Alchemical Free Energy Methods Reproduce Scaffold Modifications in a Macrocyclic Peptide PCSK9 Series

Free energy perturbation (FEP) methods are widely used in drug discovery campaigns to optimise the binding affinity of small molecule ligands. Recently, macrocyclic peptides have emerged as an increasingly attractive modality alongside small molecules and biologics. They possess the unique ability to bind to large, flat, and featureless binding sites that are inaccessible to small molecules, while offering a route to oral bioavailability, which antibodies lack. However, modelling macrocycles with FEP presents significant challenges. These campaigns frequently involve complex scaffold changes that extend beyond traditional R-group modifications and often utilise both canonical and non-canonical amino acids, necessitating specialised parameterisation protocols to handle this diverse chemical space. In this work, we address these challenges by benchmarking against the comprehensive structure activity campaign of Tucker et al. [1], which details the development of novel macrocyclic peptide inhibitors targeting PCSK9. We present open-source relative binding free energy (RBFE) protocols and methodological strategies for the range of transformations required by this series, spanning buried backbone N-methylation, olefin transformations, ring-breaking, and changes in net charge. Furthermore, we automatically parameterise these chemically diverse inhibitors using the OpenFF force fields and graph neural network partial charges. We compare three analysis protocols that differ in how equilibration and correlated sampling are treated, and report performance both edgewise and after maximum-likelihood estimation of absolute free energies. Under the most rigorous of these, in which equilibration is detected automatically, the campaign is recovered with an edgewise RMSE of 0.93 kcal mol−1 (95% CI 0.74–1.10 kcal mol−1). Reaching compound 35, the most potent compound considered here, requires five consecutive perturbations, each exercising a distinct modelling challenge and each falling well outside conventional R-group substitution. Every edge along this path is reproduced to within 1.39 kcal mol−1. Overall, this work broadens the applicability of FEP methods in drug discovery while also identifying their current limits.