LAUSR

In this paper, we report the use of a mixture of radial distribution functions (RDFs) and molecular docking descriptors (MDDs), as a new group of descriptors, to construct a predictive quantitative structure-activity relationship (QSAR) model. The performance of the proposed mixed descriptors as the independent variables was checked with QSAR modeling of the anti-cancer activities of a series of 4-anilinoquinazoline analogs as the potent epidermal growth factor receptor (EGFR) inhibitors. The RDF descriptors were calculated using the available software. The docking descriptors were extracted by docking the understudied compounds into the active site of the protein with the PDB Code of 1M17 using molecular docking software. The stepwise linear regression was used to select the most important descriptors. The selected relevant descriptors were used as the inputs in the Bayesian regularization-artificial neural network (BR-ANN) as the QSAR model. The data set was randomly divided into training (35 compounds) and external test (8 compounds) sets. The mean square error (MSE) of the training set was applied for the selection of the optimal BR-ANN model. The validation of the proposed BR-ANN model was accomplished by the prediction of pIC 50 of compounds in the external test set and all molecules through the leave-one-out (LOO) technique. The results obtained confirmed the acceptable accuracy of the model ( R test 2 = 0.90 $$ {R}_{\mathrm{test}}^2=0.90 $$ and R LOO 2 = 0.79 $$ \kern0.50em {R}_{\mathrm{LOO}}^2=0.79 $$ ).