Pith. sign in

REVIEW 3 major objections 4 minor 63 references

AquaGen generates all-atom, explicit-solvent molecular configurations from the Boltzmann distribution that match MD free-energy accuracy at a fraction of the cost.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 01:57 UTC pith:TQINM6X4

load-bearing objection First credible all-atom explicit-solvent Boltzmann generator at ~10³ atoms that actually feeds MBAR for AHFE; the ~1 kcal/mol headline is real but partly flattered by cancellation and teacher-force-field matching. the 3 major comments →

arxiv 2607.03513 v1 pith:TQINM6X4 submitted 2026-07-03 physics.chem-ph cs.LG

AquaGen: Scaling generative models to molecular dynamics precision on thousands of atoms

Emmanuel Bengio , Sanjeev Raja , Yui Tik Pang , Kerstin Klaeser , Cristian Gabellini , Nikhil Shenoy , Francesco Di Giovanni , Prudencio Tossou This is my paper
classification physics.chem-ph cs.LG
keywords generative modelsmolecular dynamicsabsolute hydration free energyflow matchingexplicit solventBoltzmann distributionalchemical free energyMBAR
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper claims that a flow-matching generative model can produce full all-atom configurations of drug-like molecules in explicit water under periodic boundaries that are close enough to true Boltzmann ensembles for free energies to be computed the same way as in molecular dynamics. Prior generative models drop solvent, use implicit solvent, or coarse-grain, so their frames are not plug-compatible with industrial force fields. By conditioning generation on an alchemical order parameter and evaluating OpenFF/TIP3P energies on the samples, AquaGen recovers absolute hydration free energies with roughly 1 kcal/mol mean error versus GPU MD on held-out compounds while using 4–10× less compute; short MD refinement from the generated starts can cut error further below 0.5 kcal/mol. A sympathetic reader cares because free energies ground decisions in drug and materials work, and a gray-box route—learned sampling plus white-box energies—gives inspectable, refinable predictions with calibrated uncertainty that pure black-box regression does not.

Core claim

AquaGen is the first generative model to sample Boltzmann-distributed all-atom configurations that include explicit solvent and periodic boundaries for systems of thousands of atoms, with geometric and energetic fidelity high enough that MBAR free-energy estimates from force-field energies on those samples match molecular-dynamics absolute hydration free energies to about 1.22 kcal/mol mean (0.93 median) error on held-out compounds, at 4–10× lower GPU cost.

What carries the argument

λ-conditional flow matching: a graph neural network learns a velocity field that transports a Gaussian prior over atomic coordinates and a cubic simulation cell into alchemical Boltzmann configurations conditioned on the order parameter λ that gradually annihilates solute–solvent interactions; uncorrelated samples at each λ are scored with the same force field used in MD and pooled by the multistate Bennett acceptance ratio (MBAR) estimator to obtain free-energy differences.

Load-bearing premise

Samples at neighboring alchemical windows must have energy overlaps faithful enough that free-energy increments from MBAR are trustworthy, not merely that opposite-signed errors along the electrostatic and van der Waals legs cancel to a small final number.

What would settle it

On held-out compounds, compare cumulative absolute error (sum of absolute adjacent-λ free-energy increments) and box-length bias of AquaGen ensembles against long MD; if CAE stays large while AHFE absolute error looks small, or if short MD refinement from generated starts fails to approach MD free energies within the reported compute budgets, the claim that the generative ensembles are Boltzmann-faithful for free-energy work fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Absolute hydration free energy can be estimated 4–10× faster than GPU MD with comparable accuracy, and still faster when short MD refinement starts from generated frames.
  • Scaling model size and the number of generated samples predictably improves free-energy accuracy (train- and test-time compute).
  • Bootstrapped confidence intervals over generated samples give well-calibrated uncertainty without deep ensembles.
  • The same high-resolution ensemble approach is positioned to extend to lipophilicity, membrane permeability, and absolute binding free energy.
  • Gray-box predictions remain refinable and inspectable via force-field energies, unlike pure black-box regressors.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If adjacent-λ energy overlaps remain faithful at protein–ligand scale, the same pipeline could cut wall-clock cost of absolute binding free energy campaigns without abandoning force-field grounding.
  • Opposite-signed errors on the electrostatic versus van der Waals legs imply that better λ conditioning may improve reliability more than further model scaling alone.
  • Explicit-solvent generative ensembles may become a practical intermediate between structure predictors and full MD for any property that is an ensemble average of potential energies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. AquaGen is presented as the first all-atom, explicit-solvent, PBC-aware generative model (flow matching on a GNN) that samples approximately Boltzmann-distributed configurations of solvated drug-like molecules (~10^3 atoms) conditioned on an alchemical parameter λ. Samples are used with force-field energy evaluations and MBAR to estimate absolute hydration free energies (AHFE) on held-out compounds, claiming ~1.22 kcal/mol mean / 0.93 kcal/mol median absolute error versus GPU MD at 4–10× lower cost, with further gains from model scaling, more samples, short MD refinement, calibrated bootstrap uncertainties, and gray-box interpretability relative to black-box regressors.

