Input Pose is Key\nto Performance of Free Energy Perturbation:\nBenchmarking with Monoacylglycerol Lipase

Free energy perturbation (FEP) methodologies have become\ncommonplace\nmethods for modeling potency in hit-to-lead and lead optimization\nstages of drug discovery. The conformational states of the initial\nposes of compounds for FEP+ calculations are often set up by alignment\nto a cocrystal structure ligand, but it is not clear if this method\nprovides the best result for all proteins or all ligands. Not only\nare ligand conformational states potential variables in modeling compound\npotency in FEP but also the selection of crystallographic water molecules\nfor inclusion in the FEP input structures can impact FEP models. Here,\nwe report the results of FEP calculations using FEP+ from Schrödinger\nand starting from maximum common substructure alignment and docked\n poses generated with an array of docking methodologies. As a benchmark\ndata set, we use monoacylglycerol lipase (MAGL), an important clinical\ndrug target in cancer malignancy, neurological diseases, and metabolic\ndisorders, and a set of 17 MAGL inhibitors. We found a large variation\namong FEP+ correlations to experimental IC<sub>50</sub> values depending\non the method used to generate the input pose and that the inclusion\nof ligand-based information in the docking process, with some methods,\nincreases the correlation between FEP+ free energies and IC<sub>50</sub> values. Upon analysis of the initial poses, we found that the differences\nin FEP+ correlations stemmed from rotation around a tertiary amide\nbond as well as translation of the compound toward the more hydrophobic\nside of the MAGL pocket. FEP+ estimation improved across all pose\nmodeling methods when hydrogen bond constraint information was added.\nHowever, simple maximum common substructure alignment in the presence\nof all crystallographic water molecules outperformed all other methods\nin correlation between estimated and experimental IC<sub>50</sub> values.\nTaken together, these findings suggest that pose selection and crystallographic\nwater inclusion greatly impact how well FEP+ estimated IC<sub>50</sub> values align with experimental IC<sub>50</sub> values and that modelers\nshould benchmark a few different pose generation methodologies and\ndifferent water inclusion strategies for their hit-to-lead and lead\noptimization drug discovery projects.