Pith. sign in

REVIEW 3 major objections 5 minor 79 references

Navigating Chemical Space: Multi-Level Bayesian Optimization with Hierarchical Coarse-Graining

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Multi-level Bayesian optimization over hierarchically coarse-grained chemical spaces outperforms single-resolution Bayesian optimization for free-energy-based molecular discovery.

desk verdict A real integration of hierarchical CG with multi-level BO, reproducible and well-demonstrated on a membrane design problem, but the headline advantage over standard BO rests on a single run and an unmeasured delta-learning premise. read the letter →

arxiv 2505.04169 v2 pith:PNCXTFO6 submitted 2025-05-07 physics.chem-ph

Luis J. Walter , Tristan Bereau This is my paper · ORCID
classification physics.chem-ph
keywords Bayesianoptimizationcoarse-grainingchemicalspaceexplorationfree-energydifferenceslipidbilayerphaseseparationmulti-fidelitylatentencodingactivelearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes a funnel-like strategy: enumerate chemical space at three coarse-grained resolutions (15, 45, and 96 bead types, spanning about 90,000 to 137 million molecules), embed each resolution in a smooth learned latent space, and run Bayesian optimization that starts at low resolution and uses lower-resolution Gaussian-process predictions as priors for higher-resolution searches. The claim is that this multi-level procedure beats standard Bayesian optimization performed only at the highest resolution, finding better candidates and a better distribution of candidates while evaluating fewer than 330 molecules. The demonstration optimizes small molecules that promote phase separation in a ternary lipid bilayer, scored by free-energy differences from molecular dynamics simulations. If true, the method offers a route to navigate very large chemical spaces when the objective is a free-energy difference, with interpretable design rules as a by-product.

What carries the argument

The load-bearing mechanism is delta learning over a hierarchy of coarse-grained resolutions. Three levels share the same atom-to-bead mapping but use 15, 45, and 96 transferable bead types, so every high-resolution molecule maps to a unique lower-resolution molecule. Each resolution's chemical space is embedded separately by a regularized graph autoencoder into a five-dimensional latent space, and a many-to-one mapping $M_l$ transfers lower-resolution predictions into higher-resolution Gaussian processes. The model at level $l$ is $f_l(x) \sim \mathcal{GP}(f_{l-1}(M_l(x)), k_l(x,x'))$, with expected improvement maximized only inside neighborhoods of promising lower-resolution points. Resolution switching is triggered when the GP prediction error stays below 0.12 kcal/mol for three consecutive evaluations, and switching back occurs when the best candidate lies farther than $2\xi_l$ from all evaluated points.

What would settle it

Compute $\Delta\Delta G$ at high resolution for roughly 50 molecules already evaluated at low resolution and measure the rank correlation between the two sets of values; if the correlation is near zero, the lower-resolution prior cannot carry the search.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that coupling hierarchical coarse-graining to Bayesian optimization turns chemical-space resolution into a controllable exploration-exploitation dial. Each level's function $f_l$ is modeled as the next-lower function plus a Gaussian-process correction, $f_l(x) = f_{l-1}(M_l(x)) + \delta_l(x)$, so low-resolution evaluations map out broad basins while higher-resolution evaluations refine them. In the bilayer application the algorithm evaluated 327 molecules, less than $3\times10^{-4}\%$ of the 137-million-molecule high-resolution space, shifted the distribution of $\Delta\Delta G$ values steadily downward, and produced top candidates with $\Delta\Delta G \le -1.3$ kcal/mol, all composed of hydrophobic C4, C5, and C6 bead types. A direct 1200 ns validation simulation showed that the best candidate reduced DPPC-DLiPC contacts more than benzene, a known demixing agent. The paper takes this as evidence that multi-level BO outperforms single-resolution BO on both the best value and the spread of good candidates.

Load-bearing premise

The funnel works only if a molecule's coarse-grained free-energy value is a sufficiently accurate guide to its finer-grained value, so the delta corrections stay small; the paper never directly measures that correspondence for identical molecules.

