Identification of lead inhibitors for 3CLpro of SARS-CoV-2 target using machine learning based virtual screening, ADMET analysis, molecular docking and molecular dynamics simulations

Abstract

The SARS-CoV-2 3CLpro is a critical target for COVID-19 therapeutics due to its role in viral replication. We employed a screening pipeline to identify novel inhibitors by combining machine learning classification with similarity checks of approved medications. A voting classifier, integrating three machine learning classifiers, was used to filter a large database (∼10 million compounds) for potential inhibitors. This ensemble-based machine learning technique enhances overall performance and robustness compared to individual classifiers. From the screening, three compounds M1, M2 and M3 were selected for further analysis. Absorption, distribution, metabolism, excretion, and toxicity (ADMET) analysis compared these candidates to nirmatrelvir and azvudine. Molecular docking followed by 200 ns MD simulations showed that only M1 (6-[2,4-bis(dimethylamino)-6,8-dihydro-5H-pyrido[3,4-d]pyrimidine-7-carbonyl]-1H-pyrimidine-2,4-dione) remained stable. For azvudine and M1, the estimated median lethal doses are 1000 and 550 mg kg⁻¹, respectively, with maximum tolerated doses of 0.289 and 0.614 log mg per kg per day. The predicted inhibitory activity of M1 is 7.35, similar to that of nirmatrelvir. The binding free energy based on Molecular Mechanics Poisson-Boltzmann Surface Area (MM-PBSA) of M1 is −18.86 ± 4.38 kcal mol⁻¹, indicating strong binding interactions. These findings suggest that M1 merits further investigation as a potential SARS-CoV-2 treatment.