QSAR-Based Stacked Ensemble Classifier for Hepatitis C NS5B Inhibitor Prediction

This study aims to develop a robust computational model for classifying potential Hepatitis C Virus (HCV) NS5B inhibitors using Quantitative Structure-Activity Relationship (QSAR) analysis. A large number of biological activity data for HCV NS5B was collected from the ChEMBL database, and molecular descriptors were calculated to characterize the compounds. A stacking classifier approach was employed for accurate classification using LightGBM + XGBoost as base models and Random Forest as the meta-classifier. The proposed model demonstrated improved performance compared to individual models, achieving an accuracy of 85.07% on the testing set. We assessed the model reliability through various evaluation procedures, including precision, recall, F1-score, and y-scrambling. The applicability domain was also analyzed to determine the model’s reliability for predicting compounds outside the training set. Overall, the developed model holds promise as a valuable tool in early-stage drug discovery for prioritizing compounds with inhibitory potential against HCV NS5B.