Editorial extensions

If this is right

  • A search that would require screening a 137-million-molecule high-resolution space can instead run a few hundred evaluations across three resolutions and still find multiple strong candidates.
  • Lower-resolution neighborhoods, each containing roughly 250 molecules, map to neighborhoods of about 378,000 high-resolution molecules, so a modest number of coarse evaluations can guide a large fraction of the fine-grained search.
  • The workflow returns interpretable chemical rules, here that hydrophobic C4, C5, and C6 beads in mixed sizes promote demixing, alongside the optimized molecules.
  • The same funnel applies to any molecular objective expressible as a free-energy difference, without requiring pretraining data for the target.
  • Because the switching thresholds are tied to the chemical space rather than to the specific application, the hyperparameters should transfer to other molecular optimization tasks.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural next test, beyond the paper's analysis, would be an explicit check of the delta-correction assumption: computing high-resolution $\Delta\Delta G$ for a sample of molecules already evaluated at low resolution would show whether low-resolution values actually predict high-resolution values for identical molecules.
  • If the delta assumption holds, the same hierarchical funnel could be applied to other expensive molecular properties, such as binding free energies or solvation properties, with the cheap proxy prior replaced accordingly.
  • The bilayer comparison to standard BO rests on single runs; the paper's toy model suggests the advantage is reproducible on average, but repeated runs on a cheaper surrogate that preserves the three-resolution structure would make the real-system claim more decisive.
  • The neighborhood restriction uses the GP lengthscale as a fixed similarity radius; a testable extension would be to let that radius adapt per region, since the chemical space is likely not uniformly smooth.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript presents a multi-level Bayesian optimization framework for small-molecule discovery. Chemical space is represented at three coarse-grained resolutions with 15/45/96 Martini3-derived bead types; each space is embedded with a graph-neural-network regularized autoencoder. A delta-learning Gaussian process (Eq. 5) propagates lower-resolution free-energy information upward, with the lowest level initialized by a bead-additivity prior (Eq. SI3). The method is demonstrated by minimizing a free-energy proxy for phase separation in DPPC/DLiPC/cholesterol bilayers. The authors report that the multi-level funnel uses 327 evaluations to outperform standard high-resolution BO, finds candidates with ΔΔG below −1.3 kcal/mol, and validates the best candidate with direct contact simulations. They also argue that lower-resolution chemical neighborhoods are larger, implying smoother landscapes.

Significance. If the efficiency claim holds, the paper is a valuable contribution: it combines transferable CG models and active learning in a way that is broadly applicable to free-energy-based molecular optimization. Strengths include the release of code, models, and data; the independent validation of the top candidate with a distinct observable (DPPC–DLiPC contacts); a multi-run toy-model comparison that partially addresses initialization effects; and the identification of interpretable chemical design rules. However, the central efficiency claim rests on a single real-system run and on an unvalidated delta-learning premise, so the quantitative advantage over single-resolution BO should be regarded as conditional rather than established.

major comments (3)
  1. [Section III D, Eq. (5)] The evidence for the central premise that lower-resolution landscapes are informative priors is not direct. The neighborhood-size analysis in Section III D derives ξ_l by fitting RBF kernels to the evaluated molecules and then reports mapped neighborhood counts of 249 → 18,700 → 378,000. As shown in SI2.6, these counts are obtained by multiplying the low-resolution neighborhood size by the average many-to-one multiplicities (75 and 20), so the growth tracks the total number of molecules per resolution (9.0×10^4, 6.7×10^6, 1.37×10^8) rather than a measured smoothness gradient. The paper never reports the distribution of δ_l(x) (Eq. 5) or the correlation between lower-resolution predictions and high-resolution observations for the same molecule. Without such a test, the restriction of EI maximization to neighborhoods of low-resolution favorites (SI1.4) could be excluding the high-resolution optimum, and the Fig. 9 advantage cannot be attributed to the hierarchy rather than to search restriction.
  2. [Section III C, Fig. 9] The headline comparison is a single run of each algorithm. The text acknowledges that averaging over multiple runs is 'computationally infeasible', but the resulting curves have no error bars, and standard BO is known to be initialization-sensitive. The toy-model average in SI2.7 uses a different synthetic score and cannot by itself establish the real-system comparison. I request either several independent standard-BO runs on the bilayer system at reduced cost, a statistical treatment of the single-run comparison, or a clear reframing of the claim as a case study rather than a general outperformance result.
  3. [Section III B / Section II F] The free-energy proxy ΔΔG is validated as a predictor of phase separation on only one molecule. The 1600 ns contact simulation shows the top candidate outperforms benzene and the control, but it does not test whether the ΔΔG ranking across the discovered candidates (Figs. 5 and 9) reflects demixing propensity. Given that the distribution-level claim in Section III C is based entirely on ΔΔG values, an additional validation on a non-top candidate, or on a small set spanning the observed range, would materially strengthen the claim that the method navigates chemical space toward the target property rather than toward an uncorrelated proxy.
minor comments (5)
  1. [Section II F and SI Eq. SI2] The main text says S is defined as a conditional weighted sum of ΔGwater−ΔGinterface and ΔGinterface−ΔGcenter, but Eq. SI2 uses conditions involving ΔGwater−ΔGcenter and ΔGinterface−ΔGcenter; please reconcile the description and the formula.
  2. [Eq. (8)] The mapping M_l is defined via discrete molecular mapping between bead-type levels, but the GP is formulated on continuous latent spaces; please clarify how M_l(x) is evaluated for latent points that do not correspond exactly to an enumerated molecule during EI maximization.
  3. [Table SI2.1 and Section SI2.6] The label 'DIPC' should be 'DLiPC' in Table SI2.1, and the sentence in SI2.6 explaining the multiplication by 75 and 20 should be more explicit that these are enumeration-count ratios rather than measurements of landscape correlation.
  4. [Figure 9] The x-axis should state explicitly whether it is the total number of MD evaluations including initialization and prior evaluations for both methods; the current caption is ambiguous.
  5. [SI Eq. SI3] Two displayed equations in SI2.2 contain raw LaTeX markup; they should be typeset properly.

