Quantitative Read-Across (q-RA) and Quantitative Read-Across Structure–Activity Relationships (q-RASAR)—Genesis and Model Development

Recently the concept of read-across has been applied to machine-learning-based supervised predictions for quantitative-read-across (q-RA) which have shown superior performance over QSAR-derived predictions in several examples. This was further extended to the generation of QSAR-like statistical models, i.e., quantitative read-across structure-activity relationship (q-RASAR) by using various similarity and error-based descriptors computed from original structural and physicochemical descriptors. Several composite functions like the RA function, Average similarity, Banerjee-Roy concordance measures (gm and gm_class), and Banerjee-Roy similarity coefficients (sm1 and sm2) have been computed for the query set from the source compounds and used for the predictions of a target property from well-validated models developed from the training set. The quality of predictions (quantitative or classification-based) is judged from the usual quality and validation metrics for QSAR models. In general, it has been found that the q-RASAR approach enhances the quality of predictions compared to the corresponding QSAR models.