pith. sign in

arxiv: 2606.30778 · v1 · pith:BCBIYIMUnew · submitted 2026-06-29 · 💻 cs.LG · physics.chem-ph

ReactionAtlas: Ab origine exploration of chemical reaction networks with machine learning

Stefan Gugler , Max Eissler , Khaled Kahouli , Klaus-Robert M\"uller This is my paper

Pith reviewed 2026-07-01 06:44 UTC · model grok-4.3

classification 💻 cs.LG physics.chem-ph
keywords chemical reaction networksmachine learningtransition statespre-biotic chemistryformose cyclegenerative modelsmachine learned force fieldscarbohydrate chemistry
0
0 comments X

The pith

ReactionAtlas builds chemical reaction networks from a handful of seed molecules using machine learning without hand-crafted rules.

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

The paper presents a method to map entire chemical reaction networks starting only from a small number of initial molecules. A generative model proposes candidate reactions while a machine-learned force field identifies which ones correspond to valid transition states, and the resulting products become new starting points for further exploration. This process is repeated iteratively to grow the network. A sympathetic reader would care because it removes the need for both reactants and products as input and for manually defined reaction rules, making it possible to explore large reaction spaces in pre-biotic carbohydrate chemistry at a scale that direct quantum calculations cannot reach.

Core claim

ReactionAtlas builds a reaction network ab origine from a handful of seed molecules and without hand-crafted rules. Specifically, our machine-learned generative model proposes reactions from kinetically sampled candidate compounds and a DFT-trained machine learned force field filters them to valid transition states, the resulting products of which enter the search as new seeds. Starting from eight pre-biotic seeds, ReactionAtlas discovers approximately 47,000 reactions among approximately 12,000 compounds. The MLFF transition states match the PBE0 references within 0.5 Å RMSD in 85 percent of the cases and can be easily brought to the PBE0 level.

What carries the argument

The closed iterative loop in which a generative model proposes reactions and an ML force field validates transition states, then feeds the new products back as seeds.

If this is right

  • The network includes charge and stereo information for compounds up to C4H8O4.
  • Alternative pathways for formose chemistry can be identified within the discovered network.
  • Reaction networks for small carbohydrate chemistry become explorable at a scale of tens of thousands of reactions.
  • Insights into many well-studied reaction paths become available without requiring reactant-product pairs as input.

Where Pith is reading between the lines

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

  • The same iterative proposal-and-validation loop could be applied to other chemical domains such as catalysis or combustion by changing only the initial seeds.
  • Increasing the number or diversity of seed molecules would likely produce even larger networks or different classes of reactions.
  • The generated network could serve as a starting point for experimental verification or for training further models that predict reaction rates.

Load-bearing premise

The machine-learned generative model combined with the DFT-trained MLFF produces a representative sample of chemically valid transition states without systematic omission of important pathways or excessive false positives that would distort the discovered network.

What would settle it

A targeted check showing that one or more well-documented reaction pathways in the formose cycle are absent from the generated network, or that a large fraction of the proposed transition states fail to optimize to stable minima under full PBE0 calculations.

