Anais da 43ª Reunião Anual Virtual da SBQ

Relative binding free energy (RBFE) calculations for small molecules are becoming commonplace. RBFE can be used to predict relative affinities of a series of ligands to a biological target, and, thus, be used to guide optimization in drug discovery campaigns. Free Energy Perturbation (FEP) using molecular dynamics (MD) to simulate non-physical transformations between ligand pairs are among the techniques most extensively used for RBFE prediction. To achieve the high accuracy and speed needed for such predictions to be useful, FEP calculations must use highly optimized MD code, enhanced sampling techniques, and modern force fields. Current hardware allows for hundreds of perturbations to be routinely simulated; however, automation of the setup process is crucial.

Here, we report PyAutoFEP, a tool to automate RBFE calculations using FEP. PyAutoFEP is flexible, allowing the user to control over most of the program options, and complete, aiding users through the whole process. PyAutoFEP is composed of three parts: a perturbation map generator, a hybrid topology builder, and an analysis module. PyAutoFEP is opensource, and can be downloaded at GitHub.

Three types of perturbation maps can be generated: star maps, wheel maps, and optimal maps. Optimal maps are generated by a modified Ant Colony Optimization, which minimizes the number of perturbed atoms, the distance between two nodes, and the number of perturbations, optionally making a map with cycle closures. The main module converts individual ligand structures and topologies into a hybrid topology,  prepares a protein-ligand complex, solvates and add counter ions to it. The hybrid topology is guided by the maximum common structure between the molecules, and multiple force fields can be parsed. PyAutoFEP allows use of arbitrary λ scaling factors for charged and Van der Waals interactions and for A and B end states. Finally, the tool prepares scripts for executing the perturbation. The ΔΔG is estimated by either Bennet Acceptance Ratio (BAR) or Multistep BAR, implemented in alchemlyb. PyAutoFEP includes tools to streamline analysis of ΔΔG convergence, Replica Exchange ratios, and ligand-receptor interactions.

To test and validate PyAutoFEP, we used it to study a series of ligands of the Farnesoid X Receptor, an RBFE challenge presented to the community by D3R. We used Amber03/GAFF with AM1BCC charges and Charmm36WYF/CgenFF to predict the ΔΔG of the series with and without enhanced sampling methods. Using Amber03/GAFF, we obtained results par with the reported values using FEP+, and far better than with pure GROMACS free energy code. Results with Charmm36WYF/CgenFF were inferior, showing poorer correlation with experimental data. The use of Hamiltonian Replica Exchange (HREX) or HREX with Solute Scaling did not improve the results.