* Leave-one-out cross-validation (LOO CV) procedure is performed using the whole PASS training set for validation of prediction quality. The prediction result is compared with known experimental data for the studied compound. The procedure is repeated for all compounds from the PASS training set; then the Accuracy (ROC AUC) values are calculated for each type of metabolic stability.
* Leave-20-out cross-validation procedure is performed using the whole GUSAR training set for validation of prediction quality. The prediction result is compared with known experimental data for the studied compound. The procedure is repeated for all compounds from the GUSAR training set; then the ccuracy (R² for continuous models and Balanced Accuracy for categorial models) were calculated.

Training sets were created on the basis of the data from ChEMBL v.27

Quality models were also built by program PASS and programm GUSAR. GUSAR program uses quantitative neighborhoods of atoms (QNA), multilevel neighborhoods of atom (MNA) and whole-molecule descriptors in combination with self-consistent regression (SCR) or a combination of Radial Basis Function with self-consistent regression (RBF-SCR). The quantitative models used the values converted to log10 to unify the data over the range of values, also the median value was used in the case of several experimental values for one compounds.