Significance. If the energetic fidelity and free-energy accuracy hold under broader scrutiny, this is a substantial advance: it closes a resolution gap relative to prior biomolecular generative models (Table 3) and shows that high-resolution ensemble generation can serve as a practical, refinable surrogate for alchemical MD free-energy workflows. The multi-layered evidence (energy histograms/decompositions, RDFs, tICA, H-bond trends, train/test-time scaling in Fig. 1c, uncertainty calibration, CAE ablations, and black-box baselines) and explicit gray-box advantages (refinability, inspectable samples) are genuine strengths that would matter for industrial free-energy tasks and for scaling toward ABFE or related properties.

major comments (3)
  1. [§4.4–4.5, Fig. 4c, Table 2] §4.4–4.5, Fig. 4c and Table 2: The headline AHFE AE (~1.22/0.93 kcal/mol) is a telescoping sum of free-energy increments and can be substantially smaller than the cumulative absolute error (CAE) because of systematic opposite-signed errors (negative on the electrostatic leg k=1–5, positive on the VDW leg k=6–20). Several ablations improve AE while worsening CAE, and the authors themselves attribute this to imperfect λ-conditioning that learns an “averaged” distribution. For the central claim that generated ensembles are faithful enough for trustworthy MBAR free energies (not merely that the endpoint difference happens to match after cancellation), the paper needs either (i) stronger evidence that adjacent-state energy overlaps and per-leg increments are accurate, or (ii) primary reporting of CAE (or equivalent) alongside AE, with clearer caveats on when cancellation is acceptable.
  2. [§4.1, Table 1] §4.1, Table 1 and FreeSolv/CombiSolv columns: Without MD refinement the model’s mean AE on public experimental sets is ~3–4 kcal/mol (worse than the internal MD-matched numbers), while short refinement brings it below 1 kcal/mol. The abstract and main claims emphasize “comparable accuracy to standard GPU-based MD” and 4–10× speedup largely on the internal MD-vs-MD comparison under the same force field. The manuscript should more carefully separate surrogate fidelity to the teacher MD ensemble from transfer to experiment, and quantify how much of the claimed speedup remains after the refinement step that is needed for competitive experimental accuracy.
  3. [§4.3, §4.5] §4.3 and §4.5 (Gaussian prior / box-size results): There is a consistent ~0.1 Å overestimation of simulation box length attributed to the N(0,2) coordinate prior and cell prior. Because free energies and densities depend on volume, and because the cell is generated jointly, this systematic bias is load-bearing for the claim of energetically accurate Boltzmann samples; the paper should either correct it (e.g., better prior or post-hoc volume reweighting) or demonstrate that residual volume error does not materially affect the reported AHFE increments and energy distributions.
minor comments (4)
  1. [Fig. 1c, §4.1] Fig. 1c and related text: Clarify wall-clock vs GPU-time accounting when claiming 4–10× speedups, especially under parallel sampling across λ versus HREX communication overhead.
  2. [§3.2, §A.1, §C.3] §3.2 / §A.1: The water-compression scheme (O retained, H reconstructed via projection) is central to scaling; a short quantitative ablation of its effect on solvent bonded/nonbonded energies (beyond the qualitative note in §C.3) would help readers assess the approximation.
  3. [§3.2, §C.1] Notation: τ is used for flow-matching time and t for physical time, but some figure captions and §C.1 still mix t/τ; unify for clarity.
  4. [Table 3, §5] Table 3 and related-work claims of “first”: The comparison is useful; a one-sentence qualification that “first at all-atom explicit-solvent Boltzmann sampling suitable for direct FF+MBAR” would avoid over-reading relative to concurrent trajectory or ensemble models.

