Pith. sign in

REVIEW 4 major objections 4 minor 74 references

Graph-Coarsening for Machine Learning Coarse-grained Molecular Dynamics

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a graph-coarsening rule based on local variation cost can replace manual bead selection in coarse-grained molecular dynamics, and that MACE potentials trained on the resulting beads reproduce reference structural…

desk verdict Clever pipeline, but the force-matching target in the methods is internally inconsistent; as written, the paper cannot support its central claim. read the letter →

arxiv 2507.16531 v1 pith:ZYVQSSH2 submitted 2025-07-22 cond-mat.soft

Soumya Mondal , Subhanu Halder , Debarchan Basu , Sandeep Kumar , Tarak Karmakar This is my paper
classification cond-mat.soft
keywords coarse-grainedmoleculardynamicsgraphcoarseningspectraltheorylocalvariationforcematchingMACEmachine-learnedpotentialsgraphs
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

This paper claims that the bottleneck of coarse-grained molecular dynamics—choosing which atoms merge into a bead—can be solved by a graph algorithm rather than chemical intuition. The algorithm contracts molecular-graph nodes in candidate sets chosen by a local variational cost, which preserves the leading spectral structure of the graph, and the resulting beads are used to train a MACE (Message Passing Atomic Cluster Expansion) potential by force matching. On aspirin, azobenzene, and 3BPA, simulations driven by these potentials reproduce the bond-length distributions, dihedral densities, orientational-angle densities, and radial distribution functions of the reference CG-mapped trajectories. The authors present the workflow as a bottom-up, deterministic, and systematically improvable way to build CG models, with the coarsening ratio as the main control parameter.

What carries the argument

The load-bearing object is the local variational cost, $\mathrm{cost}_\ell(C) = \| S_{\ell-1} \Pi^\perp_C A_{\ell-1} \|_2^2 / (|C|-1)$, computed for each candidate contraction set $C$. Here $\Pi^\perp_C$ is the orthogonal projection onto the complement of the leading-eigenvector subspace, so the cost measures how much contracting $C$ distorts the Laplacian in the directions the coarsening is meant to preserve. Greedy contraction of lowest-cost non-overlapping candidates keeps the per-level variation $\sigma_\ell$ below a threshold, and the contracted Laplacian $L_\ell = P_\ell^T L_{\ell-1} P_\ell$ preserves cut weights exactly. The resulting mapping matrix $P = P_c \dots P_1$ assigns each atom to one CG bead, with bead coordinates and forces defined as unweighted means of the atoms in each contraction set; this binary structure makes the force projection the identity and feeds directly into MACE's equivariant message passing.

What would settle it

Train the same MACE-CG pipeline on aspirin at 500 K, then simulate at 400 K or 600 K and compare bond-length, dihedral, and radial distribution statistics to a mapped all-atom reference at that temperature; the graph-spectral guarantee makes no prediction about temperature transfer, so a large structural error would show that spectral preservation is not the controlling physical variable.

Watch

Extended reading notes

Core claim

The central claim is that a coarse-grained representation can be obtained by multilevel spectral graph coarsening, and that this representation is good enough to train an accurate ML-CG potential. Concretely, at each level the method builds candidate contraction sets—one-hop neighborhoods (LVN) or maximal cliques (LVC)—and greedily contracts the lowest-cost candidates, with the local variation cost measuring the worst-case spectral distortion induced by merging a set into a supernode. The product of level-wise contraction matrices is the mapping matrix $P$; CG bead coordinates and forces are arithmetic means over each contraction set. Because $P$ has orthogonal rows, the force projection simplifies to the same binary assignment, which the paper argues preserves translational and rotational equivariance needed for MACE training. The reported evidence is that MACE-CG simulations reproduce bond lengths, N=N and C-N distances, a carboxylate–ester angle in aspirin, the C-N-N-C dihedral in azobenzene, and atomic-density $g(r)$ for all three molecules against the CG-mapped reference.

Load-bearing premise

The method assumes that preserving the low-frequency structure of a hand-built molecular graph—one whose edge weights use manually chosen parameters—is the right criterion for physical accuracy; if the graph does not encode the interactions that control molecular structure, a spectrally faithful mapping can still be physically poor.

Editorial extensions

If this is right

  • CG bead definitions become a deterministic graph computation: given only the heavy-atom coordinates, a force-and-coordinate dataset, and a target coarsening ratio, the LVN/LVC rules produce the same mapping every run, with no learned parameters.
  • Because maximal cliques are the candidate sets in LVC, cyclic motifs such as phenyl rings are contracted as units, preserving ring topology in the coarse representation (for example, three beads per ring in azobenzene and 3BPA).
  • MACE trained by instantaneous force matching on the CG-mapped coordinates reproduces the equilibrium structural statistics of the mapped reference, so the workflow removes the need for a predefined CG potential prior.
  • The same multilevel coarsening procedure applies to all-atom and no-hydrogen representations, and the paper validates it at coarsening ratios of 0.6 and 0.45, indicating the method is not locked to a single resolution.

