Factors Influencing the Binding of HIV-1 Protease Inhibitors: Insights from Machine Learning Models.

Abstract

HIV-1 protease inhibitors are crucial for antiviral therapies targeting acquired immunodeficiency syndrome (AIDS). Hundreds of HIV-1 protease complexes with various ligands have been resolved and deposited in the Protein Data Bank. However, binding affinity measurements for these ligands are not always available. This gap hinders a comprehensive understanding of inhibitor efficacy. To address this challenge, machine learning (ML) models were constructed and validated based on the crystallographic coordinates of 291 HIV-1 protease-inhibitor complexes, leveraging over 2500 molecular descriptors. The models achieved accuracy scores exceeding 0.85, and applied to predict the binding affinity of 274 additional complexes for which inhibition constants were not experimentally measured. Our analysis focused on three models, each with 8-9 features, and based on KBest with Random Forest, Recursive Feature Elimination with Random Forest, and Sequential Feature Selection with Support Vector Machine. The findings revealed key predictive features, including properties of HIV-1 protease inhibitors like charge distribution, hydrogen-bonding capability, and three-dimensional topology, as well as intrinsic properties of HIV-1 protease, such as active site symmetry and flap mutations. The study highlights the contribution of a comprehensive analysis of accumulated experimental data to enhance the structural understanding of this important molecular system.