Circularity Check

1 steps flagged · score 2.0 of 10

Central BO result is self-contained against MD ground truth; only the Section III D 'neighborhood size' confirmation restates fitted lengthscales.

  1. fitted input called prediction [Section III D (Chemical Neighborhood Sizes Across Resolutions); Figure 10; SI2.6; Conclusions]
    "To test this, we introduce the concept of chemical neighborhoods and analyze their sizes across different resolution levels. ... Here, neighborhood size is determined by the lengthscale ξl of an RBF kernel fitted in a GP regression. ... These results support our assumption of a smoother free-energy landscape at lower resolutions."

    The neighborhood size is defined as the average number of molecules within a distance d = αξl, where ξl is obtained by fitting an RBF GP to the evaluated molecules at that level. The reported sizes (249, 23, 37) and their mapped multiples (18,700; 378,000) are therefore deterministic functions of the fitted lengthscales and the known many-to-one molecule counts, not independent measurements of cross-resolution predictivity. Presenting this as a confirmation of the smoothness assumption restates the fitted lengthscale ordering; it does not test whether the delta term δl(x) in Eq. (5) is small, nor whether lower-resolution priors predict high-resolution ΔΔG values for held-out molecules. The evidence is an internal consistency check rather than an independent validation.

full rationale

The paper's central derivation chain is not circular. The optimization target ΔΔG is obtained from MD-based thermodynamic integration, and the final top candidate is validated by an independent observable: direct DPPC-DLiPC contact simulations compared with benzene and a no-solute control. Equation (5), fl(x) = fl−1(Ml(x)) + δl(x), is an explicit multi-fidelity modeling assumption (cited to Huang et al., not to the authors' own work) rather than a derived prediction, and the reported advantage over standard BO is an empirical comparison, albeit from a single run. The self-citations to Mohr et al. and Centi et al. supply architecture and scoring-proxy context, not the load-bearing conclusion. The only mild circular content is Section III D: the 'chemical neighborhood' sizes are computed directly from RBF lengthscales fitted to the same evaluated molecules, so the conclusion that lower-resolution landscapes are smoother is a restatement of the fitted kernel parameters, not an independent test of the cross-resolution delta-learning premise.

Assumptions & free parameters 8 free parameters · 6 assumptions · 0 invented entities

The method is essentially an empirical optimization, so the MD free-energy values are ground-truth data. The burden of assumption sits in the priors: the delta-learning model, the bead-additivity prior with fitted beta, and the empirical score S. No new physical entities are postulated.

