Abstarct: For model construction, two kinds of descxriptors namely molecular docking descxriptors and calculated structural descxriptors were used. The new molecular docking descxriptors were derived from molecular docking by considering the interaction between compounds (as ligands) and protein (as receptor). The structural descxriptors were calculated from structure of compounds. After generation of descxriptos, the most important descxriptors were selected by applying stepwise regression (SR) and least absolute shrinlage and selector operator (LASSO) as variable selection methods. The selected descxriptors were used as inputs for construction QSAR models using Artificial Neural Network (ANN) and Random Forest (RF) as a non-linear and linear methods. For ANN model, the data set was divided into training set, valid set and test set containing 51, 11 and 11 compounds, respectively. The data set was divided into training set and test set including 62 and 11 compounds, respectively for RF modeling. All effective parameters of ANN and RF models were optimized. The validation study of the ANN and RF models was performed by prediction of the activities of the test set compounds. The obtained results from models validation showed correlation coefficient of 0.9274 and 0.8968 for prediction the activity of test set by ANN and RF models, respectively. The mean square error of 0.0597 and 0.0849 were also found for prediction of test set activity by ANN and RF models, respectively.