Abstarct: In this thesis, a quantitative study of the structure-activity relationship for some phenylalanine derivatives was carried out. Considering that the interpretability of QSAR models is very important, therefore, by using an innovative method, it is tried to reduce the number of selected descxriptors of the SCAD penalized method. Therefore, frequency method are used to find effective descxriptors. So, in this method, the training and test data are changed 100 times and the most effective descxriptors with the highest frequency are selected, and 6 variables were selected. Considering that the relationship between effective descxriptors and drug activity is often non-linear, a powerful neural network modeling method is used to find the relationship between descxriptors and drug activity. By using the SCAD variable selection method (using frequency technique) coupled with the neural network, the interpretability and predictive power of the Multistage-SCAD-ANN model increases. After creating the model, the medicinal activity of the test series compounds was predicted and the coefficient of determination for the test series was 0.823. In the following, we evaluated the model with various statistical methods, which proves the predictive power and generalizability of the Multistage-SCAD-ANN model. And at the end, a number of new chemical compounds were proposed, whose medicinal activity was measured using the developed Multistage-SCAD-ANN model, which used molecular docking study to prove the medicinal activity of the proposed compounds.