free parameters (8)
  • beta (bead-additivity scaling) = 0.63
    Fitted by least squares to the 50 initialization molecules to scale the summed single-bead free energies in the lowest-resolution prior (SI Eq. SI3, SI Section 2.2).
  • GP prediction-error threshold = 0.12 kcal/mol
    Empirically chosen switch criterion for advancing to the next resolution; requires three consecutive evaluations below threshold (Section II E).
  • Number of consecutive low-error evaluations = 3
    Part of the switch condition; set empirically to balance exploration/exploitation (Section II E).
  • m (neighborhood top candidates) = 30
    Number of best lower-resolution molecules used to define the searchable neighborhood at the next level (SI Section 1.4).
  • alpha (neighborhood size factor) = 0.5
    Defines a chemical neighborhood as points within d = alpha * xi_l; used to compute neighborhood sizes and smoothness claims (SI Section 2.6).
  • RBF lengthscales xi_l per level = fitted via GP marginal likelihood
    Kernel lengthscales for each resolution's delta GP; fitted to the actively collected data during optimization and reused in the neighborhood analysis (Section II E, SI 2.6).
  • GP noise sigma_n = 0.05 kcal/mol
    Fixed from duplicate free-energy calculations of 14 molecules; treated as known noise in the GP (SI Section 2.3).
  • Score S scaling constants = 0.5, 25, 3, exponents 2 and 0.5
    Empirical scaling and thresholds in the penalty score that guides interface/water-localizing molecules (SI Eq. SI2).
assumptions (6)
  • domain assumption Higher-resolution free-energy functions are a delta-correction to lower-resolution ones: f_l(x) = f_{l-1}(x) + delta_l(x) (Eq. 5).
    The funnel's efficiency rests on this delta-learning model from Huang et al.; if the lower-resolution function is not a good mean for the higher-resolution function, the prior misleads the search (Section II E).
  • ad hoc to paper The lowest-resolution prior is the sum of individual bead free energies scaled by beta (Eq. SI3).
    Used to start BO at l=1; the additivity assumption ignores bead correlations and is a fitted approximation (SI Section 2.2).
  • domain assumption The free-energy proxy min(DeltaDeltaG,0) + S correlates with the actual DPPC/DLiPC demixing tendency.
    The optimization objective is a proxy, and the connection to true phase separation is validated for only one molecule (Section II F, Section III B).
  • domain assumption Martini3-derived CG force fields (without bead labels) faithfully represent small-molecule interactions relevant to bilayer demixing.
    All ground-truth free energies come from these CG simulations; errors in the force field propagate directly to the objective (Section II G).
  • domain assumption The RAE-learned latent spaces are smooth enough that RBF-kernel distances encode meaningful chemical similarity for the free-energy target.
    BO is performed in the learned five-dimensional spaces; reconstruction accuracies are 0.98-0.99 but the target-relevant smoothness is not directly measured (Section II C, SI Table SI1.3).
  • domain assumption Molecules with up to four CG beads and fixed bond lengths, without angle or dihedral terms, adequately cover the relevant small-molecule chemical space.
    The enumeration of chemical space relies on these simplifications (SI Section 1.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Navigating Chemical Space: Multi-Level Bayesian Optimization with Hierarchical Coarse-Graining." pith.science (2026). https://pith.science/paper/PNCXTFO6

@misc{pith2026250504169,
  author       = {Pith},
  title        = {Pith review of: Navigating Chemical Space: Multi-Level Bayesian Optimization with Hierarchical Coarse-Graining},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PNCXTFO6}},
  note         = {Machine review of arXiv:2505.04169}
}
read the original abstract

Molecular discovery within the vast chemical space remains a significant challenge due to the immense number of possible molecules and limited scalability of conventional screening methods. To approach chemical space exploration more effectively, we have developed an active learning-based method that uses transferable coarse-grained models to compress chemical space into varying levels of resolution. By using multiple representations of chemical space with different coarse-graining resolutions, we balance combinatorial complexity and chemical detail. To identify target compounds, we first transform the discrete molecular spaces into smooth latent representations. We then perform Bayesian optimization within these latent spaces, using molecular dynamics simulations to calculate target free energies of the coarse-grained compounds. This multi-level approach effectively balances exploration and exploitation at lower and higher resolutions, respectively. We demonstrate the effectiveness of our method by optimizing molecules to enhance phase separation in phospholipid bilayers. Our funnel-like strategy not only suggests optimal compounds but also provides insight into relevant neighborhoods in chemical space. We show how this neighborhood information from lower resolutions can guide the optimization at higher resolutions, thereby providing an efficient way to navigate large chemical spaces for free energy-based molecular optimization.

Figures

Figures reproduced from arXiv: 2505.04169 by the authors.

Figure 1
Figure 1. FIG. 1. Overview of our multi-resolution coarse-graining molecule optimization workflow. (a) Definition of multiple coarse [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Influencing phase separation in a lipid bilayer by [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Estimating the demixing behavior of molecules via [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Encoded chemical spaces and evaluated points for the three levels of resolution. The full chemical spaces are shown [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Distribution of ∆∆ [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. CG structures of the best eight high-resolution [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 9
Figure 9. Figure 9: FIG. 9. Progression of the ∆∆ [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 8
Figure 8. Figure 8: FIG. 8. Top ten most influential molecular features contribut [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]
Figure 10
Figure 10. Figure 10: FIG. 10. Chemical neighborhood sizes across different CG [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 61 canonical work pages

