A machine learning-based virtual screening for natural compounds capable of inhibiting the HIV-1 integrase

HIV-1 integrase is an essential enzyme for the replication cycle, and currently, inhibitors are in first line of treatment many guidelines. Despite discovery new inhibitors, including a class molecules with different mechanisms action, resistance still relevant problem, adding options to therapeutic arsenal fight viral Sisyphean task. Because difficulty cost vitro screenings, machine learning-driven ligand-based virtual screenings alternative that can not only cut costs but also use valuable information about active compounds yet unknown action. In this work, we describe thorough model exploration hyperparameter tuning procedure dataset imbalance show several models capable distinguishing between or inactive against integrase. The best was then used screen natural product atlas compounds, resulting myriad share features known inhibitors. Here explore strengths shortcomings our discuss applicability domain guide differentiate “predictable” “unknown” regions chemical space.