In this post, we showcase an example of new science enabled by OpenBioSim’s software stack that was recently published in the Journal of Chemical Theory and Computation.
A3FE, which stands for Automated Adaptive Absolute Binding Free Energy Calculations, is a set of algorithms that, as the name suggests, largely automate the calculation of the absolute binding free energy of drug-like small molecules to a protein target, using alchemical free energy methods. A3FE was developed through EPSRC-funded collaborative industry-academic research between the University of Edinburgh, Newcastle University, and the pharmaceutical company AstraZeneca.
Why is this important?
Estimating the strength of a molecule’s binding to a protein is a cornerstone activity in computer-aided drug design. Molecular modelers working in drug discovery spend a significant portion of their time searching for molecules that could bind to a drug target (hit finding). Binding free energy calculations are also useful for optimizing the chemical structure of hits (hit-to-lead). These calculations remain valuable even after high-affinity compounds have been identified, as the focus often shifts to achieving good selectivity between structurally related proteins or maintaining strong affinity while optimizing other drug-like properties.
In this context, absolute binding free energy (ABFE) methods have shown great potential due to their higher accuracy and broader applicability compared to alternative approaches. A limitation of ABFE methods, however, is that they are relatively computationally intensive, requiring the execution of dozens to hundreds of molecular dynamics (MD) simulations per compound. This makes it challenging to process a large number of compounds in a timely manner. For this reason, ABFE methods are ideal candidates for active learning (AL) virtual screening, where ABFE-derived binding free energy estimates are used to periodically update inexpensive machine learning models that guide the exploration of chemical space.
Thus, the more efficient ABFE methods can be made, the more likely it is that they can be effectively deployed in virtual screening campaigns.
Adaptive sampling route to time savings
A3FE improves efficiency by automating decisions that practitioners must make when running an ABFE calculation. How many simulations should be run, and how long should each simulation last to obtain a reliable binding free energy estimate? This is a difficult question to answer, as optimal simulation settings can vary from system to system. To address this, A3FE uses adaptive algorithms. As data is generated, it is analyzed to suggest how to “tune” the protocol. Time savings are achieved by avoiding unnecessary simulations and by reducing simulation durations to the minimum required for reliable results.
A3FE was developed using components of OpenBioSim’s BioSimSpace software library to streamline the setup of inputs. The SOMD code implemented in OpenBioSim’s Sire software serves as the backend to compute binding free energies over short iterations. A3FE’s code then analyzes the results to determine whether to continue submitting calculations or terminate the process.
A3FE was tested on a range of protein-ligand systems, demonstrating performance that was equal to or better than simulation protocols manually selected by expert users. This shows that the methodology is robust and can be applied to unseen datasets. One such demonstration involved computing the absolute binding free energies of a dataset of structurally diverse cyclophilin D inhibitors, achieving good ranking accuracy when compared to experimental data.
References
A3FE paper) ”Automated Adaptive Absolute Binding Free Energy Calculations” Finlay Clark, Graeme Robb, Daniel J. Cole, Julien Michel J. Chem. Theory Comput., 20, 18, 7806-7828, 2024 doi:10.1021/acs.jctc.4c00806
BioSimSpace paper) ”BioSimSpace: An interoperable Python framework for biomolecular simulation” Hedges, L. O. ; Mey, A. S. J. S. ; Laughton, C. A. ; Gervasio, F. L. ; Mulholland, A. J. ; Woods, C. J. ; Michel, j. Journal of Open Source Software, 4(43), 1831, 2019 doi:10.21105/joss.01831
Sire paper) ”Sire: An Interoperability Engine for Prototyping Algorithms and Exchanging Information Between Molecular Simulation Programs” Christopher J. Woods, Lester Hedges, Adrian Mulholland, Maturos Malaisree, Paolo Tosco, Hannes H. Loeffler, Miroslav Suruzhon, Matthew Burman, Sofia Bariami, Stefano Bosisio, Gaetano Calabro, Finlay Clark, Antonia S. J. S. Mey, Julien Michel J. Chem. Phys. , 160, 202503, 2024 doi:10.1063/5.0200458
A3FE repository) https://github.com/michellab/a3fe
Written by Julien Michel.