  1. [1]

    Kirkpatrick and C

    P. Kirkpatrick and C. Ellis, Nature 432, 823–823 (2004)

    work page 2004

  2. [2]

    Reymond, Acc

    J.-L. Reymond, Acc. Chem. Res. 48, 722–730 (2015)

    work page 2015

  3. [3]

    P. G. Polishchuk, T. I. Madzhidov, and A. Varnek, J. Comput.-Aided Mol. Des. 27, 675–679 (2013)

    work page 2013

  4. [4]

    Mishra, L

    K. Mishra, L. Ganju, M. Sairam, P. Banerjee, and R. Sawhney, Biomed. Pharmacother. 62, 94–98 (2008)

    work page 2008

  5. [5]

    Macarron, M

    R. Macarron, M. N. Banks, D. Bojanic, D. J. Burns, D. A. Cirovic, T. Garyantes, D. V. S. Green, R. P. Hertzberg, W. P. Janzen, J. W. Paslay, U. Schopfer, and G. S. Sittampalam, Nat. Rev. Drug Discovery 10, 188–195 (2011)

    work page 2011

  6. [6]

    D. C. Fara, T. I. Oprea, E. R. Prossnitz, C. G. Bologa, B. S. Edwards, and L. A. Sklar, Drug Discovery Today: Technol. 3, 377–385 (2006)

    work page 2006

  7. [7]

    Karplus and G

    M. Karplus and G. A. Petsko, Nature 347, 631–639 (1990)

    work page 1990

  8. [8]

    Hansson, C

    T. Hansson, C. Oostenbrink, and W. van Gunsteren, Curr. Opin. Struct. Biol. 12, 190–196 (2002)

    work page 2002

Show all 79 references
  1. [9]

    Bereau, Modell

    T. Bereau, Modell. Simul. Mater. Sci. Eng. 29, 023001 (2021)

    2021

  2. [10]

    Stanley and G

    N. Stanley and G. De Fabritiis, In Silico Pharmacol. 3, 10.1186/s40203-015-0007-0 (2015)

    2015 doi

  3. [11]

    H. J. Kushner, J. Basic Eng. 86, 97–106 (1964)

    1964

  4. [12]

    P. I. Frazier, A tutorial on bayesian optimization (2018)

    2018

  5. [13]

    Agarwal, H

    G. Agarwal, H. A. Doan, L. A. Robertson, L. Zhang, and R. S. Assary, Chem. Mater. 33, 8133–8144 (2021)

    2021

  6. [14]

    Thompson, W

    J. Thompson, W. P. Walters, J. A. Feng, N. A. Pabon, H. Xu, M. Maser, B. B. Goldman, D. Moustakas, M. Schmidt, and F. York, Artif. Intell. Life Sci.2, 100050 (2022)

    2022

  7. [15]

    J. E. Crivelli-Decker, Z. Beckwith, G. Tom, L. Le, S. Khuttan, R. Salomon-Ferrer, J. Beall, R. G´ omez- Bombarelli, and A. Bortolato, J. Chem. Theory Comput. 10.1021/acs.jctc.4c00399 (2024)

    2024 doi

  8. [16]

    T. N. Kipf and M. Welling, Variational graph auto- encoders (2016), arXiv:1611.07308

    2016 arXiv

  9. [17]

    G´ omez-Bombarelli, J

    R. G´ omez-Bombarelli, J. N. Wei, D. Duvenaud, J. M. Hern´ andez-Lobato, B. S´ anchez-Lengeling, D. Sheberla, J. Aguilera-Iparraguirre, T. D. Hirzel, R. P. Adams, and A. Aspuru-Guzik, ACS Cent. Sci. 4, 268–276 (2018)

    2018

  10. [18]

    Reiser, M

    P. Reiser, M. Neubert, A. Eberhard, L. Torresi, C. Zhou, C. Shao, H. Metni, C. van Hoesel, H. Schopmans, T. Sommer, and P. Friederich, Commun. Mater. 3, 10.1038/s43246-022-00315-6 (2022). 12

    2022 doi

  11. [19]

    H. L. Morgan, J. Chem. Doc. 5, 107–113 (1965)

    1965

  12. [20]

    M. K. Warmuth, J. Liao, G. R¨ atsch, M. Mathieson, S. Putta, and C. Lemmen, J. Chem. Inf. Comput. Sci. 43, 667–673 (2003)

    2003

  13. [21]

