Table 3 Performance of QSAR classification models for P4up activity in H295R assay.

QSAR-P4up model (n structures = 1700)
	Acc	bAcc	Se	Sp	MCC	AUC
Tenfold cross-validation (training set, n active = 155, n inactive = 1290)
CART	0.87	0.61	0.29	0.94	0.25	0.69
SVM-rbf	0.89	0.52	0.04	1	0.11	0.52
NN	0.89	0.5	0	1	0	0.71
LDA	0.88	0.62	0.3	0.95	0.28	0.76
DNN	0.83	0.62	0.36	0.89	0.24	0.62
RF	0.99	0.99	0.98	0.99	0.95	0.99
RF balanced	0.76	0.86	1	0.73	0.47	0.86
Fitting (training set, n active = 155, n inactive = 1290)
CART	0.97	0.87	0.75	1	0.83	0.92
SVM-rbf	0.99	0.97	0.95	1	0.97	0.97
NN	1	0.99	0.99	1	0.99	0.99
LDA	0.91	0.71	0.44	0.97	0.49	0.92
DNN	0.52	0.71	0.97	0.46	0.27	0.71
RF	0.99	0.99	0.98	0.99	0.94	0.99
RF balanced	0.76	0.86	1	0.73	0.47	0.86
External validation (test set, n active = 26, n inactive = 229)
CART	0.89	0.66	0.37	0.95	0.35	0.78
SVM-rbf	0.9	0.52	0.04	1	0.18	0.52
NN	0.91	0.56	0.11	1	0.32	0.56
LDA	0.87	0.58	0.22	0.94	0.19	0.76
DNN	0.5	0.67	0.45	0.89	0.21	0.67
RF	0.89	0.66	0.37	0.96	0.37	0.66
RF balanced	0.73	0.8	0.89	0.71	0.39	0.8

Acc accuracy, bAcc balanced accuracy, Sp specificity, Se sensitivity, AUC Area Under the Receiver Operating Characteristic Curve, MCC Matthew Coefficient Correlation, see “Material and methods”.

Quick links

Search