Large-Scale Collaborative Assessment of Binding Free Energy Calculations for Drug
Discovery Using OpenFE

Computational binding affinity calculations comprise an important and increasingly accessible component of small molecule drug discovery; transparent assessment of tool performance at scale is therefore essential for practitioners and developers alike. This article comprehensively evaluates the Open Free Energy (OpenFE) relative binding free energy (RBFE) protocol, across approximately 1700 ligands and 95 targets from public and proprietary sources.

Importantly, the article explicitly addresses only the default OpenFE RBFE protocol. The headline outcomes across performance, reproducibility, convergence and throughput support routine use of the OpenFE RBFE protocol in commercial discovery contexts.
Perhaps unsurprisingly, performance was system-dependent and found to be predominantly influenced by input data quality and the alchemical transformation type. While no single error source was identified, worse performance on proprietary (cf. public)
datasets was partially attributed to transforming larger numbers of heavy atoms, i.e. inherent difficulty.

The compared commercial software solution FEP+ was overall more accurate, defined as having better error statistics, however reassuringly, the OpenFE RBFE protocol performance was comparable to FEP+ in terms of ranking, a common practical use-case. Caveats around direct comparison to FEP+ are highlighted, for example, the prior FEP+ results employed a proprietary force-field and custom ligand parameterisation. While the later is tractable with OpenFE using bespoke fit software, the force field dependance of comparative results cannot be decoupled from other factors. Protocol optimisation opportunities and modelling challenges in proprietary datasets are identified, usefully summarising an important “hazard” checklist for non-expert users. Taken together, this article makes a robust evidence-based case that the OpenFE RBFE protocol can be a useful drug discovery tool in commercial contexts. While the protocol is amenable to extensive configuration, it is encouraging that “out-of-the-box” default settings perform well at scale.

To read the article click here