Reading between the lines

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

  • A consequence the authors leave implicit is that the variation cost computed during coarsening could serve as a per-bead or per-level confidence score, letting a practitioner choose the coarsening ratio by watching when the spectral distortion begins to grow.
  • The paper validates only equilibrium structure; a natural next test is whether LVN/LVC-mapped MACE-CG also reproduces kinetics such as diffusion, reorientational correlation times, or transition rates across the azobenzene dihedral barrier, which the current evidence does not address.
  • Because the mapping step is independent of the potential and requires no training, it could be paired with other CG potential construction methods as a drop-in bead-definition module, a combination the paper does not explore.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes an automated, graph-based coarse-graining pipeline for molecular dynamics. It constructs a weighted molecular graph from heavy atoms, applies multilevel graph coarsening (Local Variation Neighborhood/Clique, based on Loukas's spectral coarsening algorithm) to define CG beads, and then trains a MACE equivariant message-passing potential by instantaneous force matching. The method is tested on aspirin, azobenzene, and 3BPA from the MD17 dataset. In each case, the authors run CG MD simulations and compare bond-length distributions, angle/dihedral densities, and RDFs against the corresponding 'CG-Mapped trajectory' obtained by applying the chosen mapping to the original training trajectory. The central claim is that the LVN/LVC coarsening plus MACE produces accurate and efficient ML-CG potentials in a bottom-up, theoretically grounded manner.

Significance. If the central claim were fully established, the paper would offer a useful, deterministic, and interpretable alternative to learned mapping schemes in ML-CG. The combination of spectral graph coarsening with a modern equivariant ML potential is conceptually interesting and the pipeline is CPU-bound and fast. However, the validation is entirely in-sample: the reference distributions are obtained by mapping the same trajectories used to train the force field, so the reported agreement is largely a self-consistency check. Moreover, the force-matching target is not rigorously derived, and no comparison against baseline mappings or alternative CG potentials is provided. These issues substantially reduce the current significance of the work as a methods paper.

major comments (4)
  1. [Methods, Eqs. (2)-(7)] The force-projection formalism is internally inconsistent. The paper states that a binary assignment matrix satisfies P P^T = I, but for the centroid mapping defined in Eq. (6) (row entries 1/|C_r|) the matrix P P^T equals diag(1/|C_r|), not I; for an unnormalized binary assignment it equals diag(|C_r|). Consequently, the least-squares projection PF = (P P^T)^{-1} P F from Eq. (2) is the sum of atomic forces in a bead, not the average used in Eq. (7). If the implementation actually uses Eq. (7), the training target is scaled differently per bead and does not correspond to the gradient of the many-body PMF; the learned potential would not be the PMF and the reported equilibrium agreement would not be expected. The manuscript must state precisely which projection is implemented and reconcile Eqs. (2), (6), and (7).
  2. [Results, Figures 3-5 and Methods, Eq. (3)] The validation is in-sample and self-referential. The 'ground truth' data are constructed by applying the same coarsening matrix P to the same MD trajectory that provides the training forces for the MACE potential; thus the comparison measures how well the CG model reproduces the projected training data, not how well the coarsening strategy generalizes or whether it outperforms alternative mappings. No error bars are given on the plotted distributions, no out-of-sample test is performed (for example, at a temperature not used in training or on configurations not in the training set), and no baseline mapping (e.g., random or chemically intuitive bead definitions) is compared. To support the claim that graph-coarsening is beneficial, the authors should add such comparisons and report quantitative uncertainties on the JSD values.
  3. [Methods, Molecular Graph Representation and Coarse-Graining Mapping Procedure] The graph weights and coarsening parameters are not rigorously justified. The edge weights are defined as w_ij = exp[-alpha(d_ij - d_0)] with hand-chosen alpha, d_0, and cutoff d_cut, and each system uses a user-fixed coarsening ratio r. The spectral guarantees of Loukas apply to this arbitrary weighted graph, not to the molecular free-energy surface. The paper therefore does not establish that preserving the spectrum of this graph is a suitable proxy for physical accuracy. The authors should test the sensitivity of the final CG results to these parameters and ideally relate the graph Laplacian to physically relevant interactions.
  4. [Results, Aspirin/Azobenzene/3BPA subsections] The claim that the ML-CG potentials are 'accurate' is not supported by quantitative comparison against existing CG construction methods. The main text cites only visually overlapping distributions and defers JSD values to the SI, with no benchmark against other CG mappings or other ML-CG force fields. Without such comparisons, the demonstrated accuracy is only internal consistency. The authors should include quantitative measures of error (e.g., JSD values in the main text) and at least one comparison to a standard mapping (e.g., MARTINI-style or random mapping) to validate the added value of the spectral coarsening.
minor comments (4)
  1. [Abstract/Introduction] There are several typos and grammatical errors, for example 'dimesnional' (page 3), 'combinational' (page 5), 'train upto' (page 9), and 'pharmaceautical' (page 7). The manuscript should be carefully proofread.
  2. [Figure 4 caption] The caption for Figure 4a reads 'Demonstration of the coarsening strategy for the aspirin molecule', but the figure shows azobenzene; this should be corrected.
  3. [Methods, Force Matching] In the text preceding Eq. (3), the notation uses i both for the configuration index and the atom index, which is confusing. Please use distinct indices.
  4. [Methods, Coarse-Graining Mapping Procedure] The derivation of cost_l(C) in Eq. (4) is sketched very briefly; a reader cannot easily verify how S_l-1 and A_l-1 are defined from the preceding paragraph. Please provide a self-contained definition or a reference to the precise equations in Loukas (2019).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the coarsening and force-matching derivation chain is self-contained, and the validation is a standard bottom-up consistency check rather than a definitional identity.

