Skip to content
Kosha Sciences
← Research
Fisher flow matchingMolecular energetics

Learning transferable free energies

Fisher-conditioned flow matching for generalization across molecular transformations

Deciding which molecule to make next often means comparing candidates that differ by only a few atoms. Those changes can alter how a molecule interacts with its surroundings. Free-energy calculations help quantify such differences, but each comparison requires sampling many molecular configurations.

A model that learns from many such calculations could make that work reusable. There is a subtle obstacle: the coordinate normally used to describe a transformation tells us how the energy function changes, not how much the molecular distribution has changed. A point labeled “halfway” need not represent comparable progress in two different transformations.

This work tests a coordinate based on the change in those distributions. Fisher information provides a way to measure that change, giving a shared flow model a more consistent description of progress along each path. The study asks whether this helps on molecular pairs absent from training; it still uses simulation data from those held-out pairs to construct the coordinate.

The paper’s overview compares uniform lambda steps with normalized Fisher arc-length steps for two schematic molecular transformations. Their Fisher midpoints occur at different lambda values.View full-size figure

A shared coordinate for different transformations. The same coupling value can correspond to different amounts of distributional change. Normalized Fisher arc length instead measures fractional progress along each pair’s information-geometric path. The paths and profiles in this original figure are schematic, not molecular trajectories or measured Fisher profiles.

Figure 1 in the Fisher flow matching paper

The same input can mean different things

In an alchemical calculation, one molecular system is gradually transformed into another through a sequence of intermediate energy functions. These intermediates provide a computational path between the endpoint distributions.

The coupling parameter λ labels that sequence: λ = 0 is one endpoint and λ = 1 is the other. Molecular dynamics samples the intermediate distributions, and thermodynamic integration combines their average energy derivatives into a free-energy difference.

Thermodynamic integration(1)
ΔF=∫01⟨∂λUλ⟩pλ dλ\Delta F=\int_0^1\left\langle\partial_{\lambda}U_{\lambda}\right\rangle_{p_{\lambda}}\,d\lambda
The free-energy difference is the integral of the mean energy derivative along the alchemical path. Molecular dynamics supplies samples at discrete lambda windows; the paper uses trapezoidal integration of the window means.

The learning problem is to train one model across many transformations and apply it to a pair absent from training. That requires the model to understand both which pair it is working on and where it is along that pair’s path.

Raw λ gives every pair the same numerical interval, but it does not make their intermediate distributions comparable. A small step in λ may barely change one system and substantially change another. The paper asks whether a coordinate based on distributional change can make that shared learning problem easier. [1, §2]

Measure how much the distribution changes

Fisher information measures how sensitive a probability distribution is to a change in its parameter. Along an alchemical path, it is proportional to the variance of the same energy derivative used in thermodynamic integration. We write it as 𝓘 to distinguish it from free energy.

Fisher information and path length(2)
I(λ)=β2Var⁡pλ ⁣(∂λUλ),ℓ(λ)=∫0λI(u) du.\begin{aligned}\mathcal I(\lambda)&=\beta^2\operatorname{Var}_{p_{\lambda}}\!\left(\partial_{\lambda}U_{\lambda}\right),\\[6pt]\ell(\lambda)&=\int_0^{\lambda}\sqrt{\mathcal I(u)}\,du.\end{aligned}
At inverse temperature beta, Fisher information equals beta squared times the variance of the energy derivative. Integrating its square root from zero to lambda defines the cumulative Fisher arc length.

Where a distribution changes rapidly, Fisher length accumulates quickly. Where it changes slowly, the same λ increment contributes less length. The result is unchanged by a smooth monotonic relabeling of the path.

Locally, the Kullback–Leibler divergence between neighboring distributions is approximately ½𝓘(λ)(dλ)² = ½(dℓ)². Equal small steps in unnormalized Fisher length therefore correspond to equal leading-order local divergence. This is a local relationship; cumulative path length is not generally the finite KL divergence between the endpoints.

The implementation uses normalized length, ℓ̄ = ℓ/ℓmax, so every pair’s coordinate runs from zero to one. That choice matters because total lengths vary from 0.41 to 11.79 in the small-molecule corpus. Normalization aligns fractional progress along each path; it does not make the absolute information distance equal across pairs with different total lengths. [1, §3.2–3.3]

Compare the same input across two paths

Change near the end

011λℓ̄

λ 0.50Fisher progress 38%

Change near the start

011λℓ̄

λ 0.50Fisher progress 63%

Same number, different progress. At λ = 0.50, these paths have traversed 37.5% and 62.5% of their Fisher length. Switch to Fisher progress to align their fractional positions. This interactive illustration uses analytic profiles √𝓘(λ) = 0.5 + λ and 1.5 − λ, each with total length one. It is not experimental data or a reproduction of a molecular simulation.

Change the coordinate, keep the learning problem fixed

The model learns a one-dimensional distribution: the energy derivative x = ∂λU at a particular alchemical window. It does not generate atomic configurations. A conditional flow transports Gaussian samples into this scalar distribution, conditioned on the molecular pair and its position along the path.

The default transformer receives four inputs: the scalar sample x, flow time t, the within-pair coordinate s, and a vector describing the pair. Flow time t tracks the generative process; s tracks the alchemical path. The experiment changes s from raw λ to normalized Fisher length while preserving the rest of the comparison.

The transformer has four layers, four attention heads, and a hidden width of 64. The basic pair descriptor contains four atom-count quantities: total atoms, alchemical atoms, shared atoms, and the fraction of alchemical atoms. Training uses AdamW for 8,000 steps with batch size 512 and learning rate 3 × 10−3.

At evaluation, generated samples estimate the derivative’s mean at each λ window. Those means are still integrated over λ to obtain ΔF. Fisher length changes the model’s conditioning input; it does not replace the integration measure in the thermodynamic identity.

The held-out pair’s molecular-dynamics samples remain available. They provide the Fisher profile and the reference free energy used for evaluation. The demonstrated transfer removes the need to train a separate flow for that pair; it does not yet remove its simulation requirement. [1, §3.1–3.4]

The coordinate is only as good as its energy derivative

Alchemical simulations often soften interactions near the endpoints to avoid singular energies when atoms appear or disappear. Under a linear interpolation of endpoint energies, ∂λU equals UB − UA. A softcore Hamiltonian generally does not obey that identity.

Evaluating the endpoint difference on configurations sampled under softened interactions can produce extreme outliers: overlaps that are harmless in the sampled system become severe clashes in the fully interacting endpoint. Such errors contaminate both the thermodynamic-integration target and the variance used to construct Fisher length.

The paper instead evaluates a centered finite difference of the actual softcore Hamiltonian, with δ = 0.01 and one-sided differences at λ = 0 and 1. Both conditioning modes use that same corrected derivative. This isolates the coordinate comparison from errors introduced by an inconsistent energy estimator. [1, §3.5]

Evaluate transfer on held-out molecular pairs

The main chemistry experiments use five-fold cross-validation. Each fold withholds complete molecular pairs, and each configuration is trained with three seeds and evaluated with three sampling seeds. The raw-λ and Fisher-conditioned models share architecture, training data, optimizer settings, batches, and seeds.

The target is each held-out pair’s own simulation-derived ΔF. Absolute prediction errors are measured in kJ/mol. The aggregate percentage reduction compares the sum of errors under the two coordinates; a positive value favors Fisher conditioning. A pair counts as a win only when its error is less than 95% of the other mode’s error. Ties are excluded from the reported decided win rate.

A Lennard–Jones diagnostic provides a controlled physical example. Four training pairs use particle-size ratios σB/σA from 1.05 to 1.20; the held-out pair uses 1.30. The reported error falls from 1.70 ± 0.59 to 0.35 ± 0.19 kJ/mol, a reported 79.5% reduction. This is an extrapolation test on a simple interaction model, not a drug-binding benchmark. [1, §5.1]

A measurable gain—and a useful feature tradeoff

The small-molecule corpus contains 129 transformations across alkanes, alkylaromatics, alcohols, amines, thiols, and a heterogeneous remaining group. These are single-topology substituent extensions simulated in vacuum at 300 K, with AM1-BCC charges and the OpenFF Sage force field.

With a transformer and atom-count features, Fisher conditioning reduces aggregate held-out error by 5.6%, with a reported 95% interval of 0.8% to 9.9%. It wins 73 of 114 decided comparisons; the other 15 pairs are ties. The architecture and feature ablations show that this benefit is not universal.

129 small-molecule pairs · five-fold cross-validation
Architecture / featuresFisher wins / decidedError reductionReported 95% CI
MLP
Atom counts
51 / 94+0.6%[−1.9, +3.3]%
MLP
Morgan fingerprints
59 / 107+0.3%[−4.7, +4.9]%
Transformer
Atom counts
73 / 114+5.6%[+0.8, +9.9]%
Transformer
Morgan fingerprints
60 / 116−2.0%[−11.7, +6.7]%
Positive reduction favors Fisher conditioning. Aggregate reduction compares the sums of absolute errors; it is not the mean of per-pair percentages. A coordinate wins when its error is less than 95% of the other’s; remaining pairs are ties and excluded from the decided denominator. Results use three training seeds and three sampling seeds. Paper §5.2; metric definitions in Appendix J.

Morgan fingerprints improve the transformer’s absolute errors under both coordinates. Yet their inclusion removes the additional advantage of Fisher conditioning. The authors interpret this as partial substitution: chemical similarity features may already supply information that the geometric coordinate would otherwise contribute.

Transformer · mean absolute error on the 129-pair corpus (kJ/mol)
Pair featuresRaw λFisher coordinate
Atom counts7.667.24
Morgan fingerprints4.834.93
Richer chemical features improve absolute errors in both modes, while the additional benefit of Fisher conditioning disappears. These are reported means without uncertainty intervals. Paper §5.2.

The distinction is useful when interpreting the result. The transformer with atom counts is the clearest test of the coordinate hypothesis in these experiments. It is not the configuration with the lowest absolute errors among all reported feature choices.

