Virtual Screening for COX-2 Inhibitors with Random Forest Algorithm and Feature Selection
Shangjie Ai, Yong Bai, Xiande Liu · 2017
The virtual screening technology has been widely used in the drug development process to shorten the development cycle with the help of quantitative structure-activity relationship (QSAR) modelling and machine learning. When constructing the training set for machine learning model, the redundancy of molecular descriptors can seriously affect the accuracy of the established learning model. In this paper, we propose to use the F-score based feature selection to select appropriate subset of molecular descriptors as the training set, and then employ the random forest algorithm to establish the classification model for predicting potential cyclooxygenase-2 (COX-2) inhibitors. The results demonstrate that our proposed method can improve the prediction accuracy of virtual screening for COX-2 inhibitors than without feature selection, and it also shows better prediction performance compared with SVM (Support Vector Machine) based classification model.