Machine learning-aided identification of natural compounds targeting Marburg virus VP40 with potential antiviral activity.

Abstract

Marburg virus infection poses a significant threat to humans due to its high fatality rate. The application of in-silico drug design to target the essential protein target of the virus has been proven to be a fundamental technique to inhibit viral growth. Here, VP40 (a matrix protein) was used as an essential protein target of Marburg, and 2569 natural compounds were screened using the molecular docking and neural network-based DeepPurpose architecture. The top 138 compounds that exhibited a binding score of -8 kcal/mol in molecular docking were used in the best DeepPurpose model to predict the IC50. The best model in DeepPurpose was composed of Morgan and CNN-based encoding for protein and ligand, respectively. The top three compounds, NPL130 (CHEMBL2087156), NPL313 (CHEMBL76073), and NPL371 (CHEMBL54440), were selected from the machine learning model, and molecular dynamics simulation was performed for their best complex along with the control compound complex, Nilotinib (CHEMBL255863). Protein showed less than 0.3 nm of deviation in the complex formed with control and NPL130 compounds. Control had shown the best average MM/GBSA binding energy of -36.97 kcal/mol with the lowest standard deviation of 1.86. A stable complex was indicated by the negative binding free energies of -20 to -25 kcal/mol for all three hits. The free energy landscape showed that, along with the control compound, NPL130 had the biggest conformational space with the lowest free energy. Overall, this study proposed NPL130 as a potential hit compound to target Marburg viral growth.