Usage of machine learning and computational chemistry methodologies to predict the activity of potential antiviral compounds against the Zika virus protease

Abstract

Zika virus is an arbovirus, it is transmitted through blood-feeding arthropods and can cause symptoms ranging from fever and malaise to more severe neurological and immunological complications. Despite the extensive research, there are no approved vaccines or antiviral drugs from health organizations for this infection. Its genome is a single-stranded ribonucleic acid. It encodes a polyprotein that is cleaved by proteases, with the NS2B-NS3 complex being the most crucial for viral replication. The aim of this study is to predict potential antiviral compounds that inhibit this protease using an integrative approach involving statistical analysis, machine learning and computational chemistry techniques. A compound database from ChEMBL was analyzed to identify the most significant features. The Wilcoxon rank-sum test revealed that eight of these features showed statistically significant differences. Subsequently, a machine learning model, developed using the exhaustive search method with a Random Forest classifier, identified the optimal combination of seven features, achieving an accuracy of 95.46%. In computational chemistry, initially, the appropriate crystal structure of the virus protease complex was selected, along with the compounds to be tested for their inhibitory potential. Molecular docking experiments were then conducted using Webina and Maestro. Five compounds in the end emerged as promising candidates, and all were classified as active with a probability of 70%. These findings highlight the effective integration of these approaches in identifying compounds that could potentially inhibit the Zika virus protease, providing valuable insights for future experimental validation.