Development of the Adaboost-SVM Model for the Classification of Estrogen Receptor-B Ligands

A new QSAR model for the classification of estrogen receptor-ß (ERß) selective ligand has been developed with adaptive boosting (Adaboost) and support vector machine (SVM). Compound structures were drawn in Molinspiration WebME Editor and imported into the E-Dragon 1.0 software to calculate seven categories descriptors. The selection of variables for each descriptor was performed with particle swarm optimization (PSO). On a known compound data set, mathematical model was obtained by AdaBoost using SVM as the base classifier. Among all descriptors in the model, the RDF descriptor exhibited the highest accuracy in the predictions, which contained five variables. By comparing with previous study, the AdaBoost-SVM model improved the prediction accuracy of the training set and the test set to 100.0% and 92.3%, up from 92.4% and 88.5% when only SVM was applied. The results indicate that the combination of Adaboost- SVM and PSO gives a powerful tool for QSAR studies and classification investigations.