. Author manuscript; available in PMC: 2013 Jan 23.

Published in final edited form as: J Chem Inf Model. 2011 Dec 14;52(1):16–28. doi: 10.1021/ci2002507

Table 2.

Statistics of target-specific pose filters.

Targets	Training set			Num. models with CV accuracy ≥ 0.9^b	Test set
Targets	Num. native-like poses	Num. pose decoys	CV accuracy^a		Num. native-like poses	Num. pose decoys	Prediction accuracy
ace	48	49	0.93		99	13	12	0.89
ache	437	363	0.96	184	104	94	0.98
cdk2	168	245	0.97	188	44	60	0.96
cox2	125	96	0.96	148	36	20	0.99
egfr	296	504	0.94	157	74	126	0.97
fxa	384	416	0.94	150	100	100	0.94
hivrt	168	121	0.84	103	44	29	0.84
inha	296	504	0.96	116	78	122	0.96
p38	20	24	0.91	25	6	5	0.82
pde5	295	505	0.96	171	74	126	0.91
pdgfrb	276	524	0.96	149	73	127	0.95
src	444	356	0.94	157	112	88	0.94
vegfr2	132	103	0.93	129	35	27	0.95

Average CV accuracy is derived from all eligible models with CV accuracy greater than 0.9 except for the HIVRT data set which has no models with CV accuracy above 0.9. Therefore, a 0.8 threshold is applied

A 0.8 threshold is applied for the HIVRT data set