read the original abstract

Mapping a chemical reaction network, the graph of minima and transition states (TS) and the elementary reactions connecting them, is the natural language of chemistry, from catalysis to combustion to the origin of life. Constructing such a reaction network for a given chemistry has been impractical: it requires finding and characterizing tens of thousands of TS, a task for which traditional methods such as density functional theory (DFT) are typically prohibitively slow and require reactant and product as input. We introduce ReactionAtlas, which builds a reaction network $\textit{ab origine}$ from a handful of seed molecules and without hand-crafted rules. Specifically, our machine-learned generative model proposes reactions from kinetically sampled candidate compounds and a DFT-trained machine learned force field (MLFF) filters them to valid TS, the resulting products of which enter the search as new seeds. Starting from eight pre-biotic seeds (CH$_2$O, H$_2$O, OH$^-$, H$_3$O$^+$, CO$_2$, H$_2$CO$_3$, HCO$_3^-$, H), ReactionAtlas discovers $\sim$47,000 reactions among $\sim$12,000 compounds. The MLFF TSs match the PBE0 references within 0.5 \r{A} RMSD in 85% of the cases and can be easily brought to the PBE0 level. Thus, ReactionAtlas maps small carbohydrate chemistry up to C$_4$H$_8$O$_4$ at unprecedented scale and accuracy, including charge and stereo information. It enables novel insights into many well-studied reaction paths, including the formose cycle, which we highlight for its centrality to the chemical origins of life. Notably, our framework also allows establishing alternative reaction pathways for formose chemistry.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces ReactionAtlas, a machine-learning framework that constructs chemical reaction networks ab origine from a small set of seed molecules (eight pre-biotic species) without hand-crafted rules. A generative model proposes reactions from kinetically sampled candidates; a DFT-trained MLFF filters them to valid transition states whose products become new seeds. Applied to small carbohydrate chemistry, the method reports discovery of ~47,000 reactions among ~12,000 compounds up to C4H8O4, with 85% of retained MLFF transition states matching PBE0 references within 0.5 Å RMSD, and illustrates novel insights into the formose cycle.

Significance. If the generative model and MLFF together produce a representative sample of chemically valid transition states without large systematic omissions or false-positive inflation, the work would enable unprecedented automated mapping of reaction networks at the scale of tens of thousands of elementary steps, with direct relevance to pre-biotic chemistry, catalysis, and combustion. The ab origine construction from minimal seeds and the reported geometric fidelity to DFT are notable strengths; the framework's ability to recover and extend known pathways such as formose chemistry would constitute a concrete advance over rule-based or exhaustive-search approaches.

major comments (3)
  1. [Abstract] Abstract: the central claim that the discovered network is representative rests on an 85% geometric match rate for retained TS, yet the abstract supplies neither error bars on this rate, counts of rejected candidates, nor any baseline comparison against rule-based or exhaustive-search methods; without these, it is impossible to assess whether the generative model systematically omits entire reaction classes.
  2. [Abstract] Abstract and §4 (formose discussion): no coverage statistics, ablation on training-data bias, or quantitative comparison of discovered versus known pathways (e.g., the formose cycle) are provided to demonstrate that the kinetically sampled candidates plus generative model yield an unbiased distribution; the 85% match on the retained subset can hold even if large fractions of chemically relevant TS are never proposed.
  3. [Abstract] Abstract: the description of how the generative model was trained or regularized is absent, leaving open the possibility that its learned distribution is narrower than the true space of kinetically accessible reactions and thereby distorts the reported network topology.
minor comments (2)
  1. [Abstract] The manuscript should clarify the precise definition of 'kinetically sampled candidate compounds' and the criteria used by the MLFF to accept or reject a proposed TS.
  2. Figure captions and methods text should explicitly state the training corpus size and composition for both the generative model and the MLFF to allow reproducibility assessment.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive critique of the abstract and the emphasis on demonstrating representativeness. We agree that the abstract can be strengthened with additional quantitative context and will revise it accordingly. Below we respond point by point to the major comments.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the discovered network is representative rests on an 85% geometric match rate for retained TS, yet the abstract supplies neither error bars on this rate, counts of rejected candidates, nor any baseline comparison against rule-based or exhaustive-search methods; without these, it is impossible to assess whether the generative model systematically omits entire reaction classes.

    Authors: We agree the abstract should convey more information on the validation statistics. The 85% RMSD match is computed on the retained TS after MLFF filtering; the full manuscript reports the underlying validation set size and per-molecule breakdown. In revision we will add (i) error bars obtained via bootstrap resampling of the validation set and (ii) the number of candidates rejected at the MLFF stage. Direct quantitative baselines against rule-based enumerators are difficult because those methods require hand-crafted templates that do not exist for the full C4 carbohydrate space; we will nevertheless add a short paragraph contrasting the scale achieved here with published rule-based networks for smaller systems. revision: yes

  2. Referee: [Abstract] Abstract and §4 (formose discussion): no coverage statistics, ablation on training-data bias, or quantitative comparison of discovered versus known pathways (e.g., the formose cycle) are provided to demonstrate that the kinetically sampled candidates plus generative model yield an unbiased distribution; the 85% match on the retained subset can hold even if large fractions of chemically relevant TS are never proposed.