full rationale

The paper's central derivation chain is not circular. The CG mapping is produced by the Loukas multilevel graph-reduction framework [70] with LVN/LVC candidate sets, an external, parameter-free spectral coarsening method whose guarantees apply to the stated molecular graph; it is not defined in terms of the later MACE potential or the validation statistics. The MACE potential is trained by instantaneous force matching (Eq. 3) against forces obtained from the CG-mapped MD17 reference trajectory, which is the standard bottom-up construction of the many-body potential of mean force. The equilibrium bond-length, dihedral, and RDF comparisons in Figs. 3–5 are against the same mapped reference ensemble; this is a consistency check induced by the force-matching target, but it is not a definitional identity because the model minimizes a force loss rather than the displayed distributional statistics, and the agreement is a nontrivial result of optimization and MD stability. No load-bearing self-citation chain exists: the coarsening algorithm is attributed to Loukas [70], MACE to Batatia et al. [59], and the MD17 data to Christensen and von Lilienfeld [73]; the authors' own graph-coarsening papers [65–68] are cited only as examples of possible mapping choices, not as the source of the central result. The limitations section honestly states that no learned condensation scheme is used and that full ablation on large molecules is infeasible; these are scoping statements, not circular moves. One correctness issue, noted separately from circularity, is the Methods statement that for a binary assignment matrix P P^T = I together with Eq. 7's average force: for the centroid mapping defined by Eq. 6, P P^T = diag(1/|C_r|), and the least-squares projection of Eq. 2 is the sum of atomic forces, not the average. That inconsistency affects the interpretation of the trained target but does not make any claimed prediction equivalent to its input by construction.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

Central claim rests on hand-chosen graph parameters, the Loukas spectral guarantee, and standard force-matching theory. No new physical entities are introduced.

free parameters (4)
  • coarsening ratio r = 0.6 (aspirin, 3BPA), 0.45 (azobenzene)
    Controls final number of CG beads; chosen per molecule by hand; directly sets resolution and affects all subsequent results.
  • edge weight damping alpha and reference distance d0 = not specified numerically
    Chosen so bonded contacts have weight near 1 and others zero; introduced ad hoc in graph construction.
  • edge cutoff dcut = not specified numerically
    Defines graph edges; chosen to include covalent bonds and short contacts; affects the entire Laplacian and coarsening.
  • MACE architecture hyperparameters = reported in SI (not available)
    Training epochs, hidden dimensions, cutoff for MACE; these determine the fitted force field.
assumptions (4)
  • standard math The force-matching theorem: minimizing instantaneous force error reproduces the many-body PMF in the limit of sufficient sampling (Izvekov and Voth, Noid et al.)
    The entire ML-CG training relies on this known result, invoked in Methods Section 2.1.
  • standard math Loukas's spectral approximation guarantees for graph reduction (Theorem guarantees on preservation of Laplacian spectra)
    Used in Methods 'Coarse-Graining Mapping Procedure' to justify LVN/LVC contraction. The guarantee is for the graph Laplacian, not physical properties.
  • domain assumption Graph representation with heavy atoms only and cutoff-based edges is a sufficient description for CG mapping
    Hydrogens are removed before coarsening, and edges depend on a hand-chosen cutoff; these choices are not justified from physics.
  • domain assumption MACE is expressive enough to represent the CG potential of mean force
    The paper assumes the equivariant architecture can fit the target PMF; no expressivity analysis is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Graph-Coarsening for Machine Learning Coarse-grained Molecular Dynamics." pith.science (2026). https://pith.science/paper/ZYVQSSH2

@misc{pith2026250716531,
  author       = {Pith},
  title        = {Pith review of: Graph-Coarsening for Machine Learning Coarse-grained Molecular Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZYVQSSH2}},
  note         = {Machine review of arXiv:2507.16531}
}
read the original abstract

Coarse-grained (CG) molecular dynamics (MD) simulations can simulate large molecular complexes over extended timescales by reducing degrees of freedom. A critical step in CG modeling is the selection of the CG mapping algorithm, which directly influences both accuracy and interpretability of the model. Despite progress, the optimal strategy for coarse-graining remains a challenging task, highlighting the necessity for a comprehensive theoretical framework. In this work, we present a graph-based coarsening approach to develop CG models. Coarse-grained sites are obtained through edge contractions, where nodes are merged based on a local variational cost metric while preserving key spectral properties of the original graph. Furthermore, we illustrate how Message Passing Atomic Cluster Expansion (MACE) can be applied to generate ML-CG potentials that are not only highly efficient but also accurate. Our approach provides a bottom-up, theoretically grounded computational method for the development of systematically improvable CG potentials.

