Comparative Evaluation of Covalent Docking Tools

• DOI: 10.1021/acs.jcim.8b00228
• Publication Date: 2018-06-11T22:14:15Z
• Authors: Andrea Scarpino; György G. Ferenczy; György M. Keserű
• Keywords: Molecular Docking Simulation; 0301 basic medicine; 03 medical and health sciences; Protein Conformation; Proteins; Thermodynamics; Databases, Protein; Ligands; Algorithms; Software; Protein Binding

Increased interest in covalent drug discovery led to the development of computer programs predicting binding mode and affinity of covalent inhibitors. Here we compare the performance of six covalent docking tools, AutoDock4, CovDock, FITTED, GOLD, ICM-Pro, and MOE, for reproducing experimental binding modes in an unprecedently large and diverse set of covalent complexes. It was found that 40-60% of the top scoring ligand poses are within 2.0 Å RMSD from the experimental binding mode. This rate showed program dependent increase and achieved 50-90% when the best RMSD among the top ten scoring poses was considered. This performance is comparable to that of noncovalent docking tools and therefore suggests that anchoring the ligand does not necessarily improve the accuracy of the prediction. The effect of various ligand and protein features on the docking performance was investigated. At the level of warhead chemistry, higher success rate was found for Michael additions, nucleophilic additions and nucleophilic substitutions than for ring opening reactions and disulfide formation. Increasing ligand size and flexibility generally affects pose predictions unfavorably, although AutoDock4, FITTED, and ICM-Pro were found to be less sensitive up to 35 heavy atoms. Increasing the accessibility of the target cysteine tends to result in improved binding mode predictions. Docking programs show protein dependent performance suggesting a target-dependent choice of the optimal docking tool. It was found that noncovalent docking into Cys/Ala mutated proteins by ICM-Pro and Glide reproduced experimental binding modes with only slightly lower performance and at a significantly lower computational expense than covalent docking did. Overall, our results highlight the key factors influencing the docking performance of the investigated tools and they give guidelines for selecting the optimal combination of warheads, ligands, and tools for the system investigated. Results also identify the most important aspects to be considered for developing improved protocols for docking and virtual screening of covalent ligands.