QSAR study of dihydrofolate reductase inhibitors activities based on optimization of correlation weights of local graph invariants

Abstract

The aim of this work is modeling the activities of 67 E. coli dihydrofolate reductase (DHFR) inhibitors 2,4-diamino-5-(substituted benzyl) pyramidines by optimization of correlation weights of local graph invariants (cwlgis), assisting genetic algorithm (GA). The relationship between the descriptors and molecular activities was modeled using radial basis function partial least squares (rbfPLS). Proper relation between activities and the optimized molecular descriptors were shown, R training = 0.92 and r test = 0.85. The results indicate promising potential for the optimization of a correlation weights scheme. It is the first application of GA in optimization of cwlgis. The study also shows the ability of rbfPLS, combined with GA, in a factor-based nonlinear QSAR modeling using simple flexible descriptors.