The advantage depends on a meaningful molecular mapping

The second chemistry study uses transformations drawn from six target subsets of the pmx protein–ligand benchmark. Of 165 selected edges, 32 fail during system construction or simulation, leaving 133 usable pairs. The failures include 20 molecular-dynamics crashes and 12 charge-assignment failures.

This experiment uses newly run vacuum simulations. Its evaluation targets are those simulations’ ΔF values, not the published pmx binding free energies. It therefore tests generalization across drug-like transformations without establishing binding-affinity accuracy against experiment.

The main comparison focuses on 79 pairs with fewer than 30% of atoms in the alchemical region. These mappings retain a substantial shared molecular core. On that subset, the transformer with atom counts yields an 11.5% aggregate error reduction, with a reported 95% interval of 3.3% to 19.0%, and wins 46 of 69 decided comparisons.

79 pmx pairs with clean atom mappings · five-fold cross-validation
Architecture / featuresFisher wins / decidedError reductionReported 95% CI
MLP
Atom counts
27 / 56−1.3%[−4.4, +2.0]%
MLP
Morgan fingerprints
32 / 63−1.4%[−5.2, +1.8]%
Transformer
Atom counts
46 / 69+11.5%[+3.3, +19.0]%
Transformer
Morgan fingerprints
9 / 41−6.5%[−10.0, −3.6]%
Positive reduction favors Fisher conditioning. Aggregate reduction compares the sums of absolute errors; it is not the mean of per-pair percentages. A coordinate wins when its error is less than 95% of the other’s; remaining pairs are ties and excluded from the decided denominator. Results use three training seeds and three sampling seeds. Paper §5.3; metric definitions in Appendix J.

When the maximum-common-substructure mapping fails to retain a meaningful shared core, the pattern changes. Among 35 pairs with at least 70% alchemical atoms, the same configuration has a 50% decided win rate and a −3.1% aggregate reduction: Fisher conditioning is slightly worse in aggregate. The coordinate helps within the reported operating regime; it does not repair a poor molecular mapping.

Where Fisher conditioning helps

pmx · Transformer + atom counts
Fisher conditioning wins 66.7% of decided comparisons with clean and intermediate mappings, and 50.0% with broken mappings. Horizontal lines show 95% binomial intervals; the small intermediate group has the widest interval. Aggregate error reductions are +11.5%, +9.7%, and −3.1%, respectively.
Values and source
Transformer with atom-count features · pmx pairs
Mapping groupGroup size (n)Decided win rate95% intervalAggregate error reduction
Clean
<30% alchemical atoms
7966.7%54.9–76.6%+11.5%
Intermediate
30–<70% alchemical atoms
866.7%30.0–90.3%+9.7%
Broken
≥70% alchemical atoms
3550.0%33.6–66.4%−3.1%

Win rates and interval endpoints were recovered from the original vector figure and rounded to 0.1 percentage point. The intervals were not recalculated. Group sizes include ties and are not the denominators for decided win rates. Aggregate error reduction compares sums of absolute errors.

Download values and provenance ↗
The benefit depends on the mapping. Redrawn from Figure 2, preserving its 95% binomial intervals. Ties are excluded from decided win rates. The plotted group sizes cover 122 of the 133 usable pairs; the original figure does not account for the remaining 11. View in the paper or open the original figure ↗.

The paper also tests whether the square root in Fisher length matters. Replacing ∫√𝓘 dλ with ∫𝓘 dλ gives a −2.4% reduction, with a reported interval of −9.6% to 4.0%, compared with +6.5% for the canonical coordinate on the same 111-pair comparison. This supports the specific geometric construction rather than an arbitrary rescaling of λ. [1, §6]

Toward broader transfer in molecular modeling

The contribution is a concrete way to connect physical geometry to the inputs of a generative model. A coordinate that reflects distributional change can remove some of the mismatch that a shared model would otherwise need to learn from molecular descriptors alone.

Several steps remain before that becomes a replacement for a full free-energy workflow. The model would need a Fisher profile estimated from molecular features or a limited pilot simulation, rather than the held-out pair’s full simulation data. Broader validation would also require explicit solvent, appropriate hybrid topologies, improved atom mapping, and the protein-bound portion of the alchemical cycle.

For programmable biology, this work addresses one part of molecular design: learning physical quantities across related chemical systems. Its results motivate more transferable estimators while keeping the scope precise—scalar energy-derivative modeling under a stated simulation protocol, with benefits that depend on the architecture, features, and quality of the molecular path.

Sources and data

Read the paper

Results and original figures are taken from the supplied 15-page manuscript. The PDF uses “Fisher-Conditioned” in its title; the OpenReview listing uses “Fisher-constrained.” The interactive coordinate example is an explanatory illustration created for this article.

  1. Read the full manuscript ↗Fisher-Conditioned Flow Matching for Transferable Free Energy Estimation. Methods, results, simulation protocols, and appendices. PDF, 1.6 MB.
  2. OpenReview publication record ↗Rohin Maganti and Rahul Maganti. SPIGM workshop at ICML 2026. The public record lists the work under CC BY 4.0.