Predicting New Anti-Norovirus Inhibitor With the Help of Machine Learning Algorithms and Molecular Dynamics Simulation-Based Model

Oluwakemi Ebenezer; Nkululeko Damoyi; Michael Shapi

doi:10.3389/fchem.2021.753427

Predicting New Anti-Norovirus Inhibitor With the Help of Machine Learning Algorithms and Molecular Dynamics Simulation-Based Model

Front Chem. 2021 Nov 17:9:753427. doi: 10.3389/fchem.2021.753427. eCollection 2021.

Authors

Oluwakemi Ebenezer¹, Nkululeko Damoyi¹, Michael Shapi¹

Affiliation

¹ Department of Chemistry, Faculty of Natural Science, Mangosuthu University of Technology, Durban, South Africa.

Abstract

Hepatitis C virus (HCV) inhibitors are essential in the treatment of human norovirus (HuNoV). This study aimed to map out HCV NS5B RNA-dependent RNA polymerase inhibitors that could potentially be responsible for the inhibitory activity of HuNoV RdRp. It is necessary to develop robust machine learning and in silico methods to predict HuNoV RdRp compounds. In this study, Naïve Bayesian and random forest models were built to categorize norovirus RdRp inhibitors from the non-inhibitors using their molecular descriptors and PubChem fingerprints. The best model observed had accuracy, specificity, and sensitivity values of 98.40%, 97.62%, and 97.62%, respectively. Meanwhile, an external test set was used to validate model performance before applicability to the screened HCV compounds database. As a result, 775 compounds were predicted as NoV RdRp inhibitors. The pharmacokinetics calculations were used to filter out the inhibitors that lack drug-likeness properties. Molecular docking and molecular dynamics simulation investigated the inhibitors' binding modes and residues critical for the HuNoV RdRp receptor. The most active compound, CHEMBL167790, closely binds to the binding pocket of the RdRp enzyme and depicted stable binding with RMSD 0.8-3.2 Å, and the RMSF profile peak was between 1.0-4.0 Å, and the conformational fluctuations were at 450-460 residues. Moreover, the dynamic residue cross-correlation plot also showed the pairwise correlation between the binding residues 300-510 of the HuNoV RdRp receptor and CHEMBL167790. The principal component analysis depicted the enhanced movement of protein atoms. Moreover, additional residues such as Glu510 and Asn505 interacted with CHEMBL167790 via water bridge and established H-bond interactions after the simulation. http://zinc15.docking.org/substances/ZINC000013589565.

Keywords: anti-norovirus; hepatitis C virus (HCV); machine learning; molecular docking; molecular dynamics.