Abstarct: Molecular docking was used to construct QSAR models for the prediction of the biological activities of three groups of medicinal compounds, and the molecular descxriptors were extracted from the enzyme-ligand interactions using Autodock4.2 software. Since that the molecular docking descxriptors do not represent the structural properties of ligands alone and also there is no strong relationship between molecular docking and biological activity, using structural descxriptors (calculated by DRAGON software) in addition to docking descxriptors is essential. In the first part, the study of the structure-activity of 43 quinazoline derivatives as cancer inhibitors was studied. 10 molecular docking descxriptors were extracted for all compounds. The data set was divided into two training (35 compounds) and test (8 compounds) sets. The stepwise regression (SR) as the variable selection method was applied to the train set data. The artificial neural network (ANN) model was developed with two selected descxriptors. The low determination coefficient (R2) and high mean square error (MSE) obtained for the test set (0.52 and 0.66 respectively) indicate the inappropriate predictability of the ANN model. In order to improve the results of the ANN model, radial distribution function (RDF) structural descxriptors were added to the molecular docking descxriptors. RDF descxriptors represent the difference of the distribution of atoms in the molecule and the effect of this distribution on the inhibitory activity of the compounds. SR method was implemented on 160 descxriptors (150 RDF and 10 molecular docking descxriptors) and 8 effective descxriptors were selected and used as the inputs of the ANN model. The high R2 and low MSE of the test set (0.90 and 0.15, respectively) prove the predictability and generalizability of the developed ANN model with a mixture of molecular-structure docking descxriptors. In the second part, the structure-activity of 73 inhibitors of AIDS that are analog structure of Diarylpirimidine (DAPY like) was studied. After extracting the molecular docking descxriptors, the data set was divided into three training (53 compounds), validation (10 compounds) and test (10 compounds) sets. Two SR selected descxriptors were introduced to the ANN model and all parameters of the ANN were optimized. R2 and MSE parameters of the test set were obtained 0.50 and 0.24, respectively. In order to improve the ANN model, the functional group counts structural descxriptors were calculated for HIV inhibitors. SR method was applied on 164 descxriptors (154 functional group counts and 10 molecular docking descxriptors), and the ANN model was developed using 7 selected descxriptors. The biological activities of the test set data were predicted in the optimum ANN model. R2 and MSE of the test set were obtained 0.89 and 0.16 respectively that showed the improvement of the results of the ANN model with a mixture of molecular-structure docking descxriptors. In the third section, a quantitative structure- activity relationship of the azine derivative as HIV inhibitors, was investigated. For this purpose, 10 molecular docking descxriptors were obtained using the docking of the studied compounds in the active site of the protein. The data set was divided into three training (53 compounds), validation (10 compounds), and test (10 compounds) sets. Among the 10 molecular docking descxriptors, 2 descxriptors were selected as the most effective descxriptors using SR method on the train set data. The ANN model was optimized using docking derived descxriptors. The biological activity values of the test set compounds were predicted using the optimum ANN model. R2 and MSE parameters of the test set were obtained 0.79 and 0.23 respectively. Then, for the purpose of improving the ANN model, 16 drug like indexes were calculated for the studied compounds and added to 10 molecular docking descxriptors. SR was employed to select the most effective descxriptors on the train set data. 4 important descxriptors selected from the total molecular docking and drug like indexes and they were used as ANN model inputs. The biological activities of the test set data were predicted using the optimum ANN model. R2 and MSE parameters of the test set are obtained and equal to 0.86 and 0.11 respectively. Therefore, using molecular docking descxriptors and drug like indexes simultaneously, improved the efficiency of the ANN model.