    Authors: We acknowledge that the abstract and §4 do not contain explicit coverage statistics or training-data ablations. The iterative seed-expansion procedure is designed to reduce bias by continually adding newly discovered products, and §4 shows that the method recovers the canonical formose cycle plus several alternative routes. In the revision we will (i) report the fraction of literature-reported formose reactions that appear in our network and (ii) add a brief discussion of possible training-data bias together with the observation that the MLFF filter is trained on a broad DFT dataset independent of the generative model. A full ablation study would require additional compute and is noted as future work. revision: partial

  3. Referee: [Abstract] Abstract: the description of how the generative model was trained or regularized is absent, leaving open the possibility that its learned distribution is narrower than the true space of kinetically accessible reactions and thereby distorts the reported network topology.

    Authors: The abstract is intentionally concise; the training procedure (including the kinetic sampling of candidates, the architecture, and the regularization terms used to encourage diversity) is fully described in the Methods section. We will expand the abstract by one sentence summarizing the generative-model training objective and the diversity-promoting regularization to make this information immediately visible. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper describes an iterative discovery process that begins with a fixed set of eight external seed molecules and applies a generative model plus MLFF (both trained against independent DFT reference data) to propose and filter reactions, adding new products as seeds. The reported counts (~47k reactions, ~12k compounds) and the 85% RMSD match statistic are direct outputs of running this loop; they are not defined in terms of themselves, fitted to a subset and then re-predicted, or justified solely by self-citation. No equations or steps reduce the central claims to inputs by construction, and the method relies on standard external benchmarks rather than internal re-labeling.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim depends on the trained generative model and MLFF being sufficiently accurate and unbiased; these models contain large numbers of fitted parameters whose values are not reported. The only explicit domain assumption is that the MLFF can serve as a reliable proxy for PBE0 transition-state searches.

free parameters (2)
  • generative model parameters
    Weights of the machine-learned generative model that proposes reactions; fitted during training on unspecified data.
  • MLFF parameters
    Weights of the machine-learned force field trained on PBE0 data; used to filter transition states.
axioms (1)
  • domain assumption An ML force field trained on PBE0 calculations can identify valid transition states with sufficient accuracy for network exploration.
    Invoked when the MLFF is used to accept or reject candidate reactions.

pith-pipeline@v0.9.1-grok · 5865 in / 1446 out tokens · 33593 ms · 2026-07-01T06:44:48.580834+00:00 · methodology

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

164 extracted references · 30 canonical work pages · 2 internal anchors

  1. [1]

    J., Stark, S

    Broadbelt, L. J., Stark, S. M. & Klein, M. T. Computer generated pyrolysis modeling: On-the-fly generation of species, reactions, and rates.Ind. Eng. Chem. Res.33, 790–799 (1994)

    1994

  2. [2]

    J., Stark, S

    Broadbelt, L. J., Stark, S. M. & Klein, M. T. Computer generated reac- tion modelling: Decomposition and encoding algorithms for determining species uniqueness.Comput. Chem. Eng.20, 113–129 (1996)

    1996

  3. [3]

    Broadbelt, L. J. & Pfaendtner, J. Lexicography of kinetic modeling of complex reaction networks.AIChE J.51, 2112–2121 (2005)

    2005

  4. [4]

    Fialkowski, M., Bishop, K. J. M., Chubukov, V. A., Campbell, C. J. & Grzy- bowski, B. A. Architecture and Evolution of Organic Chemistry.Angew. Chem. Int. Ed.44, 7263–7269 (2005). URL https://onlinelibrary.wiley.com/doi/abs/ 10.1002/anie.200502272

    work page doi:10.1002/anie.200502272 2005

  5. [5]

