Machine Learning for Antiviral Drug Discovery: Application of XGBoost-TPE for Predicting Bioactivity of Hepatitis C Inhibitors

The development of effective antiviral medications remains a priority, particularly in combating diseases caused by viral infections such as the Hepatitis C virus (HCV). Despite the success of direct-acting antivirals, challenges such as drug-resistant strains and high treatment costs necessitate ongoing efforts to discover novel antiviral agents. Traditional drug discovery methods are laborious, costly, and resource-intensive, highlighting the need for more efficient approaches. Machine learning (ML) offers a promising alternative, with Quantitative Structure-Activity Relationship (QSAR) modeling widely used to predict the bioactivity of new compounds based on their chemical properties. However, the performance of ML models like Extreme Gradient Boosting (XGBoost) depends on complex hyperparameter optimization. This study demonstrates that combining XGBoost with the Tree-structured Parzen Estimator (TPE) significantly enhances the performance and efficiency of predicting the bioactivity of HCV NS3 inhibitors. The XGBoost-TPE model achieved superior performance with an R-squared of 0.856, root mean square error (RMSE) of 0.640, and mean absolute error (MAE) of 0.452, outperforming baseline models such as Support Vector Regression and standard XGBoost. The study highlights the potential of combining XGBoost with TPE to enhance the efficiency and accuracy of antiviral drug discovery, providing a scalable methodology for screening large libraries of chemical compounds and contributing to developing more effective antiviral therapies.