Benchmarking the Ability of Common Docking Programs to Correctly Reproduce and Score Binding Modes in SARS-CoV-2 Protease Mpro

Abstract

The coronavirus SARS-CoV-2 main protease, M pro , is conserved among coronaviruses with no human homolog and has therefore attracted significant attention as an enzyme drug target for COVID-19. The number of studies targeting M pro for in silico screening has grown rapidly, and it would be of great interest to know in advance how well docking methods can reproduce the correct ligand binding modes and rank these correctly. Clearly, current attempts at designing drugs targeting M pro with the aid of computational docking would benefit from a priori knowledge of the ability of docking programs to predict correct binding modes and score these correctly. In the current work, we tested the ability of several leading docking programs, namely, Glide, DOCK, AutoDock, AutoDock Vina, FRED, and EnzyDock, to correctly identify and score the binding mode of M pro ligands in 193 crystal structures. None of the codes were able to correctly identify the crystal structure binding mode (lowest energy pose with root-mean-square deviation < 2 Å) in more than 26% of the cases for noncovalently bound ligands (Glide: top performer), whereas for covalently bound ligands the top score was 45% (EnzyDock). These results suggest that one should perform in silico campaigns of M pro with care and that more comprehensive strategies including ligand free energy perturbation might be necessary in conjunction with virtual screening and docking.

Characterization of the non-covalent interaction between the PF-07321332 inhibitor and the SARS-CoV-2 main protease

“…We have found that in about ' 35% of these 28 structures Vina is able to identify the correct (crystallographic) binding pose with a tolerance of less than 3 Å RMSD, and in ' 70% of the cases the crystallographic binding mode was found among the poses within 0.3 kcal/mol of the best scoring pose. As observed in Ref., [ 38 ], Vina performances tend to improve for bulky ligands such as PF-07321332. Full results of this analysis can be found in the ESI.…”

“…Vina is known to improve the average accuracy of the (non covalent) binding mode predictions on the well established DUD-E benchmark set [ 37 ] of more than 50% compared to AutoDock4, and it was found to be a strong competitor against popular commercial programs, resulting at the top of the pack in many cases. In a recent paper [ 38 ], the ability to correctly reproduce the binding modes in SARS-CoV-2 3CL pro of popular docking programs, including Vina, was questioned. In this study [ 38 ], 85 co-crystal 3CL pro structures with non-covalently bound ligands were examined.…”

“…In a recent paper [ 38 ], the ability to correctly reproduce the binding modes in SARS-CoV-2 3CL pro of popular docking programs, including Vina, was questioned. In this study [ 38 ], 85 co-crystal 3CL pro structures with non-covalently bound ligands were examined. The authors found that in only 15% of these structures, Vina was able to identify the correct crystallographic pose (within a rather conservative 2 Å RMSD tolerance) with the lowest scoring function.…”

Best practices for repurposing studies

“…Are there experimentally determined values of affinity to compare to? Perhaps we can hope that docking will at least give us a reliable docking pose: sadly, for the case of mPro in SARS-CoV-2, this has been shown not to be the case [22], so even more care needs to be applied when using docking as a positive filter rather than a negative one.…”

Influenza viruses and coronaviruses: Knowns, unknowns, and common research challenges

“…Combined with extensive next generation sequencing, it will guide drug discovery and help understand the biological significance of amino acid variations, in particular in the context of a viral pandemic. Advanced computational and machine learning methods should also help improve the performance of in silico docking programs, whose limitations have been highlighted by the COVID-19 pandemic [ 155 ].…”