    Muegge and P

    I. Muegge and P. Mukherjee, Expert Opin. Drug Discov- ery 11, 137–148 (2015)

    2015

  14. [22]

    Gorantla, A

    R. Gorantla, A. Kubincov´ a, B. Suutari, B. P. Cossins, and A. S. J. S. Mey, J. Chem. Inf. Model. 64, 1955–1965 (2024)

    2024

  15. [23]

    W. G. Noid, J. Chem. Phys. 139, 10.1063/1.4818908 (2013)

    2013 doi

  16. [25]

    K. H. Kanekal and T. Bereau, J. Chem. Phys. 151, 10.1063/1.5119101 (2019)

    2019 doi

  17. [27]

    Huang, T

    D. Huang, T. T. Allen, W. I. Notz, and R. A. Miller, Struct. Multidiscip. Optim. 32, 369–382 (2006)

    2006

  18. [28]

    C. Fare, P. Fenner, M. Benatan, A. Varsi, and E. O. Pyzer-Knapp, npj Comput. Mater. 8, 10.1038/s41524- 022-00947-9 (2022)

    2022 doi

  19. [29]

    Mikkola, J

    P. Mikkola, J. Martinelli, L. Filstroff, and S. Kaski, in The 26th International Conference on Artificial Intelli- gence and Statistics , Vol. 206 (Valencia, Spain, 2023) pp. 7425–7454

    2023

  20. [30]

    Gantzler, A

    N. Gantzler, A. Deshwal, J. R. Doppa, and C. M. Simon, Digital Discovery 2, 1937–1956 (2023)

    2023

  21. [31]

    Sanchez-Lengeling and A

    B. Sanchez-Lengeling and A. Aspuru-Guzik, Science 361, 360–365 (2018)

    2018

  22. [33]

    Centi, A

    A. Centi, A. Dutta, S. H. Parekh, and T. Bereau, Bio- phys. J. 118, 1321–1332 (2020)

    2020

  23. [34]

    Izvekov and G

    S. Izvekov and G. A. Voth, J. Phys. Chem. B 109, 2469–2473 (2005)

    2005

  24. [35]

    T. C. Moore, C. R. Iacovella, and C. McCabe, J. Chem. Phys. 140, 10.1063/1.4880555 (2014)

    2014 doi

  25. [36]

    Pulawski, M

    W. Pulawski, M. Jamroz, M. Kolinski, A. Kolinski, and S. Kmiecik, J. Chem. Inf. Model. 56, 2207–2215 (2016)

    2016

  26. [37]

    Zi¸ eba, M.´Slusarz, R

    K. Zi¸ eba, M.´Slusarz, R. ´Slusarz, A. Liwo, C. Czaplewski, and A. K. Sieradzan, J. Phys. Chem. B 123, 7829–7839 (2019)

    2019

  27. [38]

    Alessandri, F

    R. Alessandri, F. Gr¨ unewald, and S. J. Marrink, Adv. Mater. 33, 10.1002/adma.202008635 (2021)

    2021 doi

  28. [39]

    L. R. Kjølbye, G. P. Pereira, A. Bartocci, M. Pannuzzo, S. Albani, A. Marchetto, B. Jim´ enez-Garc´ ıa, J. Mar- tin, G. Rossetti, M. Cecchini, S. Wu, L. Monticelli, and P. C. T. Souza, QRB discov. 3, 10.1017/qrd.2022.16 (2022)

    2022 doi

  29. [40]

    van Hilten, J

    N. van Hilten, J. Methorst, N. Verwei, and H. J. Risse- lada, Sci. Adv. 9, 10.1126/sciadv.ade8839 (2023)

    2023 doi

  30. [41]

    Methorst, N

    J. Methorst, N. van Hilten, A. Hoti, K. S. Stroh, and H. J. Risselada, J. Chem. Theory Comput.20, 1763–1776 (2024)

    2024

  31. [42]

    L¨ utge, M

    S. L¨ utge, M. Krebs, and H. J. Risselada, J. Phys. Chem. B 10.1021/acs.jpcb.4c08200 (2025)

    2025 doi

  32. [43]

    Ghosh, M

    P. Ghosh, M. S. M. Sajjadi, A. Vergari, M. Black, and B. Scholkopf, in International Conference on Learning Representations (2020)

    2020

  33. [44]

    D. P. Kingma and M. Welling, Auto-encoding variational bayes (2013), arXiv:1312.6114

    2013 arXiv

  34. [47]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, in Advances in Neural Information Pro-...

