Table 2.

Classification Performance of SVM Models Developed Based on Dataset Train2 with Various Descriptor Sets

	MOE	Schrödinger	ISIDA fragments
10-fold cross-validation (58 actives and 71 inactives)
Parameters	C = 40 γ = 0.1	C = 40 γ = 0.1	C = 40 γ = 0.001
Accuracy	93.01%	91.54%	90.71%
Sensitivity	94.83%	93.10%	87.93%
Specificity	91.55%	90.14%	92.96%
Prediction on external test set Test2 (44 actives and 51 inactives)
Accuracy	92.63%	89.47%	96.84%
Sensitivity	90.90%	81.82%	100.00%
Specificity	94.12%	96.08%	92.68%
Prediction on external test set DBB (284 weak actives)
Accuracy	19.37%	31.34%	34.51%
Prediction on external test set DBC (847 inactives)
Accuracy	96.22%	99.29%	86.89%