    M.et al.Rewiring chemistry: Algorithmic discovery and experimen- tal validation of one-pot reactions in the network of organic chemistry.Angew

    Gothard, C. M.et al.Rewiring chemistry: Algorithmic discovery and experimen- tal validation of one-pot reactions in the network of organic chemistry.Angew. Chem. Int. Ed.51, 7922–7927 (2012)

    2012

  6. [6]

    Kowalik, M.et al.Parallel optimization of synthetic pathways within the network of organic chemistry.Angew. Chem. Int. Ed.51, 7928–7932 (2012)

    2012

  7. [7]

    Wo los, A.et al.Synthetic connectivity, emergence, and self-regeneration in the network of prebiotic chemistry.Science369, eaaw1955 (2020)

    2020

  8. [8]

    Feinberg, M.Foundations of Chemical Reaction Network TheoryApplied Mathematical Sciences (Springer International Publishing, Cham, Switzerland, 2019)

    2019

  9. [9]

    L., Arg¨ uelles, A

    Dewyer, A. L., Arg¨ uelles, A. J. & Zimmerman, P. M. Methods for exploring reaction space in molecular systems.WIREs Comput. Mol. Sci.8, e1354 (2018). URL https://onlinelibrary.wiley.com/doi/abs/10.1002/wcms.1354

    work page doi:10.1002/wcms.1354 2018

  10. [10]

    Dewyer, A. L. & Zimmerman, P. M. Finding reaction mechanisms, intuitive or otherwise.Org. Biomol. Chem.15, 501–504 (2017). URL http://dx.doi.org/10. 1039/C6OB02183B

    2017

  11. [11]

    J., Bell, A

    Garza, A. J., Bell, A. T. & Head-Gordon, M. Mechanism of CO 2 reduction at copper surfaces: Pathways to C 2 products.ACS Catal.8, 1490–1499 (2018). URL http://dx.doi.org/10.1021/acscatal.7b03477

    work page doi:10.1021/acscatal.7b03477 2018

  12. [12]

    W., Bui, J

    Lees, E. W., Bui, J. C., Song, D., Weber, A. Z. & Berlinguette, C. P. Con- tinuum model to define the chemistry and mass transfer in a bicarbonate electrolyzer.ACS Energy Lett.7, 834–842 (2022). URL http://dx.doi.org/10. 1021/acsenergylett.1c02522

    2022

  13. [13]

    Wen, M.et al.Chemical reaction networks and opportunities for machine learning.Nat. Comput. Sci.3, 12–24 (2023)

    2023

  14. [14]

    & Caflisch, A

    Rao, F. & Caflisch, A. The protein folding network.J. Mol. Biol.342, 299–306 (2004)

    2004

  15. [15]

    & Fischer, S

    No´ e, F. & Fischer, S. Transition networks for modeling the kinetics of con- formational change in macromolecules.Curr. Opin. Struct. Biol.18, 154–162 56 (2008)

    2008

  16. [16]

    Cerjan, C. J. & Miller, W. H. On finding transition states.J. Chem. Phys.75, 2800–2806 (1981). URL http://dx.doi.org/10.1063/1.442352

    work page doi:10.1063/1.442352 1981

  17. [17]

    & Shepard, R

    Banerjee, A., Adams, N., Simons, J. & Shepard, R. Search for stationary points on surfaces.J. Phys. Chem.89, 52–57 (1985). URL http://dx.doi.org/10.1021/ j100247a015

    1985

  18. [18]

    An algorithm for the location of transition states.J

    Baker, J. An algorithm for the location of transition states.J. Comput. Chem. 7, 385–395 (1986). URL http://dx.doi.org/10.1002/jcc.540070402

    work page doi:10.1002/jcc.540070402 1986

  19. [19]

    Bofill, J. M. Updated hessian matrix and the restricted step method for locating transition structures.J. Comput. Chem.15, 1–11 (1994)

    1994

  20. [20]

