Open-source QSAR models for pKa prediction using multiple machine learning approaches

Table 1 Performance of SVM models using three data options with continuous descriptors, fingerprints and fragment counts

Data option	Data set	Feature sets	Number of features	Train		Fivefold CV		Test
Data option	Data set	Feature sets	Number of features	R²	RMSE	Q²	RMSE	R²	RMSE
1	Acidic	Continuous	870	0.96	0.65	0.58	2.18	0.68	1.91
1	Acidic	Fingerprints	1548	0.91	1.00	0.64	2.02	0.71	1.81
1	Acidic	Fingerprints + counts	2104	0.94	0.80	0.64	2.02	0.72	1.80
1	Basic	Continuous	876	0.96	0.64	0.65	1.94	0.65	1.93
1	Basic	Fingerprints	1535	0.91	0.99	0.69	1.84	0.69	1.83
1	Basic	Fingerprints + counts	2079	0.93	0.87	0.72	1.73	0.70	1.80
2	Acidic	Continuous	913	0.98	0.49	0.61	2.10	0.69	1.89
2	Acidic	Fingerprints	1552	0.9	1.05	0.63	2.04	0.69	1.87
2	Acidic	Fingerprints + counts	2141	0.94	0.85	0.63	2.05	0.71	1.81
2	Basic	Continuous	913	0.97	0.52	0.67	1.88	0.66	1.88
2	Basic	Fingerprints	1534	0.90	1.02	0.68	1.83	0.75	1.63
2	Basic	Fingerprints + counts	2085	0.93	0.88	0.71	1.76	0.78	1.53
3	Acidic	Continuous	510	0.96	0.66	0.59	2.17	0.57	2.20
3	Acidic	Fingerprint	1580	0.91	1.00	0.64	2.01	0.68	1.91
3	Acidic	Fingerprints + counts	2395	0.93	0.86	0.65	1.99	0.69	1.87
3	Basic	Continuous	510	0.95	0.75	0.61	2.01	0.6	2.09
3	Basic	Fingerprints	1543	0.91	0.94	0.72	1.72	0.67	1.90
3	Basic	Fingerprints + counts	2358	0.93	0.84	0.73	1.67	0.71	1.79

ISSN: 1758-2946