectively