FEP after docking: what free energy perturbation adds, and what it costs

Docking is a fast, rough filter: it sorts a large library into plausible and implausible, quickly and cheaply, but its score is not a reliable potency. Free energy perturbation, or FEP, sits at the other end of the trade-off. It simulates the physics in detail and predicts how a small chemical change shifts binding affinity, often within about one kilocalorie per mole of experiment. The catch is cost: FEP is orders of magnitude more expensive per compound. The skill is knowing which question each tool is for.

What FEP actually computes

Most practical FEP is relative (RBFE): instead of asking "how tightly does this molecule bind," it asks "how much does turning molecule A into molecule B change the binding." It does this by simulating a gradual, unphysical morph of one ligand into the other, both while bound to the protein and while free in water, and measuring the free-energy cost of each. The difference between those two legs is the change in binding affinity. Because it compares two close analogs, many systematic errors cancel, which is why RBFE is accurate enough to guide real lead optimization.

Why it is accurate but slow

FEP earns its accuracy by not cutting the corners docking cuts. It models the protein flexing, explicit water molecules moving in and out, and the entropy of the whole system, by running molecular dynamics across many intermediate steps. That is a lot of simulation for a single pair of molecules, which is why a single FEP comparison can take hours of GPU time while a docking pose takes seconds. You do not run FEP on a million compounds; you run it on the handful where getting the potency ranking right actually changes a decision.

Where it fits in the funnel

The sensible pipeline is a staircase. Docking (and cheaper machine- learning scores) triage the large library down to a shortlist. Only then, on that shortlist, does FEP earn its keep, resolving the close calls that docking cannot: which of two similar analogs is actually more potent, whether a substituent helps or hurts, whether a series is worth continuing. Used the other way around, FEP is simply too expensive to be a first pass. Used correctly, it is the last filter before you commit chemistry.

The honest limits

FEP is not magic. Every method in this space, the commercial ones included, carries an error floor around one kilocalorie per mole, so on a change whose true effect is smaller than that, the prediction is a coin flip on direction. It also needs a good starting pose and a well-behaved system; a large conformational change or a big charge difference makes it far harder and less reliable. FEP is a sharp tool for the right edges, not a universal oracle, and treating it as one is how teams waste compute.

Start with the fast pass

FEP only makes sense once you have a shortlist, and the shortlist comes from docking. Open Studio to run molecular docking online, free and in the browser: rank a set of compounds, and, because Liganx is mutation-aware, compare wild- type versus mutant binding in the same run. That shortlist is exactly the input a free-energy calculation is worth spending on.

Primary sources

• Wang L, Wu Y, Deng Y, et al. Accurate and reliable prediction of relative ligand binding potency in prospective drug discovery by way of a modern free- energy calculation protocol and force field. J Am Chem Soc 137, 2695-2703 (2015). doi:10.1021/ja512751q
• Cournia Z, Allen B, Sherman W. Relative binding free energy calculations in drug discovery: recent advances and practical considerations. J Chem Inf Model 57, 2911-2937 (2017). doi:10.1021/acs.jcim.7b00564