Circularity Check

0 steps flagged

No load-bearing circular derivation; AHFE is obtained by standard MBAR on force-field energies of samples from a generative model trained to match MD configurations, not by fitting free energies or defining the target in terms of itself.

full rationale

AquaGen is trained via flow matching to transport a Gaussian prior to MD-sampled configurations conditioned on alchemical λ (Eqs. 7–10, §3.2–3.3), using >1B frames from OpenFF/TIP3P HREX trajectories. AHFE is then computed exactly as in classical alchemical MD: evaluate U(λk) with the same force field on the generated frames and apply the unbiased MBAR estimator (Eqs. 4–6, §2.2, §3.1). The model is never trained on free-energy labels; black-box regressors that are trained on AHFE are shown as separate baselines (Table 1). Evaluation against held-out MD AHFE (and recomputed FreeSolv/CombiSolv under the identical protocol) therefore measures fidelity of the learned ensemble, which is the intended surrogate objective rather than a tautology. Error cancellation along the alchemical path (Fig. 4c, CAE vs AE, §4.4–4.5) is an acknowledged accuracy limitation, not a definitional reduction. No uniqueness theorems, self-citation chains, or ansatzes are load-bearing; the free-energy estimator itself is the external, standard MBAR of Shirts & Chodera. Mild surrogate-to-teacher character is inherent to any Boltzmann generative model and does not meet the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 2 invented entities

The central claim rests on classical statistical mechanics (NPT Boltzmann, alchemical free energies, MBAR), standard force fields and water models, and flow matching as a transport method. Load-bearing free choices include the Gaussian prior variance, cell prior, exponential ODE schedule, water-compression architecture, and the discrete λ path. No new physical particles or forces are postulated; AquaGen is a learned surrogate for existing MD ensembles. Proprietary training data and the assumption that MD under OpenFF/TIP3P is the evaluation target are domain assumptions that limit external falsifiability of the headline numbers.

free parameters (5)
  • Gaussian coordinate prior variance σ² = 2
    Chosen as σ²=2 after ablations of N(0,1) and auto-variance priors; strongly affects path curvature, box-size bias, and AHFE/CAE (Table 2, §4.5, §C.1).
  • Cell prior mean/variance (μ_c, σ_c²) = μ_c=7.84, σ_c²=0.05
    Cubic Gaussian cell prior with μ_c=7.84, σ_c²=0.05 set so the prior box holds ~95% of N(0,2) coordinates (§A.1); drives systematic box-length overestimation (~0.1 Å).
  • Exponential integration schedule α = α=4
    t(u)=(exp(αu)-1)/(exp(α)-1) with α=4 concentrates steps near τ≈0 where angular velocity is high (§C.1); uniform timesteps collapse VDW CAE and AHFE AE to 6.75 kcal/mol.
  • Alchemical λ schedules (20 windows) = 20 λ values; elec first 5, vdW remaining 15
    Fixed OpenFE-style elec-then-vdW schedules (Eqs. 16–17) define the conditional distributions the model must learn; uneven spacing contributes to signed error pattern in Fig. 4c.
  • Model capacity (40M / 80M / 160M) and GNN design knobs = main results 160M; ablations 40M
    Width/depth/horizon and water O-H-H compression are engineering choices that change AHFE vs compute curves (Fig. 1c) and energy decompositions (Fig. 7).
axioms (6)
  • domain assumption NPT MD with the chosen thermostat/barostat ergodically samples the isothermal-isobaric Boltzmann distribution p_NPT (Eq. 2).
    Standard MD premise used throughout §2.1; training data and ground-truth AHFE both rely on it.
  • domain assumption MBAR on reduced energies from overlapping alchemical states yields an unbiased minimum-variance free-energy estimate (Shirts & Chodera).
    §2.2; both MD reference and AquaGen AHFE use the same estimator, so model quality is judged by sample quality under that estimator.
  • domain assumption OpenFF 2.1.1 + TIP3P + stated soft-core/annihilation path define the target physics for AHFE.
    §B; experimental FreeSolv/CombiSolv are secondary; headline accuracy is force-field-relative.
  • ad hoc to paper Linear conditional flow-matching path with velocity target x1−x0 is an adequate transport for joint (coordinates, cell) generation.
    §3.2 Eqs. 9–10; standard FM choice but not forced; prior/path interact with observed box bias and curvature.
  • ad hoc to paper Water molecules may be compressed to a single message-passing node with reconstructed H latents without destroying Boltzmann-relevant energetics.
    §A.1 water compression; enables scale but interacts with lack of rigid-water constraint and bonded solvent energy mismatch (§C.3).
  • domain assumption Vacuum ΔG from classical MD can be combined with generative solvated ΔG via the thermodynamic cycle (Eq. 4).
    §3.1; standard AHFE cycle; generative effort is only on the solvated leg.
invented entities (2)
  • AquaGen (λ-conditional all-atom explicit-solvent flow model) independent evidence
    purpose: Generate uncorrelated configurations approximating alchemical Boltzmann distributions for gray-box free-energy estimation.
    Named system introduced by the paper; independent evidence is empirical match to MD energies/structures and AHFE, not a new physical object.
  • Cumulative Absolute Error (CAE) metric for alchemical pathways no independent evidence
    purpose: Measure total |ΔG_i,i+1| deviation robust to error cancellation that AHFE AE misses.
    Defined in §4.4 Eq. 12; diagnostic construct, not a physical entity; useful but paper-specific.

