Iptp1b-3l: Virtual Screening For Ptp1b Inhibitors And Evaluation Of Their Their Therapeutic Efficacy Using Ensemble Learning Models

Abstract

Drug discovery brings challenges and opportunities for research in health science, particularly, finding the medicine for treating diabetes. In recent years, with the strong development of technology, machine learning has become an indispensable trend in all fields, especially, machine learning not only processes the huge amount of data but also solves the difficulties of discovering drugs that exist in many years. This research aims to construct 3 layers base on the dataset of 12 types of molecular fingerprints with an ensemble method. Besides, three common algorithms include Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB) with tuning parameters are applied to build both of classification model (layer 1 and layer 2) and regression model (layer 3). The result shows the efficient ability of model performance in layer 1 and layer 2 in the ensemble technique. In particular, the score of AUC in layer 1 and layer 2 are 0.903 and 0.911, respectively. However, the regression model in layer 3 not work effectively with only obtain 0.393 for MSE and 0.536 for R2.