Figures

Figures reproduced from arXiv: 2507.16531 by the authors.

Figure 1
Figure 1. The overall framework of multilevel graph-coarsening for bottom-up ML-CG molecular dynamics [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Toy model for developing a coarsened graph using the multilevel graph coarsening technique. a) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. a). Demonstration of the coarsening process for the aspirin molecule using LVN coarsening [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: a). Demonstration of the coarsening strategy for the aspirin molecule using the LVN coarsening [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: a). Schematic representation of the coarsening strategy for the 3BPA molecule using the LVC [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

74 extracted references · 72 canonical work pages

  1. [1]

    Computer simulation of liquids

    Michael P Allen and Dominic J Tildesley. Computer simulation of liquids . Oxford university press, 2017

    work page 2017

  2. [2]

    Understanding molecular simulation: from algorithms to applications

    Daan Frenkel and Berend Smit. Understanding molecular simulation: from algorithms to applications. Elsevier, 2023

    work page 2023

  3. [3]

    The martini force field: coarse grained model for biomolecular simulations

    Siewert J Marrink, H Jelger Risselada, Serge Yefimov, D Peter Tieleman, and Alex H De Vries. The martini force field: coarse grained model for biomolecular simulations. The journal of physical chem- istry B, 111(27):7812–7824, 2007

    work page 2007

  4. [4]

    Coarse-grained protein models and their applications

    Sebastian Kmiecik, Dominik Gront, Michal Kolinski, Lukasz Wieteska, Aleksandra Elzbieta Dawid, and Andrzej Kolinski. Coarse-grained protein models and their applications. Chemical reviews, 116(14):7898–7936, 2016

    work page 2016

  5. [5]

    Uncertainty driven active learning of coarse grained free energy models

    Blake R Duschatko, Jonathan Vandermause, Nicola Molinari, and Boris Kozinsky. Uncertainty driven active learning of coarse grained free energy models. npj Computational Materials, 10(1):9, 2024

    work page 2024

  6. [6]

    K-means clustering coarse-graining (kmc-cg): A next generation methodology for determining optimal coarse-grained mappings of large biomolecules

    Jiangbo Wu, Weizhi Xue, and Gregory A V oth. K-means clustering coarse-graining (kmc-cg): A next generation methodology for determining optimal coarse-grained mappings of large biomolecules. Journal of Chemical Theory and Computation, 19(23):8987–8997, 2023

    work page 2023

  7. [7]

    Krishnakanth Baratam and Anand Srivastava. Sop-multi: A self-organized polymer-based coarse- grained model for multidomain and intrinsically disordered proteins with conformation ensem- ble consistent with experimental scattering data. Journal of Chemical Theory and Computation , 20(22):10179–10198, 2024

    work page 2024

  8. [8]

    Insight into the density-dependence of pair potentials for predictive coarse-grained models

    Maria C Lesniewski and WG Noid. Insight into the density-dependence of pair potentials for predictive coarse-grained models. The Journal of Physical Chemistry B, 128(5):1298–1316, 2024

    work page 2024

Show all 74 references
  1. [9]

    Unveiling interactions of a peptide-bound monolayer-protected metal nanocluster with a lipid bilayer

    Soumya Mondal and Tarak Karmakar. Unveiling interactions of a peptide-bound monolayer-protected metal nanocluster with a lipid bilayer. The Journal of Physical Chemistry Letters, 16(13):3351–3358, 2025

    2025

  2. [10]

    Evolutionary algorithm in the optimization of a coarse-grained force field

    Filip Leonarski, Fabio Trovato, Valentina Tozzini, Andrzej Les, and Joanna Trylska. Evolutionary algorithm in the optimization of a coarse-grained force field. Journal of Chemical Theory and Com- putation, 9(11):4874–4889, 2013. 14

    2013

  3. [11]

    Temporally coher- ent backmapping of molecular trajectories from coarse-grained to atomistic resolution

    Kirill Shmilovich, Marc Stieffenhofer, Nicholas E Charron, and Moritz Hoffmann. Temporally coher- ent backmapping of molecular trajectories from coarse-grained to atomistic resolution. The Journal of Physical Chemistry A, 126(48):9124–9139, 2022

    2022

  4. [12]

    Multiscale coarse graining of liquid-state systems

    Sergei Izvekov and Gregory A V oth. Multiscale coarse graining of liquid-state systems. The Journal of chemical physics, 123(13), 2005

    2005

  5. [13]

    Structure of a tractable stochastic mimic of soft particles

    Galen T Craven, Alexander V Popov, and Rigoberto Hernandez. Structure of a tractable stochastic mimic of soft particles. Soft Matter, 10(29):5350–5361, 2014

    2014

  6. [14]

    Derivation of coarse-grained potentials via multistate iterative boltzmann inversion

    Timothy C Moore, Christopher R Iacovella, and Clare McCabe. Derivation of coarse-grained potentials via multistate iterative boltzmann inversion. The Journal of chemical physics, 140(22), 2014

    2014

  7. [15]

    Bottom-up coarse-grained models that accurately describe the structure, pressure, and compressibility of molecular liquids.The Journal of chemical physics, 143(24), 2015

