Identification of hot-spot residues in protein-protein interactions by computational docking

BMC Bioinformatics

Table 3 Benchmarking NIP hot-spot predictions on different datasets

	*PPV*	S
Initial dataset (21 cases)^a
NIP ≥ 0.2	68%	43%
NIP ≥ 0.4	78%	24%
FOLDEF	73%	46%
ROBETTA^b	71%	69%
Li's dataset (15 cases)^c
NIP ≥ 0.2	75%	42%
NIP ≥ 0.4	91%	26%
FOLDEF^d	70%	45%
ROBETTA^d	64%	60%
Additional dataset (all 22 cases)^e
NIP ≥ 0.2	59%	34%
NIP ≥ 0.4	78%	15%
Additional dataset (X-ray subunits)^e
NIP ≥ 0.2	73%	44%
NIP ≥ 0.4	80%	19%
Additional dataset (NMR subunits)^e
NIP ≥ 0.2	0%	0%
NIP ≥ 0.4	0%	0%
Additional dataset (modeled subunits)^e
NIP ≥ 0.2	59%	33%
NIP ≥ 0.4	75%	15%

Comparison to ROBETTA and FOLDEF
^a Our initial dataset (Table 1); ^b Values from Kortemme and Baker,¹⁴ on a sub-set of 19 complexes from our initial dataset; ^c Dataset compiled by Li et al.;^{28 d} Values from Li et al.;^{28 e} Our additional dataset (Table 6).

ISSN: 1471-2105