ABSTRACT
As the ultimate goal of developing quantitative structure-activity relationship (QSAR) models is to minimize as much as possible the errors between the predicted and the observed endpoint values, carrying out several validation measures is a mandatory task for QSAR practitioners. After generating QSAR models, the internal validation procedure is usually performed to select initial QSAR models. In the case of leave-one-out (LOO) cross-validation, which is the most routinely used method for initial evaluation of a QSAR model, one compound is removed from the entire data set followed by training the model using the rest of the studied molecules. The aim of bootstrapping is to generalize the relationship within the developed model. In this type of validation method, original data set is randomly divided into train and test sets several times. In conclusion, no QSAR model is considered applicable with high confidence if the required validation criteria are not satisfied.
