Table 7. Predicted Accuracy of the External Validation Set Using the Models Built in This Work.

models	descriptors	methodsa	accuracy (%)	SE	SP
model 1A	MACCS fingerprint	k-NN	63.62	0.69	0.64
model 1B	PubChem fingerprint	k-NN	75.70	0.63	0.76
model 1C	CORINA Symphony	k-NN	94.05	0.58	0.94
model 2A	MACCS fingerprint	DT	77.60	0.62	0.78
model 2B	PubChem fingerprint	DT	84.56	0.45	0.85
model 2C	CORINA Symphony	DT	84.30	0.59	0.84
model 3A	MACCS fingerprint	RF	95.83	0.71	0.96
model 3B	PubChem fingerprint	RF	94.60	0.59	0.95
model 3C	CORINA Symphony	RF	98.37	0.43	0.98
model 4A	MACCS fingerprint	SVM	93.32	0.59	0.94
model 4B	PubChem fingerprint	SVM	90.60	0.44	0.91
model 4C	CORINA Symphony	SVM	98.14	0.46	0.98
model 5A	MACCS fingerprint	DNN	80.79	0.71	0.81
model 5B	PubChem fingerprint	DNN	80.67	0.52	0.81
model 5C	CORINA Symphony	DNN	97.97	0.57	0.98

^ak-NN: k-nearest neighbors; DT: decision tree; RF: random forest; SVM: support vector machine; DNN: deep neural net.