pith-pipeline@v1.1.0-grok45 · 26296 in / 4311 out tokens · 38064 ms · 2026-07-12T01:57:27.513081+00:00 · methodology

0 comments
read the original abstract

We present AquaGen, the first all-atom, explicit solvent, periodic-boundary-condition-aware generative model that produces molecular configurations from the Boltzmann distribution at a fraction of the cost of molecular dynamics (MD). This is in contrast with existing generative models that remove degrees of freedom by operating on coarse-grained, vacuum, or implicit solvent systems. Operating at this resolution allows for post-processing through force field energy evaluations and MD simulations, and enables the prediction of relevant properties in a gray-box manner (as ensemble averages of potential energy evaluations over generated samples). We demonstrate the utility of this paradigm on absolute hydration free energy (AHFE), producing estimates 4-10x faster and with comparable accuracy to standard GPU-based MD. By generating uncorrelated samples from alchemical Boltzmann distributions, we create more accurate, interpretable, and refinable ensemble predictions with calibrated uncertainty estimates, unlike regression methods which are entirely black-box predictors. Our approach also yields predictable benefits from increasing train- and test-time compute, realized by scaling model size and generating more samples, respectively. We believe that this approach demonstrates the utility of high-resolution ensemble generation for free energy estimation, with future potential to replace MD in tasks such as the prediction of lipophilicity, membrane permeability, or absolute binding free energy (ABFE) -- whose grounding and interpretability may be critical for the development of new drugs and materials.

Figures

Figures reproduced from arXiv: 2607.03513 by Cristian Gabellini, Emmanuel Bengio, Francesco Di Giovanni, Kerstin Klaeser, Nikhil Shenoy, Prudencio Tossou, Sanjeev Raja, Yui Tik Pang.

