Summary of predictive performance for QSAR model of ERα inhibitory activity.

Fingerprint class	Training set		External set		R Tr 2 − QExt2
	R Tr 2	RMSETr	Q Ext 2	RMSEExt
AtomPairs 2D count	0.93	0.38	0.73	0.53	0.20
AtomPairs 2D	0.85	0.54	0.68	0.62	0.17
CDK fingerprinter	0.87	0.51	0.71	0.56	0.16
CDK extended	0.84	0.55	0.67	0.65	0.18
CDK graph only	0.81	0.60	0.70	0.58	0.11
E-state	0.80	0.63	0.64	0.71	0.16
Klekota–Roth count	0.91	0.41	0.72	0.54	0.19
Klekota–Roth	0.82	0.60	0.70	0.59	0.12
MACCS	0.86	0.52	0.71	0.58	0.15
PubChem	0.84	0.57	0.71	0.56	0.12
Substructure count	0.94	0.34	0.73	0.52	0.21
Substructure	0.87	0.51	0.68	0.63	0.19