Table 3.

Number of correct sites identified.

Method	# protein-peptide interaction sites (maximum recall)	# protein-protein interaction sites
InterPep*	255 (50.7%)	32 (14.2%)
PeptiMap	230 (45.8%)	39 (17.3%)
GalaxyPepDock*	218 (43.4%)	24 (10.6%)
PepSite2**	176 (35.0%)	54 (23.9%)
InterPep-TMonly*	160 (31.9%)	48 (21.2%)
Surf-Cons	118 (23.5%)	32 (14.2%)
Random	54 (10.8%) σ = 5.9	20 (8.8%) σ = 2.6
Total sites	502 (100.0%)	226 (100.0%)

The number of target pairs where the correct peptide-binding site was identified in the top scorer (cluster for InterPep) for the different methods. #protein-peptide interaction sites measures the number of predictions that hit a protein-peptide interaction site, while #protein-protein interaction sites measures the number of predictions that hit a protein-protein interaction site that is an interaction but not the correct protein-peptide interaction site. Random is what would be expected by chance, and was generated by randomly predicting n number of residues on the surface of the protein as interacting, where n is the number of interacting residues in the native structure. Total is the total number of sites in the test set. *For template methods, no templates matching the target with BLAST E < 0.001 were used. **To make sure PepSite2 was not disadvantaged for longer peptides, it was run with a sliding window of size 10 for the whole peptide and the best prediction was chosen.