Figure 1
Figure 1. Figure 1: Overview of AquaGen generative modeling framework and results on absolute hydration free energy (AHFE) estimation. (a) A flow matching model is trained to generate uncorrelated, all-atom, explicit-solvent, PBC-aware configurations x¯ (λk) i ∼ pθ(· | λk), i ∈ [1, N], k ∈ [1, K] which approximate samples from data distributions pdata(· | λk) along an alchemical pathway governed by λk. Potential energy evalua… view at source ↗
Figure 2
Figure 2. Figure 2: Structural and energetic accuracy of samples generated by AquaGen at the fully interacting endpoint (λk, k = 1) (a) True vs generated potential energy for 4 randomly chosen compounds. (b) Various decompositions of compound 4 energy. In clockwise order, starting from top left: total potential energy excluding solvent bonded terms (since the reference simulations are performed with rigid water), solute-only … view at source ↗
Figure 3
Figure 3. Figure 3: Structural and energetic accuracy of all-atom, explicit water samples generated by AquaGen along the alchemical pathway (λk, k ∈ [1, 20]). (a) Time-lagged independent component analysis (tICA) plots of samples from reference MD simulations (contours) and flow matching model (points - darker indicates higher density) across λk, k ∈ {1, 6, 11, 16, 20}. The model samples move across the first tICA component a… view at source ↗
Figure 4
Figure 4. Figure 4: Analysis of uncertainty estimation, generalization across compounds, and error cancellation when using AquaGen for AHFE prediction. (a) The width of bootstrapped 90% confidence intervals, computed from random 32-sample subsets of 128 generated configurations, correlates strongly with the AHFE prediction error, indicating that AquaGen provides well-calibrated uncertainty estimates. (b) AHFE prediction error… view at source ↗
Figure 5
Figure 5. Figure 5: Angular velocity of vθ(x, τ |λ) during integration. The large velocity near t = 0 can be counteracted by an exponential integration schedule for the N(0, 1) and N(0, 2) models. The auto-variance prior reduces the angular velocity near t = 0, but leads to higher angular velocities near t = 1 and higher overall CAE values. The auto-variance prior has lower angular velocity at t = 0 but higher angular velocit… view at source ↗
Figure 6
Figure 6. Figure 6: tICA decomposition over the alchemical trajectories. AquaGen samples track the change in principal components as λ evolves. C.3 Additional energy decompositions at fully-interacting distribution. In [PITH_FULL_IMAGE:figures/full_fig_p019_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Energy decompositions at the fully-interacting endpoint (λk, k = 1). For each compound and energy component, we plot the reference MD distribution (gray), the distribution from the 40M AquaGen model (red), and the distributon from the 160M AquaGen model (blue). C.4 Error cancellation along the alchemical pathway The free energy difference between the alchemical endpoints can be written as a sum over the fi… view at source ↗
Figure 8
Figure 8. Figure 8: plots the correlation between AHFE Mean AE and Tanimoto similarity for the regression baselines. The baselines exhibit a stronger negative correlation than AquaGen (Figure 4b), suggesting they may be less reliable in out-of-distribution settings. 0.2 0.3 0.4 0.5 0.6 Max Tanimoto similarity to training set 0 2 4 6 8 10 12 14 AHFE error (kcal/mol) Vacuum best fit, R²: 0.04 p=0.00 0.2 0.3 0.4 0.5 0.6 Max Tani… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

63 extracted references · 4 linked inside Pith

  1. [1]

    arXiv preprint arXiv:2401.04246 , year=

    Scalable normalizing flows enable boltzmann generators for macromolecules , author=. arXiv preprint arXiv:2401.04246 , year=

    Pith/arXiv arXiv

  2. [2]

    Advances in Neural Information Processing Systems , volume=

    Generative modeling of molecular dynamics trajectories , author=. Advances in Neural Information Processing Systems , volume=

  3. [3]

    Highly accurate protein structure prediction with

    Jumper, John and Evans, Richard and Pritzel, Alexander and Green, Tim and Figurnov, Michael and Ronneberger, Olaf and Tunyasuvunakool, Kathryn and Bates, Russ and. Highly accurate protein structure prediction with. nature , volume=. 2021 , publisher=

    2021

  4. [4]

    Accurate structure prediction of biomolecular interactions with

    Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans, Richard and Green, Tim and Pritzel, Alexander and Ronneberger, Olaf and Willmore, Lindsay and Ballard, Andrew J and Bambrick, Joshua and others , journal=. Accurate structure prediction of biomolecular interactions with. 2024 , publisher=

    2024

  5. [5]

    Nature Machine Intelligence , volume=

    Predicting equilibrium distributions for molecular systems with deep learning , author=. Nature Machine Intelligence , volume=. 2024 , publisher=

    2024

  6. [6]