    & Schenter, G

    Mills, G., J´ onsson, H. & Schenter, G. K. Reversible work transition state theory: Application to dissociative adsorption of hydrogen.Surf. Sci. 324, 305–337 (1995). URL https://www.sciencedirect.com/science/article/pii/ 0039602894007314

    1995

  21. [21]

    & Jacobsen, K

    Jonsson, H., Mills, G. & Jacobsen, K. W. Nudged elastic band method for finding minimum energy paths of transitions (1998). InClassical and Quantum Dynamics in Condensed Phase Simulations, World Scientific, 385–404

    1998

  22. [22]

    Henkelman, G., Uberuaga, B. P. & J´ onsson, H. A climbing image nudged elastic band method for finding saddle points and minimum energy paths.J. Chem. Phys.113, 9901–9904 (2000)

    2000

  23. [23]

    Reliable Transition State Searches Integrated with the Growing String Method.J

    Zimmerman, P. Reliable Transition State Searches Integrated with the Growing String Method.J. Chem. Theory Comput.9, 3043–3050 (2013). URL https: //doi.org/10.1021/ct400319w

    work page doi:10.1021/ct400319w 2013

  24. [24]

    Zimmerman, P. M. Single-ended transition state finding with the growing string method.J. Comput. Chem.36, 601–611 (2015). URL https://onlinelibrary. wiley.com/doi/abs/10.1002/jcc.23833

    work page doi:10.1002/jcc.23833 2015

  25. [25]

    & J´ onsson, H

    Henkelman, G. & J´ onsson, H. A dimer method for finding saddle points on high dimensional potential surfaces using only first derivatives.J. Chem. Phys.111, 7010–7022 (1999)

    1999

  26. [26]

    & Harabuchi, Y

    Maeda, S. & Harabuchi, Y. Exploring paths of chemical transformations in molecular and periodic systems: An approach utilizing force.WIREs Comput. Mol. Sci.11, e1538 (2021). URL http://dx.doi.org/10.1002/wcms.1538

    work page doi:10.1002/wcms.1538 2021

  27. [27]

    & Ohno, K

    Maeda, S. & Ohno, K. Global mapping of equilibrium and transition structures on potential energy surfaces by the scaled hypersphere search method: Appli- cations to ab initio surfaces of formaldehyde and propyne molecules.J. Phys. Chem. A109, 5742–5753 (2005)

    2005

  28. [28]

    & Morokuma, K

    Maeda, S., Komagawa, S., Uchiyama, M. & Morokuma, K. Finding reac- tion pathways for multicomponent reactions: The passerini reaction is a four-component reaction.Angew. Chem. Int. Ed.50, 644–649 (2011)

    2011

  29. [29]

    & Morokuma, K

    Maeda, S. & Morokuma, K. Finding reaction pathways of type A + B→X: Toward systematic prediction of reaction mechanisms.J. Chem. Theory Comput. 7, 2335–2345 (2011). 57

    2011

  30. [30]

    & Morokuma, K

    Maeda, S., Taketsugu, T. & Morokuma, K. Exploring transition state structures for intramolecular pathways by the artificial force induced reaction method.J. Comput. Chem.35, 166–173 (2014)

    2014

  31. [31]

    Sameera, W. M. C., Maeda, S. & Morokuma, K. Computational catalysis using the artificial force induced reaction method.Acc. Chem. Res.49, 763–773 (2016)

    2016

  32. [32]

    N., Proppe, J

    Bergeler, M., Simm, G. N., Proppe, J. & Reiher, M. Heuristics-guided explo- ration of reaction mechanisms.J. Chem. Theory Comput.11, 5712–5722 (2015)

    2015

  33. [33]

    Simm, G. N. & Reiher, M. Context-Driven Exploration of Complex Chemical Reaction Networks.J. Chem. Theory Comput.13, 6108–6119 (2017)

    2017

  34. [34]

