Proteochemometric modeling in a Bayesian framework

Table 2 Internal and external validation metrics for the PCM models

		Adenosine Receptors Dataset
	$R_{int}^{2}$	RMSEP_int	$R_{0 ext}^{2}$	RMSEP_ext
GP Bessel	0.64	0.70	0.70	0.63
GP Laplacian	0.67	0.68	0.67	0.66
GP Norm. Polynomial (NP)	0.69	0.65	0.75	0.58
GP Polynomial	0.70	0.64	0.70	0.63
GP PUK	0.57	0.79	0.56	0.77
GP Radial	0.65	0.69	0.65	0.68
PLS	0.29	0.97	0.30	1.00
SVM Norm. Polynomial (NP)	0.70	0.64	0.73	0.60
SVM Polynomial	0.71	0.63	0.71	0.62
SVM Radial	0.68	0.65	0.70	0.64
Family QSAR	0.31	0.70	0.31	0.96
		Aminergic GPCRs Dataset
	$R_{int}^{2}$	RMSEP _int	$R_{0 ext}^{2}$	RMSEP _ext
GP Bessel	0.56	0.83	0.56	0.80
GP Laplacian	0.62	0.78	0.63	0.75
GP Norm. Polynomial (NP)	0.69	0.68	0.72	0.66
GP Polynomial	0.68	0.71	0.70	0.68
GP PUK	0.46	0.93	0.46	0.90
GP Radial	0.69	0.69	0.71	0.66
PLS	0.69	0.69	0.27	1.05
SVM Norm. Polynomial (NP)	0.69	0.68	0.72	0.66
SVM Polynomial	0.69	0.69	0.71	0.66
SVM Radial	0.69	0.69	0.72	0.66
Family QSAR	0.38	0.98	0.38	0.97
		Dengue virus NS3 proteases Dataset
	$R_{int}^{2}$	RMSEP _int	$R_{0 ext}^{2}$	RMSEP _ext
GP Bessel	0.91	0.43	0.92	0.44
GP Laplacian	0.88	0.54	0.91	0.50
GP Linear	0.91	0.45	0.91	0.48
GP Norm. Polynomial (NP)	0.88	0.50	0.91	0.48
GP Polynomial	0.91	0.42	0.92	0.44
GP PUK	0.77	1.10	0.81	1.13
GP Radial	0.91	0.45	0.91	0.45
PLS	0.90	0.45	0.91	0.49
SVM Norm. Polynomial (NP)	0.86	0.54	0.91	0.46
SVM Polynomial	0.89	0.46	0.90	0.51
SVM Radial	0.90	0.48	0.90	0.48
Family QSAR	0.29	1.19	0.48	1.13

For the three datasets, the best models are obtained with GP, being the lowest RMSEP_ext and highest R²_{0 ext} values: (i) adenosine receptors: 0.58 and 0.75 with NP kernel, (ii) GPCRs: 0.66 and 0.72 with NP kernel, and (iii) Dengue virus NS3 proteases 0.44 and 0.92 with Bessel kernel. Overall, GP models for the three datasets agree with the validation criteria.
Abbreviations: RMSEP root mean square error in prediction, Ext. external, Norm Normalized.

ISSN: 1758-2946