CovCIFDock: Covalent Docking with CIFDock and Hybrid QM/MM Minimizations

Although reversible inhibitors represent the majority of approved and investigational drug therapeutics, irreversible inhibitors (or targeted covalent inhibitors, TCIs) are increasingly being considered due to a greater potency and higher selectivity. Many currently available commercial programs excel at modeling traditional reversible inhibitors. However, many of these non-covalent docking programs are unsuitable for docking TCIs because they do not model the post-reaction protein-ligand structure. Attempts to modify existing methods or formulate new ones that can simulate covalent docking often utilize classical mechanical potentials that inherently lack the ability to modulate bonding topologies in an effort to reduce computational overhead. Not only are such modifications ill-suited to covalent docking, they can lead to a reduction in docking accuracy and impose limitations on the transferrability of a method to other protein systems. Herein we aim to address these shortcomings by utilizing a hybrid simulation workflow that employs both classical and quantum mechanical (QM/MM) potentials capable of bond rearrangement. The workflow, named CovCIFDock, is comprised of modules natively available in CHARMM and is an extension of the recently published CIFDock flexible docking method. Protein-ligand complexes are docked classically in the pre-reactive state in order to facilitate computational speed, after which a QM/MM minimization is conducted to form the protein-ligand bond and refine the final pose. CovCIFDock was validated against 46 protein receptors belonging to Cathepsin K, HCV NS3, EGFR, and XPO1 targets. Results show that CovCIFDock is able to replicate the binding mode of these TCI complexes within 2 Angstrom of the crystal structure, and is comparable in accuracy to leading commercial docking programs.