    P., Grimmel, S

    Unsleber, J. P., Grimmel, S. A. & Reiher, M. Chemoton 2.0: Autonomous exploration of chemical reaction networks.J. Chem. Theory Comput.18, 5393– 5409 (2022)

    2022

  35. [35]

    J., Zubarev, D

    Rappoport, D., Galvin, C. J., Zubarev, D. Y. & Aspuru-Guzik, A. Complex Chemical Reaction Networks from Heuristics-Aided Quantum Chemistry.J. Chem. Theory Comput.10, 897–907 (2014)

    2014

  36. [36]

    & Savoie, B

    Zhao, Q. & Savoie, B. M. Deep reaction network exploration of glucose pyrolysis. Proc. Natl. Acad. Sci. U.S.A.120, e2305884120 (2023)

    2023

  37. [37]

    & Savoie, B

    Woulfe, M. & Savoie, B. M. Chemical Reaction Networks from Scratch with Reaction Prediction and Kinetics-Guided Exploration.J. Chem. Theory Comput.21, 1276–1291 (2025)

    2025

  38. [38]

    Chem.6, 1044–1048 (2014)

    Wang, L.-P.et al.Discovering chemistry with an ab initio nanoreactor.Nat. Chem.6, 1044–1048 (2014)

    2014

  39. [39]

    T., Pande, V

    Wang, L.-P., McGibbon, R. T., Pande, V. S. & Mart´ ınez, T. J. Automated discovery and refinement of reactive molecular dynamics pathways.J. Chem. Theory Comput.12, 638–649 (2016)

    2016

  40. [40]

    & Ochsenfeld, C

    Stan, A., von der Esch, B. & Ochsenfeld, C. Fully automated generation of prebiotically relevant reaction networks from optimized nanoreactor simulations. J. Chem. Theory Comput.18, 6700–6712 (2022)

    2022

  41. [41]

    & Ochsenfeld, C

    Stan-Bernhardt, A., Glinkina, L., Hulm, A. & Ochsenfeld, C. Exploring Chemi- cal Space Using Ab Initio Hyperreactor Dynamics.ACS Cent. Sci.10, 302–314 (2024)

    2024

  42. [42]

    Chem.16, 727–734 (2024)

    Zhang, S.et al.Exploring the frontiers of condensed-phase chemistry with a general reactive machine learning potential.Nat. Chem.16, 727–734 (2024)

    2024

  43. [43]

    & Stirnemann, G

    Benayad, Z., David, R. & Stirnemann, G. Prebiotic chemical reactivity in solu- tion with quantum accuracy and microsecond sampling using neural network potentials.Proc. Natl. Acad. Sci. U.S.A.121, e2322040121 (2024)

    2024

  44. [44]

    & Saitta, A

    Huet, L., Devergne, T., Magrino, T. & Saitta, A. M. A New Route to the Pre- biotic Synthesis of Glycine via Ab Initio-Based Machine Learning Calculations. J. Phys. Chem. Lett.15, 8697–8705 (2024). 58

    2024

  45. [45]

    Modeling prebiotic chemistries with quantum accuracy at classical costs.Proc

    Tiwary, P. Modeling prebiotic chemistries with quantum accuracy at classical costs.Proc. Natl. Acad. Sci. U.S.A.121, e2408742121 (2024)

    2024

  46. [46]

    Zenodo (2022)

    Baiardi, A.et al.Qcscine/Utilities: Release 5.0.0. Zenodo (2022)

    2022

  47. [47]

    N., Vaucher, A

    Simm, G. N., Vaucher, A. C. & Reiher, M. Exploration of reaction pathways and chemical transformation networks.J. Phys. Chem. A123, 385–399 (2019)

    2019

  48. [48]

    T¨ urtscher, P. L. & Reiher, M. Pathfinder—navigating and analyzing chemical reaction networks with an efficient graph-based approach.J. Chem. Inf. Model. 63, 147–160 (2023)

    2023

  49. [49]

    & Reiher, M

    Steiner, M. & Reiher, M. A human-machine interface for automatic exploration of chemical reaction networks.Nat. Commun.15, 3680 (2024)