    Neumann, Mark and Gin, James and Rhodes, Benjamin and Bennett, Steven and Li, Zhiyi and Choubisa, Hitarth and Hussey, Arthur and Godwin, Jonathan , journal=. Orb:

  7. [7]

    Boltzmann generators:

    No. Boltzmann generators:. Science , volume=. 2019 , publisher=

    2019

  8. [8]

    The Journal of Chemical Physics , volume=

    Targeted free energy perturbation with machine learning , author=. The Journal of Chemical Physics , volume=. 2020 , publisher=

    2020

  9. [9]

    International Conference on Learning Representations , year=

    Flow Matching for Generative Modeling , author=. International Conference on Learning Representations , year=

  10. [10]

    Advances in Neural Information Processing Systems , year=

    Equivariant Flow Matching , author=. Advances in Neural Information Processing Systems , year=

  11. [11]

    St. Equi. International Conference on Machine Learning , pages=. 2022 , organization=

    2022

  12. [12]

    Advances in Neural Information Processing Systems , volume=

    Hassan, Majdi and Shenoy, Nikhil and Lee, Jungyoon and St. Advances in Neural Information Processing Systems , volume=

  13. [13]

    Boltz-2:

    Passaro, Saro and Corso, Gabriele and Wohlwend, Jeremy and Reveiz, Mateo and Thaler, Stephan and Somnath, Vignesh Ram and Getz, Noah and Portnoi, Tally and Roy, Julien and Stark, Hannes and others , journal=. Boltz-2:

  14. [14]

    Science , volume=

    Scalable emulation of protein equilibrium ensembles with generative deep learning , author=. Science , volume=. 2025 , publisher=

    2025

  15. [15]

    The Journal of Chemical Physics , volume=

    Statistically optimal analysis of samples from multiple equilibrium states , author=. The Journal of Chemical Physics , volume=. 2008 , publisher=

    2008

  16. [16]

    Journal of chemical information and modeling , volume=

    Extended-connectivity fingerprints , author=. Journal of chemical information and modeling , volume=. 2010 , publisher=

    2010

  17. [17]

    Neural thermodynamic integration:

    Mate, Balint and Fleuret, Francois and Bereau, Tristan , journal=. Neural thermodynamic integration:. 2024 , publisher=

    2024

  18. [18]

    The Journal of Chemical Physics , volume=

    Solvation free energies from neural thermodynamic integration , author=. The Journal of Chemical Physics , volume=. 2025 , publisher=

    2025

  19. [19]

    arXiv preprint arXiv:2504.11516 , year=

    He, Jiajun and Du, Yuanqi and Vargas, Francisco and Wang, Yuanqing and Gomes, Carla P and Hern. arXiv preprint arXiv:2504.11516 , year=

    arXiv

  20. [20]

    Thaler, Stephan and Wu, Zhiyi and Glass, William G and Bradshaw, Richard T and Bartlett, Gail and Tossou, Prudencio and Wood, Geoffrey PF , journal=. Boltz-. 2026 , publisher=

    2026

  21. [21]

    How well does

    Tan, Chunhu and Yang, Lijiang and Luo, Ray , journal=. How well does. 2006 , publisher=

    2006

  22. [22]

    arXiv preprint arXiv:2603.17633 , year=

    Atomic Trajectory Modeling with State Space Models for Biomolecular Dynamics , author=. arXiv preprint arXiv:2603.17633 , year=

    arXiv

  23. [23]

    Jing, Bowen and Berger, Bonnie and Jaakkola, Tommi , journal=. Alpha

  24. [24]

    All-atom diffusion transformers:

    Joshi, Chaitanya K and Fu, Xiang and Liao, Yi-Lun and Gharakhanyan, Vahe and Miller, Benjamin Kurt and Sriram, Anuroop and Ulissi, Zachary W , journal=. All-atom diffusion transformers:

  25. [25]

    Advances in Neural Information Processing Systems , volume=

    Guiding a diffusion model with a bad version of itself , author=. Advances in Neural Information Processing Systems , volume=

  26. [26]

    arXiv preprint arXiv:2207.12598 , year=

    Classifier-free diffusion guidance , author=. arXiv preprint arXiv:2207.12598 , year=

    Pith/arXiv arXiv

  27. [27]