    Nicholas JH Dunn and WG Noid. Bottom-up coarse-grained models that accurately describe the structure, pressure, and compressibility of molecular liquids.The Journal of chemical physics, 143(24), 2015

    2015

  8. [16]

    Transfer-learning-based coarse-graining method for simple fluids: toward deep inverse liquid-state theory

    Alireza Moradzadeh and Narayana R Aluru. Transfer-learning-based coarse-graining method for simple fluids: toward deep inverse liquid-state theory. The journal of physical chemistry letters , 10(6):1242–1250, 2019

    2019

  9. [17]

    Influence of topology on effective potentials: coarse-graining ring polymers

    Arturo Narros, Angel J Moreno, and Christos N Likos. Influence of topology on effective potentials: coarse-graining ring polymers. Soft Matter, 6(11):2435–2441, 2010

    2010

  10. [18]

    Effective surface coverage of coarse- grained soft matter

    Galen T Craven, Alexander V Popov, and Rigoberto Hernandez. Effective surface coverage of coarse- grained soft matter. The Journal of Physical Chemistry B, 118(49):14092–14102, 2014

    2014

  11. [19]

    Solvent entropy and coarse-graining of polymer lattice models

    Pengfei Zhang and Qiang Wang. Solvent entropy and coarse-graining of polymer lattice models. Soft Matter, 9(47):11183–11187, 2013

    2013

  12. [20]

    Integrating machine learning in the coarse-grained molecular simulation of polymers

    Eleonora Ricci and Niki Vergadou. Integrating machine learning in the coarse-grained molecular simulation of polymers. The Journal of Physical Chemistry B, 127(11):2302–2322, 2023

    2023

  13. [21]

    Coarse grained protein- lipid model with application to lipoprotein particles

    Amy Y Shih, Anton Arkhipov, Lydia Freddolino, and Klaus Schulten. Coarse grained protein- lipid model with application to lipoprotein particles. The Journal of Physical Chemistry B , 110(8):3674– 3684, 2006

    2006

  14. [22]

    Coarse-graining methods for computational biology.Annual review of biophysics, 42(1):73–93, 2013

    Marissa G Saunders and Gregory A V oth. Coarse-graining methods for computational biology.Annual review of biophysics, 42(1):73–93, 2013

    2013

  15. [23]

    Perspective: Coarse-grained models for biomolecular systems

    William George Noid. Perspective: Coarse-grained models for biomolecular systems. The Journal of chemical physics, 139(9), 2013

    2013

  16. [24]

    Flow-matching: Effi- cient coarse-graining of molecular dynamics without forces

    Jonas Kohler, Yaoyi Chen, Andreas Kramer, Cecilia Clementi, and Frank No ´e. Flow-matching: Effi- cient coarse-graining of molecular dynamics without forces. Journal of Chemical Theory and Compu- tation, 19(3):942–952, 2023

    2023

  17. [25]

    Exploring the landscape of model representations

    Thomas T Foley, Katherine M Kidder, M Scott Shell, and WG Noid. Exploring the landscape of model representations. Proceedings of the National Academy of Sciences, 117(39):24061–24068, 2020

    2020

  18. [26]

    A data-driven perspective on the hierarchical assembly of molecular structures

    Lorenzo Boninsegna, Ralf Banisch, and Cecilia Clementi. A data-driven perspective on the hierarchical assembly of molecular structures. Journal of Chemical Theory and Computation , 14(1):453–460, 2018. 15

    2018

  19. [27]

    Openmscg: A software tool for bottom-up coarse-graining

    Yuxing Peng, Alexander J Pak, Aleksander EP Durumeric, Patrick G Sahrmann, Sriramvignesh Mani, Jaehyeok Jin, Timothy D Loose, Jeriann Beiter, and Gregory A V oth. Openmscg: A software tool for bottom-up coarse-graining. The Journal of Physical Chemistry B, 127(40):8537–8550, 2023

    2023

  20. [28]

    Graph neural network based coarse-grained mapping prediction

    Zhiheng Li, Geemi P Wellawatte, Maghesree Chakraborty, Heta A Gandhi, Chenliang Xu, and An- drew D White. Graph neural network based coarse-grained mapping prediction. Chemical science, 11(35):9524–9531, 2020

    2020

  21. [29]

    Coarse-graining auto-encoders for molecular dynamics

    Wujie Wang and Rafael G ´omez-Bombarelli. Coarse-graining auto-encoders for molecular dynamics. npj Computational Materials, 5(1):125, 2019

    2019

  22. [30]

    Martini 3: a general purpose force field for coarse-grained molecular dynamics

    Paulo CT Souza, Riccardo Alessandri, Jonathan Barnoud, Sebastian Thallmair, Ignacio Faustino, Fabian Gr¨unewald, Ilias Patmanidis, Haleh Abdizadeh, Bart MH Bruininks, Tsjerk A Wassenaar, et al. Martini 3: a general purpose force field for coarse-grained molecular dynamics. Nat...