    2024

  50. [50]

    Zhao, Q.et al.Comprehensive exploration of graphically defined reaction spaces. Sci. Data10, 145 (2023)

    2023

  51. [51]

    Mart´ ınez-N´ u˜ nez, E.et al.AutoMeKin2021: An open-source program for automated reaction discovery.J. Comput. Chem.42, 2036–2048 (2021)

    2036

  52. [52]

    & Savoie, B

    Zhao, Q., Xu, Y., Greeley, J. & Savoie, B. M. Deep reaction network exploration at a heterogeneous catalytic interface.Nat. Commun.13, 4860 (2022). URL http://dx.doi.org/10.1038/s41467-022-32514-7

    work page doi:10.1038/s41467-022-32514-7 2022

  53. [53]

    & Kim, W

    Kim, S., Woo, J. & Kim, W. Y. Diffusion-based generative AI for exploring transition states from 2D molecular graphs.Nat. Commun.15, 341 (2024). URL https://www.nature.com/articles/s41467-023-44629-6

    2024

  54. [54]

    & Kulik, H

    Duan, C., Du, Y., Jia, H. & Kulik, H. J. Accurate transition state genera- tion with an object-aware equivariant elementary reaction diffusion model.Nat. Comput. Sci.3, 1045–1055 (2023). URL https://www.nature.com/articles/ s43588-023-00563-7

    2023

  55. [55]

    Duan, C.et al.Optimal transport for generating transition states in chemical reactions.Nat. Mach. Intell.7, 615–626 (2025). URL http://dx.doi.org/10. 1038/s42256-025-01010-0

    2025

  56. [56]

    Drift-React: One-step Generation of Reaction Pathways via SE(3) Drifting Fields

    Schlama, R. & Schwaller, P. Drift-React: One-step Generation of Reaction Pathways via SE(3) Drifting Fields. arXiv:2605.22990 (2026). arXiv:2605.22990

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  57. [57]

    A., Pattanaik, L

    Grambow, C. A., Pattanaik, L. & Green, W. H. Reactants, products, and tran- sition states of elementary chemical reactions based on quantum chemistry.Sci. Data7, 137 (2020). URL https://www.nature.com/articles/s41597-020-0460-4

    2020

  58. [58]

    D.et al.Optimizing transition states via kernel-based machine learn- ing.J

    Pozun, Z. D.et al.Optimizing transition states via kernel-based machine learn- ing.J. Chem. Phys.136, 174101 (2012). URL https://aip.scitation.org/doi/10. 1063/1.4707167

    2012

  59. [59]

    A., Jennings, P

    Garrido Torres, J. A., Jennings, P. C., Hansen, M. H., Boes, J. R. & Bligaard, T. Low-Scaling Algorithm for Nudged Elastic Band Calculations Using a Surrogate Machine Learning Model.Phys. Rev. Lett.122, 156001 (2019). URL https: //link.aps.org/doi/10.1103/PhysRevLett.122.156001

    work page doi:10.1103/physrevlett.122.156001 2019

  60. [60]

    & K¨ astner, J

    Denzel, A. & K¨ astner, J. Gaussian process regression for transition state search. J. Chem. Theory Comput.14, 5777–5786 (2018). URL http://dx.doi.org/10. 1021/acs.jctc.8b00708. 59

    2018

  61. [61]

    Peterson, A. A. Acceleration of saddle-point searches with machine learning.J. Chem. Phys.145, 074106 (2016). URL https://doi.org/10.1063/1.4960708

    work page doi:10.1063/1.4960708 2016

  62. [62]

    B.,´Asgeirsson, V., Vehtari, A

    Koistinen, O.-P., Dagbjartsd´ ottir, F. B.,´Asgeirsson, V., Vehtari, A. & J´ onsson, H. Nudged elastic band calculations accelerated with gaussian process regression. J. Chem. Phys.147, 152720 (2017). URL http://dx.doi.org/10.1063/1.4986787

    work page doi:10.1063/1.4986787 2017

  63. [63]

