The use of combined machine learning and in-silico molecular approaches for the study and the prediction of anti-HIV activit

Abstract

While the number of AIDS-related deaths continues to rise, efforts have been made to transform the disease into a manageable chronic condition. HIV protease inhibitors have become central to combination therapy. As a result, these inhibitors have become a major focus of anti-HIV drug development. This research takes a data-driven approach to drug development through the use of quantitative structure-activity relationship (QSAR) analysis. A dataset of 450 anti-HIV drugs was used to construct and validate models. Using extensive validation methods and various machine learning algorithms, the results clearly showed that the "ET" regression outperformed the other models (“XGB”, “LGBM”, “DT”, “RF”, “GB”, “Bag”, and “HGB”) in terms of goodness of fit, predictivity, generalizability, and model robustness. Promising compounds were subjected to molecular docking and molecular dynamics simulation, resulting in drugs with favourable pharmacokinetic and pharmacodynamic properties that consistently interact with the therapeutic target.