    2021

  23. [31]

    Machine learning coarse- grained potentials of protein thermodynamics

    Maciej Majewski, Adri `a P ´erez, Philipp Th ¨olke, Stefan Doerr, Nicholas E Charron, Toni Giorgino, Brooke E Husic, Cecilia Clementi, Frank No ´e, and Gianni De Fabritiis. Machine learning coarse- grained potentials of protein thermodynamics. Nature communications, 14(1):5739, 2023

    2023

  24. [32]

    Cgcompiler: auto- mated coarse-grained molecule parametrization via noise-resistant mixed-variable optimization

    Kai Steffen Stroh, Paulo CT Souza, Luca Monticelli, and Herre Jelger Risselada. Cgcompiler: auto- mated coarse-grained molecule parametrization via noise-resistant mixed-variable optimization. Jour- nal of Chemical Theory and Computation, 19(22):8384–8400, 2023

    2023

  25. [33]

    Automated coarse-grained mapping algorithm for the martini force field and benchmarks for membrane–water partitioning

    Thomas D Potter, Elin L Barrett, and Mark A Miller. Automated coarse-grained mapping algorithm for the martini force field and benchmarks for membrane–water partitioning. Journal of Chemical Theory and Computation, 17(9):5777–5791, 2021

    2021

  26. [34]

    Generative coarse-graining of molecular conformations

    Wujie Wang, Minkai Xu, Chen Cai, Benjamin Kurt Miller, Tess Smidt, Yusu Wang, Jian Tang, and Rafael G ´omez-Bombarelli. Generative coarse-graining of molecular conformations. arXiv preprint arXiv:2201.12176, 2022

    2022 arXiv

  27. [35]

    Coarsenconf: Equivariant coarsening with aggre- gated attention for molecular conformer generation

    Danny Reidenbach and Aditi S Krishnapriyan. Coarsenconf: Equivariant coarsening with aggre- gated attention for molecular conformer generation. Journal of Chemical Information and Modeling, 65(1):22–30, 2024

    2024

  28. [36]

    An information-theory- based approach for optimal model reduction of biomolecules

    Marco Giulini, Roberto Menichetti, M Scott Shell, and Raffaello Potestio. An information-theory- based approach for optimal model reduction of biomolecules. Journal of chemical theory and compu- tation, 16(11):6795–6813, 2020

    2020

  29. [37]

    Srinivasan S Iyengar, Timothy C Ricard, and Xiao Zhu. Reformulation of all oniom-type molecu- lar fragmentation approaches and many-body theories using graph-theory-based projection operators: Applications to dynamics, molecular potential surfaces, machine learning, and quantu...

    2024

  30. [38]

    Graph theory meets ab initio molecular dynamics: Atomic structures¡? format?¿ and transformations at the nanoscale

    Fabio Pietrucci and Wanda Andreoni. Graph theory meets ab initio molecular dynamics: Atomic structures¡? format?¿ and transformations at the nanoscale. Physical review letters, 107(8):085504, 2011

    2011

  31. [39]

    Deepcg: Constructing coarse-grained models via deep neural networks

    Linfeng Zhang, Jiequn Han, Han Wang, Roberto Car, et al. Deepcg: Constructing coarse-grained models via deep neural networks. The Journal of chemical physics, 149(3), 2018. 16

    2018

  32. [40]

    Machine-learned coarse- grained models

    Karteek K Bejagam, Samrendra Singh, Yaxin An, and Sanket A Deshmukh. Machine-learned coarse- grained models. The journal of physical chemistry letters, 9(16):4667–4672, 2018

    2018

  33. [41]

    Neural network based prediction of conformational free energies-a new route toward coarse-grained simulation models

    Tobias Lemke and Christine Peter. Neural network based prediction of conformational free energies-a new route toward coarse-grained simulation models. Journal of chemical theory and computation , 13(12):6213–6221, 2017

    2017

  34. [42]

    Machine learning of coarse-grained molecular dynamics force fields

    Jiang Wang, Simon Olsson, Christoph Wehmeyer, Adri`a P´erez, Nicholas E Charron, Gianni De Fabri- tiis, Frank No´e, and Cecilia Clementi. Machine learning of coarse-grained molecular dynamics force fields. ACS central science, 5(5):755–767, 2019

    2019

  35. [43]

    Investigating molecu- lar kinetics by variationally optimized diffusion maps

    Lorenzo Boninsegna, Gianpaolo Gobbo, Frank No ´e, and Cecilia Clementi. Investigating molecu- lar kinetics by variationally optimized diffusion maps. Journal of chemical theory and computation , 11(12):5947–5960, 2015

    2015

  36. [44]

    Deep coarse-grained potentials via relative entropy minimization

    Stephan Thaler, Maximilian Stupp, and Julija Zavadlav. Deep coarse-grained potentials via relative entropy minimization. The Journal of Chemical Physics, 157(24), 2022

    2022

  37. [45]

    Graph-based approach to systematic molecular coarse-graining

    Michael A Webb, Jean-Yves Delannoy, and Juan J De Pablo. Graph-based approach to systematic molecular coarse-graining. Journal of chemical theory and computation, 15(2):1199–1208, 2018

    2018

  38. [46]

