Comparative molecular field analysis (CoMFA) studies have been carried out on 2,4-diamino-5-(2_-arylpropargyl)pyrimidine derivatives such as dihydrofolate reductase (DHFR) anticancer inhibitors. Because of the interoperable results of CoMFA models, they are noteworthy tools in rational drug designs. However, the huge amount of fields generated by this method contributes to its poor predictive ability. In this study, we applied a CoMFA region focusing (CoMFA-ReF) approach to weight and enhance or attenuate the contribution of lattice points on standard CoMFA interactions. In addition, the genetic algorithm (GA) is used to select interactions that are responsible for the inhibitory activities of these compounds. The selected fields were introduced to multiple linear regression, partial least squares (PLS), support vector machines (SVM), and random forest (RF) regression methods. Among the constructed models and in terms of root mean square predictions (RMSEP), the predictive power of the (CoMFA-ReF)-PLS (RMSEP = 0.252) was better than that of the others. The performances of the GA-RF regression model (RMSEP = 0.383) and GA-SVM (RMSEP = 0.387) were comparable. The pharmacophore-based alignment has been used as an intelligent alignment algorithm in the construction of a CoMFA standard model to improve the accuracies according to the (DHFR) protein environment. Docking studies clarified the role of these compounds in the inhibitory and anticancer activities of DHFR. Copyright © 2013 John Wiley & Sons, Ltd.