Session: COMP Poster Session:

A coronavirus disease 2019 (COVID-19) caused by a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) remains a major public health threat. As COVID-19 death toll reached two million worldwide and continues to increase, development of efficient SARS-CoV-2 therapeutics remains an urgent challenge. One potential COVID-19 drug target is a papain-like protease (PLpro), a viral enzyme that catalyzes hydrolysis of a viral polyprotein and is essential for coronavirus replication. Several naphthalene-based PLpro inhibitors were found effective at halting SARS-CoV-2 replication in cells. Recently resolved X-ray structures of a SARS-CoV-2 PLpro in complex with three such inhibitors revealed similarity of their binding modes. This suggests that other potent PLpro inhibitors can be found among ligands with the same scaffold.
In silico identification of tight-binding ligands is best predicted with molecular dynamics based free energy methods in explicit solvent. However, their practical application in drug discovery remains limited due to their high computational cost and a lack of automation. Here, we present an efficient automated workflow for predicting binding free energies (BFE) for large sets of congeneric ligands. To achieve high performance our workflow builds upon modern machine learning (ML) algorithms for BFE prediction based on molecular descriptors. A ML model is trained on alchemical BFE calculation results using active learning approach. We have applied this workflow to predict BFEs of approximately 3×10⁵ commercially available congeneric compounds to SARS-CoV-2 PLpro. The ligands with the lowest BFEs will be reported. Identification of potent SARS-CoV-2 PLpro inhibitors will make an important contribution to the design of specific SARS-CoV-2 therapeutics. In addition, our workflow allows for BFE prediction for congeneric compounds to any protein with known structure, which enables its use in other structure-based drug design projects.

Maria Kurnikova

Appearances