    Encoding and selecting coarse-grain mapping operators with hierarchical graphs

    Maghesree Chakraborty, Chenliang Xu, and Andrew D White. Encoding and selecting coarse-grain mapping operators with hierarchical graphs. The Journal of Chemical Physics, 149(13), 2018

    2018

  39. [47]

    Bottom-up coarse-graining: Principles and perspectives

    Jaehyeok Jin, Alexander J Pak, Aleksander EP Durumeric, Timothy D Loose, and Gregory A V oth. Bottom-up coarse-graining: Principles and perspectives. Journal of chemical theory and computation, 18(10):5759–5791, 2022

    2022

  40. [48]

    Top-down machine learning of coarse- grained protein force fields

    Carles Navarro, Maciej Majewski, and Gianni De Fabritiis. Top-down machine learning of coarse- grained protein force fields. Journal of Chemical Theory and Computation, 19(21):7518–7526, 2023

    2023

  41. [49]

    Machine learning implicit solvation for molecular dynamics.The Journal of Chemical Physics, 155(8), 2021

    Yaoyi Chen, Andreas Kr ¨amer, Nicholas E Charron, Brooke E Husic, Cecilia Clementi, and Frank No´e. Machine learning implicit solvation for molecular dynamics.The Journal of Chemical Physics, 155(8), 2021

    2021

  42. [50]

    Physnet: A neural network for predicting energies, forces, dipole moments, and partial charges

    Oliver T Unke and Markus Meuwly. Physnet: A neural network for predicting energies, forces, dipole moments, and partial charges. Journal of chemical theory and computation, 15(6):3678–3693, 2019

    2019

  43. [51]

    Ac- curate machine learned quantum-mechanical force fields for biomolecular simulations

    Oliver T Unke, Martin St ¨ohr, Stefan Ganscha, Thomas Unterthiner, Hartmut Maennel, Sergii Kashu- bin, Daniel Ahlin, Michael Gastegger, Leonardo Medrano Sandonas, Alexandre Tkatchenko, et al. Ac- curate machine learned quantum-mechanical force fields for biomolecular simulatio...

    2022 arXiv

  44. [52]

    Machine learning for molecular simulation

    Frank No ´e, Alexandre Tkatchenko, Klaus-Robert M¨uller, and Cecilia Clementi. Machine learning for molecular simulation. Annual review of physical chemistry, 71(1):361–390, 2020

    2020

  45. [53]

    Multi- body effects in a coarse-grained protein force field

    Jiang Wang, Nicholas Charron, Brooke Husic, Simon Olsson, Frank No´e, and Cecilia Clementi. Multi- body effects in a coarse-grained protein force field. The Journal of Chemical Physics, 154(16), 2021

    2021

  46. [54]

    The multiscale coarse-graining method

    William George Noid, Jhih-Wei Chu, Gary S Ayton, Vinod Krishna, Sergei Izvekov, Gregory A V oth, Avisek Das, and Hans C Andersen. The multiscale coarse-graining method. i. a rigorous bridge be- tween atomistic and coarse-grained models. The Journal of chemical physics, 128(24)...

    2008

  47. [55]

    Coarse graining molecular dynamics with graph neural networks

    Brooke E Husic, Nicholas E Charron, Dominik Lemm, Jiang Wang, Adri `a P´erez, Maciej Majewski, Andreas Kr ¨amer, Yaoyi Chen, Simon Olsson, Gianni De Fabritiis, et al. Coarse graining molecular dynamics with graph neural networks. The Journal of chemical physics, 153(19), 2020

    2020

  48. [56]

    Machine learned coarse- grained protein force-fields: Are we there yet? Current opinion in structural biology , 79:102533, 2023

    Aleksander EP Durumeric, Nicholas E Charron, Clark Templeton, F´elix Musil, Klara Bonneau, Aldo S Pasos-Trejo, Yaoyi Chen, Atharva Kelkar, Frank No´e, and Cecilia Clementi. Machine learned coarse- grained protein force-fields: Are we there yet? Current opinion in structural bi...

    2023

  49. [57]

    Navigating protein landscapes with a machine-learned transferable coarse-grained model

    Nicholas E Charron, Felix Musil, Andrea Guljas, Yaoyi Chen, Klara Bonneau, Aldo S Pasos-Trejo, Ja- copo Venturin, Daria Gusew, Iryna Zaporozhets, Andreas Kr¨amer, et al. Navigating protein landscapes with a machine-learned transferable coarse-grained model. arXiv preprint arXi...

    arXiv

  50. [58]

    Many-body coarse-grained molecular dynam- ics with the atomic cluster expansion

    Yangshuai Wang, Gabor Csanyi, and Christoph Ortner. Many-body coarse-grained molecular dynam- ics with the atomic cluster expansion. arXiv preprint arXiv:2502.04661, 2025

    2025 arXiv

  51. [59]

    Mace: Higher order equivariant message passing neural networks for fast and accurate force fields

    Ilyes Batatia, David P Kovacs, Gregor Simm, Christoph Ortner, and G ´abor Cs ´anyi. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields. Advances in neural information processing systems, 35:11423–11436, 2022

    2022

  52. [60]

    Crash testing machine learning force fields for molecules, materials, and interfaces: Model analysis in the tea challenge 2023

