Computing how tightly a drug binds to a protein is one of the fundamental problems in computational chemistry. The standard approach — alchemical free energy perturbation — works by slowly transforming one molecule into another through unphysical intermediate states. The molecule doesn't actually morph; the simulation walks through a sequence of fictitious hybrid Hamiltonians, collecting statistical samples at each step. It's accurate but expensive: each ligand requires many intermediate simulations (“lambda windows”), and each window must independently equilibrate.
Brocidiacono et al. (arXiv:2603.12253) bypass the alchemy entirely. Their Direct Binding Free Energy (DBFE) method requires exactly one simulation of the protein-ligand complex per ligand. No intermediates. No transformations. The receptor and solvent simulations can be precomputed once and reused across all candidate molecules.
The method performs comparably to standard alchemical approaches on protein-ligand benchmarks while using a fraction of the computation. On host-guest systems it actually outperforms double decoupling.
The structural insight is about the cost model. Alchemical methods are expensive per-ligand but amortize well when comparing similar molecules (relative binding free energies between close analogs). DBFE is cheap per-ligand but doesn't share information between compounds. This makes the two approaches complementary along a different axis than accuracy: alchemy is better for optimizing within a chemical series, DBFE is better for screening across diverse scaffolds. The bottleneck in drug discovery isn't comparing similar molecules — it's deciding which scaffolds to explore. The direct method attacks the expensive part of the pipeline.