Predictive modeling by deep learning, virtual screening and molecular dynamics study of natural compounds against SARS-CoV-2 main protease

Abstract

The whole world is facing a great challenging time due to Coronavirus disease (COVID-19) caused by SARS-CoV-2. Globally, more than 14.6 M people have been diagnosed and more than 595 K deaths are reported. Currently, no effective vaccine or drugs are available to combat COVID-19. Therefore, the whole world is looking for new drug candidates that can treat the COVID-19. In this study, we conducted a virtual screening of natural compounds using a deep-learning method. A deep-learning algorithm was used for the predictive modeling of a CHEMBL3927 dataset of inhibitors of Main protease (Mpro). Several predictive models were developed and evaluated based on R², MAE MSE, RMSE, and Loss. The best model with R²=0.83, MAE = 1.06, MSE = 1.5, RMSE = 1.2, and loss = 1.5 was deployed on the Selleck database containing 1611 natural compounds for virtual screening. The model predicted 500 hits showing the value score between 6.9 and 3.8. The screened compounds were further enriched by molecular docking resulting in 39 compounds based on comparison with the reference (X77). Out of them, only four compounds were found to be drug-like and three were non-toxic. The complexes of compounds and Mpro were finally subjected to Molecular dynamic (MD) simulation for 100 ns. The MMPBSA result showed that two compounds Palmatine and Sauchinone formed very stable complex with Mpro and had free energy of −71.47 kJ mol⁻¹ and −71.68 kJ mol⁻¹ respectively as compared to X77 (−69.58 kJ mol⁻¹). From this study, we can suggest that the identified natural compounds may be considered for therapeutic development against the SARS-CoV-2.

Communicated by Ramaswamy H. Sarma

Keywords:

• COVID-19
• deep learning
• molecular docking
• main protease (MPro)
• natural compounds

Acknowledgements

The authors are thankful to the Department of Botany, Kumaun University, S.S.J Campus, Almora for providing the facility, space, and resources for this work. The authors also acknowledge Kumaun University, Nainital for providing high-speed internet facilities. We also extend our acknowledge to Rashtriya Uchchattar Shiksha Abhiyan (RUSA), Ministry of Human Resource Development, Government of India to provide Computational infrastructure for the establishment of Bioinformatics Centre in Kumaun University, S.S.J Campus, Almora.

Disclosure statement

The authors declare that there is no conflict of interest in this paper.