Table 2. AUC Values Determined on Training and Test Sets of Best Performing Ensembles Selected To Maximize AUC^a.

	androgen receptor			CDK2			PPAR-δ
method	size	training	test	size	training	test	size	training	test
exhaustive	6	0.894 ± 0.05	0.850 ± 0.04	4	0.934 ± 0.014	0.919 ± 0.019	6	0.950 ± 0.020	0.923 ± 0.023
slow heuristic	5	0.893 ± 0.04	0.850 ± 0.04	4	0.934 ± 0.014	0.919 ± 0.019	6	0.950 ± 0.020	0.923 ± 0.02
fast heuristic	3	0.890 ± 0.03	0.850 ± 0.04	4	0.934 ± 0.014	0.919 ± 0.019	8	0.942 ± 0.022	0.928 ± 0.03

^aThe column labeled “size” gives the number of target conformations in the optimally performing ensemble identified by each method; 95% confidence intervals are given. Androgen receptor ensembles were constructed from 10 MD conformations identified using pocket volume clustering and a crystal structure. CDK2 ensembles were constructed from five MD conformations identified using RMSD-based pocket clustering and a crystal structure. PPAR-δ ensembles were constructed from 12 crystal structures.