Hierarchical Modeling of Binding Affinity Prediction Using Machine LearningTechniques

Abstract

Predicting the binding affinity of compounds is an essential task in drug discovery. In silico QSAR regression and classification models to predict drug-target interaction can help speed up identifying the most potent compounds. Machine learning-based QSAR models were developed to predict the binding affinity of compounds against different targets using the experimental values or labels. In this work, we modeled the binding affinity prediction of SARS-3CL protease inhibitors using hierarchical modeling. We developed the Base classification and regression models using KNN, SVM, RF, and XGBoost techniques. Further, the predictions of the base models were concatenated and provided as inputs for the stacked models. The results indicate that stacking of models hierarchically leads to improved performances on both classification and regression endpoints.