Modeling of non-additive mixture properties using the Online CHEmical database and Modeling environment (OCHEM)

Table 1 Statistical parameters of Online Chemical Modeling Environment (OCHEM, http://ochem.eu ) models with the lowest RMSE according to the “ compounds-out ” cross-validation protocol for the prediction of binary mixture densities

	“Compounds out” validation		“Mixtures out” validation		Test set prediction (“compounds out” validation)
Method/descriptors	R²	RMSE	R²	RMSE	R²	RMSE
LibSVM/Dragon	0.69 ± 0.05*	0.014 ± 0.001	0.81 ± 0.09	0.011 ± 0.002	0.88 ± 0.04	0.0089 ± 0.001
ASNN/Inductive descriptors [12]	0.68 ± 0.04	0.014 ± 0.0008	0.72 ± 0.04	0.0131 ± 0.0009	0.81 ± 0.06	0.011 ± 0.001
LibSVM/Inductive descriptors	0.71 ± 0.05	0.014 ± 0.001	0.81 ± 0.03	0.0109 ± 0.0005	0.88 ± 0.04	0.0084 ± 0.001
ASNN/Dragon [11]	0.56 ± 0.06	0.016 ± 0.001	0.69 ± 0.04	0.014 ± 0.001	0.85 ± 0.04	0.0099 ± 0.001
ASNN/ChemAxon [13]	0.55 ± 0.06	0.017 ± 0.001	0.69 ± 0.04	0.0137 ± 0.0009	0.88 ± 0.03	0.0088 ± 0.001

* 95% confidence intervals were calculated using a bootstrap procedure based on 1,000 replicas (implemented for all models in OCHEM). RMSE – Root Mean Squared Error; R² – square of the Pearson correlation coefficient.

ISSN: 1758-2946