    Adjoint matching:

    Domingo-Enrich, Carles and Drozdzal, Michal and Karrer, Brian and Chen, Ricky TQ , journal=. Adjoint matching:

  28. [28]

    arXiv preprint arXiv:2501.06848 , year=

    A general framework for inference-time scaling and steering of diffusion models , author=. arXiv preprint arXiv:2501.06848 , year=

    Pith/arXiv arXiv

  29. [29]

    High-temperature equation of state by a perturbation method

    Zwanzig, Robert W , journal=. High-temperature equation of state by a perturbation method. 1954 , publisher=

    1954

  30. [30]

    Efficient estimation of free energy differences from

    Bennett, Charles H , journal=. Efficient estimation of free energy differences from. 1976 , publisher=

    1976

  31. [31]

    Nonphysical sampling distributions in

    Torrie, Glenn M and Valleau, John P , journal=. Nonphysical sampling distributions in. 1977 , publisher=

    1977

  32. [32]

    Exchange

    Hukushima, Koji and Nemoto, Koji , journal=. Exchange. 1996 , publisher=

    1996

  33. [33]

    Understanding Molecular Simulation:

    Frenkel, Daan and Smit, Berend , year=. Understanding Molecular Simulation:

  34. [34]

    Science , volume=

    How fast-folding proteins fold , author=. Science , volume=. 2011 , publisher=

    2011

  35. [35]

    Communications of the ACM , volume=

    Anton, a special-purpose machine for molecular dynamics simulation , author=. Communications of the ACM , volume=. 2008 , publisher=

    2008

  36. [36]

    Journal of the American Chemical Society , volume=

    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 , author=. Journal of the American Chemical Society , volume=. 2015 , publisher=

    2015

  37. [37]

    Science , volume=

    Atomic-level characterization of the structural dynamics of proteins , author=. Science , volume=. 2010 , publisher=

    2010

  38. [38]

    Alchemical free energy calculations:

    Shirts, Michael R and Mobley, David L and Chodera, John D , journal=. Alchemical free energy calculations:. 2007 , publisher=

    2007

  39. [39]

    Chithrananda, Seyone and Grand, Gabriel and Ramsundar, Bharath , journal=. Chem

  40. [40]

    Ahmad, Walid and Simon, Elana and Chithrananda, Seyone and Grand, Gabriel and Ramsundar, Bharath , journal=. Chem

  41. [41]

    Chemprop:

    Heid, Esther and Greenman, Kevin P and Chung, Yunsie and Li, Shih-Cheng and Graff, David E and Vermeire, Florence H and Wu, Haoyang and Green, William H and McGill, Charles J , journal=. Chemprop:. 2023 , publisher=

    2023

  42. [42]

    Approaches for calculating solvation free energies and enthalpies demonstrated with an update of the

    Duarte Ramos Matos, Guilherme and Kyu, Daisy Y and Loeffler, Hannes H and Chodera, John D and Shirts, Michael R and Mobley, David L , journal=. Approaches for calculating solvation free energies and enthalpies demonstrated with an update of the. 2017 , publisher=

    2017

  43. [43]

    Mobley, David L and Guthrie, J Peter , journal=. Free. 2014 , publisher=

    2014

  44. [44]

    Chemical Engineering Journal , volume=

    Transfer learning for solvation free energies: From quantum chemistry to experiments , author=. Chemical Engineering Journal , volume=. 2021 , publisher=

    2021

  45. [45]

    Kapu. Mar. International Conference on Learning Representations (ICLR) , year=

  46. [46]

    2007 , publisher=

    Free energy calculations , author=. 2007 , publisher=

    2007

  47. [47]

    Rhodes, Benjamin and Vandenhaute, Sander and. Orb-v3:. arXiv preprint arXiv:2504.06231 , year=

    Pith/arXiv arXiv

  48. [48]

    Advances in neural information processing systems , volume=

    Uncertainty quantification and deep ensembles , author=. Advances in neural information processing systems , volume=

  49. [49]

    Calculation of protein-ligand binding affinities , author=. Annu. Rev. Biophys. Biomol. Struct. , volume=. 2007 , publisher=

    2007

  50. [50]

