Machine-Learning Approach to Identify Potential Dengue Virus Protease Inhibitors: A Computational Perspective

Abstract

The global prevalence of dengue virus (DENV), a widespread flavivirus, has led to varied epidemiological impacts, economic burdens, and health consequences. The alarming increase in infections is exacerbated by the absence of approved antiviral agents against the DENV. Within flaviviruses, the NS3/NS2B serine protease plays a pivotal role in processing the viral polyprotein into distinct components, making it an attractive target for antiviral drug development. In this study, machine-learning (ML) techniques were employed to build predictive models for the screening of a library containing 32,000 protease inhibitors. Utilizing GNINA for structure-based virtual screening, the top potential candidates underwent a subsequent evaluation of their absorption, distribution, metabolism, excretion, and toxicity properties. Selected compounds were subjected to molecular dynamics simulations and binding free energy calculations via MM/GBSA. The results suggest that comp530 possesses binding potential to DENV protease as a noncovalent inhibitor with multiple positions for chemical substitutions, presenting opportunities for optimizing their selectivity and specificity. However, other compounds predicted via ML models may still provide a promising start for covalent inhibitors.