    2019

  35. [48]

    Fey and J

    M. Fey and J. E. Lenssen, in ICLR Workshop on Repre- sentation Learning on Graphs and Manifolds (New Or- leans, USA, 2019)

    2019

  36. [49]

    D. R. Jones, M. Schonlau, and W. J. Welch, J. Glob. Optim. 13, 455–492 (1998)

    1998

  37. [51]

    Torrie and J

    G. Torrie and J. Valleau, J. Comput. Phys. 23, 187–199 (1977)

    1977

  38. [52]

    Chipot and A

    C. Chipot and A. Pohorille, eds., Free Energy Calcula- tions: Theory and Applications in Chemistry and Biology (Springer Berlin Heidelberg, 2007)

    2007

  39. [53]

    A. S. Mey, B. K. Allen, H. E. Bruce Macdonald, J. D. Chodera, D. F. Hahn, M. Kuhn, J. Michel, D. L. Mobley, L. N. Naden, S. Prasad, A. Rizzi, J. Scheen, M. R. Shirts, G. Tresadern, and H. Xu, Living J. Comp. Mol. Sci. 2, 10.33011/livecoms.2.1.18378 (2020)

    2020 doi

  40. [54]

    Menichetti, K

    R. Menichetti, K. H. Kanekal, and T. Bereau, ACS Cent. Sci. 5, 290–298 (2019)

    2019

  41. [55]

    Hoffmann, A

    C. Hoffmann, A. Centi, R. Menichetti, and T. Bereau, Sci. Data 7, 10.1038/s41597-020-0391-0 (2020)

    2020 doi

  42. [58]

    Borges-Ara´ ujo, A

    L. Borges-Ara´ ujo, A. C. Borges-Ara´ ujo, T. N. Ozturk, D. P. Ramirez-Echemendia, B. F´ abi´ an, T. S. Carpenter, S. Thallmair, J. Barnoud, H. I. Ing´ olfsson, G. Hummer, D. P. Tieleman, S. J. Marrink, P. C. T. Souza, and M. N. Melo, J. Chem. Theory Comput. 19, 7387–7404 (2023)

    2023

  43. [59]

    T. N. Ozturk, M. K¨ onig, T. S. Carpenter, K. B. Peder- sen, T. A. Wassenaar, H. I. Ing´ olfsson, and S. J. Marrink, Biophysical approaches for the study of membrane struc- ture—part b: Theory and simulations (Elsevier, 2024) Chap. 7, p. 237–285

    2024

  44. [61]

    M. R. Shirts and J. D. Chodera, J. Chem. Phys. 129, 10.1063/1.2978177 (2008)

    2008 doi

  45. [62]

    Z. Wu, D. L. Dotson, I. Alibay, B. K. Allen, M. S. Barhaghi, J. H´ enin, T. T. Joseph, I. M. Kenney, H. Lee, H. Li, V. Lim, S. Liu, D. Marson, P. T. Merz, A. Schlaich, 13 D. Mobley, M. R. Shirts, and O. Beckstein, J. Open Source Softw. 9, 6934 (2024)

    2024

  46. [63]

    T. A. Wassenaar, H. I. Ing´ olfsson, R. A. B¨ ockmann, D. P. Tieleman, and S. J. Marrink, J. Chem. Theory Comput. 11, 2144–2155 (2015)

    2015

  47. [65]

    A. May, R. Pool, E. van Dijk, J. Bijlard, S. Abeln, J. Heringa, and K. A. Feenstra, Bioinformatics 30, 326–334 (2013)

    2013

  48. [66]

    Alessandri, S

    R. Alessandri, S. Thallmair, C. G. Herrero, R. Mera- Adasme, S. J. Marrink, and P. C. T. Souza, A practical introduction to martini 3 and its application to protein- ligand binding simulations, in A Practical Guide to Re- cent Advances in Multiscale Modeling and Simulation of ...

  49. [67]

    M. D. Wilkinson, M. Dumontier, I. J. Aalbersberg, G. Appleton, M. Axton, A. Baak, N. Blomberg, J.-W. Boiten, L. B. da Silva Santos, P. E. Bourne, J. Bouw- man, A. J. Brookes, T. Clark, M. Crosas, I. Dillo, O. Dumon, S. Edmunds, C. T. Evelo, R. Finkers, A. Gonzalez-Beltran, A. ...

    2016

  50. [68]

    Layer name

    T. Bereau, L. J. Walter, and J. F. Rudzinski, J. Chem. Inf. Model 64, 9413–9423 (2024). 1 Supplementary Information for Navigating Chemical Space: Multi-Level Bayesian Optimization with Hierarchical Coarse-Graining Luis J. Walter Institute for Theoretical Physics, Heidelberg U...

