Abstarct: Tyrosine kinases are important mediators of the signaling cascade, which play key roles in diverse biological processes. Researches show that inhibition of this enzyme has curative effects on some cancers. In the first section, quantitative structure-activity relationship (QSAR) models for 60 analogues of 4-(3-bromoanilino)-6,7-dimethoxyquinazoline on epidermal growth factor receptor tyrosine kinase were constructed using Bayesian Regularized Artificial Neural Networks (BR-ANN). For comparison purpose, the multiple linear regression )MLR) was also tested. Among a large number of calculated descxriptors, only nine significant molecular descxriptors were obtained by stepwise regression, as the most feasible descxriptors, and then they were used as inputs for neural network. The data set was randomly divided into 45 training and 10 validation and 5 test sets and the neural network architecture and its parameters were optimized. The prediction ability of the model was evaluated using the validation, test data sets, and leave-one-out (LOO) cross-validation method. The mean square errors for the validation, test data sets and leave-one-out method were 0.1329, 0.1937 and 0.3367, respectively. The results obtained showed the excellent prediction ability and stability of the proposed model in the prediction of inhibitory ‏activity data of the corresponding analogues. In the second section, some derivatives of necroptosis inhibitors (Necrostatin-5) were studied (Necroptosis is the third cell death pathway apart from apoptosis and necrosis). In this study, A data set including 161 compounds were used to design predictive models. The models that were established using Bayesian Regularized Artificial Neural Networks, could divide the compounds into two active and inactive groups with the accuracy 100%. In the after stage, were established models with data set for 52 active compounds each described with a diverse set of descxriptors. The most effective descxriptors were selected using stepwise regresion and the predictive models were established using artificial neural networks. This model could predict the p(EC_50)(M) of the compounds with the mean square errors for the validation, test data sets and leave-one-out method 0.0651, 0.0767 and 0.0598, respectively. The prediction results are in good agreement with the experimental values.