Free energy perturbation (FEP) is a physics-based method for predicting how small chemical changes will alter a ligand’s binding affinity to a protein. In practice, FEP uses molecular dynamics simulations in explicit solvent to evaluate relative binding free energies between closely related analogs through a thermodynamic cycle, enabling quantitative ranking of design ideas within a congeneric series. This makes FEP particularly well suited to lead optimization, where chemists are repeatedly choosing among many plausible substitutions and want a predictive, assay-calibrated way to prioritize what to make next.
(The above discussion doesn't do justice to the elegant statistical physics that undergird FEP; for details, see this review by Darren York and references therein.)
While other methods can also predict how small chemical modifications will affect binding affinity, like docking and certain co-folding methods, FEP consistently ranks among the most accurate and practically useful methods for binding-affinity prediction along a congeneric series. Unlike ML-based methods, FEP doesn't require training data, making it less susceptible to activity cliffs for ligands with structural features poorly represented in the training data. A 2023 paper from Gregory Ross and co-workers argues that FEP can even achieve performance comparable to experimental binding affinity assays.
The practical impact of FEP is best understood as an improvement in decision quality per synthesis cycle. In a fully prospective study on cathepsin L inhibitors, 36 novel compounds were synthesized and tested, and FEP-guided selections improved affinity for 8 of 10 picks, compared with 1 of 10 for other prioritization approaches evaluated alongside it. In other words, when synthesis and biology are the rate-limiting steps, increasing the hit rate of synthesized compounds directly reduces the number of design–make–test–analyze iterations required to reach a desired level of potency. This allows pre-clinical teams to work faster and spend more of their discovery budget optimizing crucial in vivo properties like permeability, bioavailability, and toxicity.
At the program level, FEP is often deployed as a pre-synthesis filter: use computation to evaluate far more ideas than can be experimentally made, then synthesize only the best supported candidates. In a large, prospective industrial deployment at Merck KGaA (2016–2019), FEP was applied across 12 targets and 23 chemical series, generating predictions for over 6,000 chemical entities and leading to more than 400 blindly predicted, novel molecules being synthesized and tested. The same report describes screening at least 50–100 ideas for custom-built libraries and aiming to screen 5–10 times more ideas than the maximum number of compounds that can be selected for synthesis. This "compute wide, synthesize narrow" strategy expands the evaluated design space without expanding wet-lab throughput and lets teams concentrate experimental effort on compounds with the highest predicted probability of delivering the next SAR advance.
Rowan is developing a modern low-cost FEP workflow designed to enable rapid and high-accuracy binding-affinity predictions at scale. We anticipate releasing this workflow via private beta in early 2026; if you're interested in being a part of this, please reach out!