The comparison of machine learning methods for prediction study of type 2 diabetes mellitus’s drug design

Abstract

Dipeptidyl peptidase-4 (DPP-4) inhibitor is an important target Diabetes Mellitus (DM) drug discovery. A quantitative Structure-activity Relationship (QSAR) model using molecular descriptors can be developed with the Machine Learning (ML) approach which Extreme Gradient Boosting (XGBoost) represents one of the most promising tools to establish it. The other tools that are used to construct the QSAR model are Support Vector Regressor (SVR) and Neural Network (NN), which the result obtained will be compared with XGBoost. The prediction results are comparable with the experimental value of the DPP4 inhibitor, in which the results reveal the superiority of the XGBoost over SVR and NN with the R-square value of XGBoost is 0.94.