    I Poltavsky, A Charkin-Gorbulin, M Puleva, G Cordeiro Fonseca, I Batatia, NJ Browning, S Chmiela, M Cui, JT Frank, S Heinen, et al. Crash testing machine learning force fields for molecules, materials, and interfaces: Model analysis in the tea challenge 2023. chemrxiv 2024

    2023

  53. [61]

    Projection of diffusions on submanifolds: Application to mean force computation

    Giovanni Ciccotti, Tony Lelievre, and Eric Vanden-Eijnden. Projection of diffusions on submanifolds: Application to mean force computation. Communications on Pure and Applied Mathematics: A Jour- nal Issued by the Courant Institute of Mathematical Sciences, 61(3):371–408, 2008

    2008

  54. [62]

    Evaluation of the mace force field architecture: From medicinal chemistry to materials science

    D ´avid P´eter Kov´acs, Ilyes Batatia, Eszter S´ara Arany, and G´abor Cs´anyi. Evaluation of the mace force field architecture: From medicinal chemistry to materials science. The Journal of Chemical Physics , 159(4), 2023

    2023

  55. [63]

    Trans- ferability of data sets between machine-learned interatomic potential algorithms

    Samuel P Niblett, Panagiotis Kourtis, Ioan-Bogdan Magd ˘au, Clare P Grey, and G´abor Cs´anyi. Trans- ferability of data sets between machine-learned interatomic potential algorithms. Journal of Chemical Theory and Computation, 2025

    2025

  56. [64]

    Transferable machine learn- ing interatomic potential for bond dissociation energy prediction of drug-like molecules

    Elena Gelzinyte, Mario ¨Oeren, Matthew D Segall, and G ´abor Cs ´anyi. Transferable machine learn- ing interatomic potential for bond dissociation energy prediction of drug-like molecules. Journal of Chemical Theory and Computation, 20(1):164–177, 2023

    2023

  57. [65]

    Featured graph coarsening with similarity guarantees

    Manoj Kumar, Anurag Sharma, Shashwat Saxena, and Sandeep Kumar. Featured graph coarsening with similarity guarantees. In Proceedings of the 40th International Conference on Machine Learning, ICML’23. JMLR.org, 2023

    2023

  58. [66]

    A unified framework for optimization-based graph coarsening

    Manoj Kumar, Anurag Sharma, and Sandeep Kumar. A unified framework for optimization-based graph coarsening. Journal of Machine Learning Research, 24(118):1–50, 2023

    2023

  59. [67]

    Optimization frame- work for semi-supervised attributed graph coarsening

    Manoj Kumar, Subhanu Halder, Archit Kane, Ruchir Gupta, and Sandeep Kumar. Optimization frame- work for semi-supervised attributed graph coarsening. In Negar Kiyavash and Joris M. Mooij, editors, Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence ...

    2024

  60. [68]

    Multi-component coarsened graph learning for scaling graph machine learning

    Subhanu Halder, Manoj Kumar, and Sandeep Kumar. Multi-component coarsened graph learning for scaling graph machine learning. In Companion Proceedings of the ACM on Web Conference 2025 , WWW ’25, page 1001–1004, New York, NY , USA, 2025. Association for Computing Machinery

    2025

  61. [69]

    Graph coarsening with preserved spectral properties

    Yu Jin, Andreas Loukas, and Joseph JaJa. Graph coarsening with preserved spectral properties. In Sil- via Chiappa and Roberto Calandra, editors, Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics , volume 108 of Proceedings of Ma...

    2020

  62. [70]

    Graph reduction with spectral and cut guarantees

    Andreas Loukas. Graph reduction with spectral and cut guarantees. Journal of Machine Learning Research, 20(116):1–42, 2019

    2019

  63. [71]

    Amit Ghosh and Saraswathi Vishveshwara. Variations in clique and community patterns in protein structures during allosteric communication: investigation of dynamically equilibrated structures of methionyl trna synthetase complexes. Biochemistry, 47(44):11398–11407, 2008

    2008

  64. [72]

    The atomic simu- lation environment—a python library for working with atoms

    Ask Hjorth Larsen, Jens Jørgen Mortensen, Jakob Blomqvist, Ivano E Castelli, Rune Christensen, Marcin Dułak, Jesper Friis, Michael N Groves, Bjørk Hammer, Cory Hargus, et al. The atomic simu- lation environment—a python library for working with atoms. Journal of Physics: Conde...

    2017

  65. [73]

    On the role of gradients for machine learning of molecular energies and forces

    Anders S Christensen and O Anatole V on Lilienfeld. On the role of gradients for machine learning of molecular energies and forces. Machine Learning: Science and Technology, 1(4):045018, 2020

    2020

  66. [74]

    Thermodynamic transferability in coarse-grained force fields using graph neural networks

    Emily Shinkle, Aleksandra Pachalieva, Riti Bahl, Sakib Matin, Brendan Gifford, Galen T Craven, and Nicholas Lubbers. Thermodynamic transferability in coarse-grained force fields using graph neural networks. Journal of Chemical Theory and Computation, 20(23):10524–10539, 2024. 19

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.