Evaluating and improving the robustness of alchemical binding free energy calculations using adaptive enhanced sampling methods

Abstract

Alchemical protein–ligand binding free energy calculations are currently a topic in computational chemistry which requires expert knowledge and a multitude of initial choices and parameters set by the researcher. While the impact of many of these decisions on the resulting free energy values has been explored in recent years, the influence of the initial protein crystal structure, as well as the protonation, tautomeric and rotameric states of the amino acid side chains on the free energy values have been underexplored. To perform these studies, a Python library (ProtoCaller) supporting an arbitrary level of automation for setting up and running binding free energy calculations is first developed and presented. Afterwards, it is shown that the choice of initial protein crystal structure can significantly impact the resulting free energy values at short timescales, while ligand rare events can induce discrepancies at longer timescales. Similarly, different initial histidine protonation, tautomeric and rotameric states can also result in free energy discrepancies, showcasing the need for enhanced sampling methods on protein and ligand degrees of freedom. To address these sampling problems, an alchemical variant of the sequential Monte Carlo (SMC) enhanced sampling method is presented and validated on a range of test cases. This methodology is then augmented with long-timescale sampling provided by simulated tempering (ST), whose initial parameters are obtained from a preliminary exploratory SMC simulation and are afterwards refined over time in an adaptive fashion. The resulting method—fully adaptive simulated tempering (FAST)—is completely automatable and does not require any system-dependent parameters, making it generally applicable to the ligand sampling problem. Finally, FAST is applied to relative protein–ligand binding free energy calculations, enabling their full automation in combination with adaptive enhanced sampling. This methodology improves free energy reproducibility by decreasing the number of initial choices made by the researcher and can also be readily generalised to other sampling scenarios, making it a highly relevant contribution to the field.

More information

Identifiers

ORCID for Jonathan Essex: orcid.org/0000-0003-2639-2746

Catalogue record

Export record