qsar study on dyrk1a inhibitors for regenerative therapy in diabetes

The qsar models were developed for predicting dyrk1a biological activity (ec50) with a series of 1,5-naphthyridines derivatives as highly potent dyrk1a-dependent inducers of human β-cell replication using multiple linear regressions (mlr) as a linear method and support vector machine (svm) as a nonlinear method. the 49 chemicals in data set were randomly partitioned into training and test subsets. for the selection of molecular descriptors, the genetic algorithm (ga) feature selection approach was used, followed by mlr and svm. testing the prediction abilities of the obtained models were conducted using the tests of cross-validation, y-randomization, and an external test set. by comparing the results of ga-mlr and ga-svm models, it is clear that ga-svm produced better results (r2train= 0.946, ftrain= 78.641, rmse train= 0.203), although both models had adequate predictive quality. using the predicted results of this study, new and potent dyrk1a inhibitors can be designed. in addition, this study provides insight into a new strategy to design diabetes drugs.