    2024

  51. [69]

    P. C. T. Souza, R. Alessandri, J. Barnoud, S. Thallmair, I. Faustino, F. Gr¨ unewald, I. Patmanidis, H. Abdizadeh, B. M. H. Bruininks, T. A. Wassenaar, P. C. Kroon, J. Melcr, V. Nieto, V. Corradi, H. M. Khan, J. Doma´ nski, M. Javanainen, H. Martinez-Seara, N. Reuter, R. B. Be...

    2021

  52. [70]

    Non-bonded interactions - GROMACS 2025.1 documentation, https://manual.gromacs.org/current/ reference-manual/functions/nonbonded-interactions.html#equation-eqnsigeps , [Accessed 03-05-2025]

    2025

  53. [71]

    B. Mohr, K. Shmilovich, I. S. Kleinw¨ achter, D. Schneider, A. L. Ferguson, and T. Bereau, Chem. Sci.13, 4498–4511 (2022)

    2022

  54. [72]

    itertools — Functions creating iterators for efficient looping, https://docs.python.org/3/library/itertools.html, [Ac- cessed 04-05-2025]

    2025

  55. [73]

    W. L. Hamilton, R. Ying, and J. Leskovec, Inductive representation learning on large graphs (2017), arXiv:1706.02216

    2017 arXiv

  56. [74]

    K. Cho, B. van Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio, Learning phrase representations using rnn encoder-decoder for statistical machine translation (2014), arXiv:1406.1078

    2014 arXiv

  57. [75]

    Gilmer, S

    J. Gilmer, S. S. Schoenholz, P. F. Riley, O. Vinyals, and G. E. Dahl, Neural message passing for quantum chemistry (2017), arXiv:1704.01212

    2017 arXiv

  58. [76]

    Vinyals, S

    O. Vinyals, S. Bengio, and M. Kudlur, Order matters: Sequence to sequence for sets (2015), arXiv:1511.06391

    2015 arXiv

  59. [77]

    Ghosh, M

    P. Ghosh, M. S. M. Sajjadi, A. Vergari, M. Black, and B. Scholkopf, in International Conference on Learning Representa- tions (2020)

    2020

  60. [78]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Des- maison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, and S. Chintala, in Advances in Neural Information Pr...

    2019

  61. [79]

    Fey and J

    M. Fey and J. E. Lenssen, in ICLR Workshop on Representation Learning on Graphs and Manifolds (New Orleans, USA, 2019)

    2019

  62. [80]

    B. Xu, N. Wang, T. Chen, and M. Li, Empirical evaluation of rectified activations in convolutional network (2015), arXiv:1505.00853

    2015 arXiv

  63. [81]

    D. P. Kingma and J. Ba, Adam: A method for stochastic optimization (2014), arXiv:1412.6980

    2014 arXiv

  64. [82]

    Centi, A

    A. Centi, A. Dutta, S. H. Parekh, and T. Bereau, Biophys. J. 118, 1321–1332 (2020)

    2020

  65. [83]

    M. J. Abraham, T. Murtola, R. Schulz, S. P´ all, J. C. Smith, B. Hess, and E. Lindahl, SoftwareX 1–2, 19–25 (2015)

    2015

  66. [84]

    P´ all, A

    S. P´ all, A. Zhmurov, P. Bauer, M. Abraham, M. Lundborg, A. Gray, B. Hess, and E. Lindahl, J. Chem. Phys. 153, 10.1063/5.0018516 (2020)

    2020 doi

  67. [85]

    Bernetti and G

    M. Bernetti and G. Bussi, J. Chem. Phys. 153, 10.1063/5.0020514 (2020)

    2020 doi

  68. [86]

    H. Kim, B. F´ abi´ an, and G. Hummer, J. Chem. Theory Comput.19, 8919–8929 (2023)

    2023

  69. [87]

    Fiorin, M

    G. Fiorin, M. L. Klein, and J. H´ enin, Mol. Phys. 111, 3345–3362 (2013)

    2013

  70. [88]

    Barnoud, G

    J. Barnoud, G. Rossi, S. J. Marrink, and L. Monticelli, PLoS Comput. Biol. 10, e1003873 (2014)

    2014

  71. [89]

    J. R. Gardner, G. Pleiss, D. Bindel, K. Q. Weinberger, and A. G. Wilson, in Advances in Neural Information Processing Systems (Montr´ eal, Canada, 2018)

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.