“Everything Out” Validation Approach for Qsar Models of Chemical Mixtures

Established strategies for validating QSAR models of binary mixtures of chemicals are not applicable to the most challenging case, which is the prediction of binary mixtures created by two compounds not present in the initial training set. In this study, we have addressed this challenge by introducing the “Everything Out” validation strategy where the external sets are deliberately formed by all binary combinations of two compounds excluded from the training set. The model accuracy is evaluated by the error of prediction for the external sets. We show that the “Everything Out” approach affords lower error of prediction for binary mixtures formed by two new compounds and similar error of prediction for mixtures with one new compound as compared to the alternative “Compound Out” validation approach. We posit that “Everything Out” should be employed as the preferred approach to validating QSAR models of binary mixtures.