    Briefings in bioinformatics , volume=

    Prediction of protein-ligand binding affinity via deep learning models , author=. Briefings in bioinformatics , volume=. 2024 , publisher=

    2024

  51. [51]

    Relative binding free energy calculations in drug discovery:

    Cournia, Zoe and Allen, Bryce and Sherman, Woody , journal=. Relative binding free energy calculations in drug discovery:. 2017 , publisher=

    2017

  52. [52]

    ACS Medicinal Chemistry Letters , volume=

    Recent advances in alchemical binding free energy calculations for drug discovery , author=. ACS Medicinal Chemistry Letters , volume=. 2023 , publisher=

    2023

  53. [53]

    Advances in neural information processing systems , volume=

    Understanding and improving layer normalization , author=. Advances in neural information processing systems , volume=

  54. [54]

    Mathematical correlations for describing solute transfer into functionalized alkane solvents containing hydroxyl, ether, ester or ketone solvents

    Grubbs, Laura M and Saifullah, Mariam and De La Rosa, Nohelli E and Ye, Shulin and Achi, Sai S and Acree, Jr, William E and Abraham, Michael H. Mathematical correlations for describing solute transfer into functionalized alkane solvents containing hydroxyl, ether, ester or ketone solvents. Fluid Phase Equilib

  55. [55]

    Minnesota Solvation Database ( MNSOL ) version 2012

    Marenich, Aleksandr V and Kelly, Casey P and Thompson, Jason D and Hawkins, Gregory D and Chambers, Candy C and Giesen, David J and Winget, Paul and Cramer, Christopher J and Truhlar, Donald G. Minnesota Solvation Database ( MNSOL ) version 2012

    2012

  56. [56]

    Estimation of solvation quantities from experimental thermodynamic data: Development of the comprehensive CompSol databank for pure and mixed solutes

    Moine, Edouard and Privat, Romain and Sirjean, Baptiste and Jaubert, Jean-No \"e l. Estimation of solvation quantities from experimental thermodynamic data: Development of the comprehensive CompSol databank for pure and mixed solutes. J. Phys. Chem. Ref. Data

  57. [57]

    Large-scale assessment of binding free energy calculations in active drug discovery projects

    Schindler, Christina E M and Baumann, Hannah and Blum, Andreas and B \"o se, Dietrich and Buchstaller, Hans-Peter and Burgdorf, Lars and Cappel, Daniel and Chekler, Eugene and Czodrowski, Paul and Dorsch, Dieter and Eguida, Merveille K I and Follows, Bruce and Fuch , Thomas and Gr \"a dler, Ulrich and Gunera, Jakub and Johnson, Theresa and Jorand Lebrun, ...

  58. [58]

    Accuracy comparison of several common implicit solvent models and their implementations in the context of protein-ligand binding

    Katkova, E V and Onufriev, A V and Aguilar, B and Sulimov, V B. Accuracy comparison of several common implicit solvent models and their implementations in the context of protein-ligand binding. J. Mol. Graph. Model

  59. [59]

    Are AMBER force fields and implicit solvation models additive? A folding study with a balanced peptide test set

    Robinson, Melina K and Monroe, Jacob I and Shell, M Scott. Are AMBER force fields and implicit solvation models additive? A folding study with a balanced peptide test set. J. Chem. Theory Comput

  60. [60]

    A modified TIP3P water potential for simulation with Ewald summation

    Price, Daniel J and Brooks, 3rd, Charles L. A modified TIP3P water potential for simulation with Ewald summation. J. Chem. Phys

  61. [61]

    The Journal of chemical physics , volume=

    Statistical mechanics of fluid mixtures , author=. The Journal of chemical physics , volume=. 1935 , publisher=

    1935

  62. [62]

    2022 , howpublished =

    Absolute Solvation Protocol , author =. 2022 , howpublished =

    2022

  63. [63]

    The Open Free Energy library

    Alibay, Irfan and Gowers, Richard J and Swenson, David W H and Henry, Michael M and Ries, Benjamin and Baumann, Hannah M and Eastwood, James R B and Mitchell, Ashley and Dotson, David and Horton, Joshua T and Thompson, Matthew and Travitz, Alyssa. The Open Free Energy library