    & Hauser, A

    Meyer, R. & Hauser, A. W. Geometry optimization using Gaussian process regression in internal coordinate systems.J. Chem. Phys.152, 084112 (2020). URL https://doi.org/10.1063/1.5144603

    work page doi:10.1063/1.5144603 2020

  64. [64]

    & K¨ astner, J

    Born, D. & K¨ astner, J. Geometry Optimization in Internal Coordinates Based on Gaussian Process Regression: Comparison of Two Approaches.J. Chem. Theory Comput.17, 5955–5967 (2021). URL https://doi.org/10.1021/acs.jctc.1c00517

    work page doi:10.1021/acs.jctc.1c00517 2021

  65. [65]

    C.-Y.et al.Deep Learning of ab initio Hessians for Transition State Optimization (2024)

    Yuan, E. C.-Y.et al.Deep Learning of ab initio Hessians for Transition State Optimization (2024). URL http://arxiv.org/abs/2405.02247. ArXiv:2405.02247

    work page arXiv 2024

  66. [66]

    Li, B., Xiao, J., Gao, Y., Zhang, J. Z. H. & Zhu, T. Transition State Searching Accelerated by Neural Network Potential.J. Chem. Inf. Model.65, 2297–2303 (2025)

    2025

  67. [67]

    B., Grambow, C

    Pattanaik, L., Ingraham, J. B., Grambow, C. A. & Green, W. H. Gener- ating transition states of isomerization reactions with deep learning.Phys. Chem. Chem. Phys.22, 23618–23626 (2020). URL http://dx.doi.org/10.1039/ D0CP04670A

    2020

  68. [68]

    Z., Verma, N., Larson, E

    Mako´ s, M. Z., Verma, N., Larson, E. C., Freindorf, M. & Kraka, E. Generative adversarial networks for transition state geometry prediction.J. Chem. Phys. 155, 024116 (2021). URL https://doi.org/10.1063/5.0055094

    work page doi:10.1063/5.0055094 2021

  69. [69]

    Prediction of transition state structures of gas-phase chemical reactions via machine learning.Nat

    Choi, S. Prediction of transition state structures of gas-phase chemical reactions via machine learning.Nat. Commun.14, 1168 (2023). URL https://www. nature.com/articles/s41467-023-36823-3

    2023

  70. [70]

    & Pearson, J

    Jackson, R., Zhang, W. & Pearson, J. TSNet: Predicting transition state structures with tensor field networks and transfer learning.Chem. Sci. 12, 10022–10040 (2021). URL https://pubs.rsc.org/en/content/articlelanding/ 2021/sc/d1sc01206a

    2021

  71. [71]

    Lemm, D., Rudorff, G. F. & von Lilienfeld, O. A. Machine learning based energy-free structure predictions of molecules, transition states, and solids. Nat. Commun.12, 4468 (2021). URL https://www.nature.com/articles/ s41467-021-24525-7

    2021

  72. [72]

    Heinen, S., von Rudorff, G. F. & von Lilienfeld, O. A. Transition state search and geometry relaxation throughout chemical compound space with quantum machine learning.J. Chem. Phys.157, 221102 (2022)

    2022

  73. [73]

    & Winther, O

    Schreiner, M., Bhowmik, A., Vegge, T., Busk, J. & Winther, O. Transition1x— a dataset for building generalizable reactive machine learning potentials.Sci. Data9, 779 (2022)

    2022

  74. [74]

    W., Pemberton, M

    Beaglehole, I. W., Pemberton, M. J., Farrar, E. H. E. & Grayson, M. N. Machine Learning Transition State Geometries and Applications in Reaction Property Prediction.WIREs Comput. Mol. Sci.15, e70025 (2025). 60

    2025

  75. [75]

    Eissler, M.et al.How simple can you go? an off-the-shelf transformer approach to molecular dynamics.J. Chem. Phys.164, 094308 (2026)

    2026

  76. [76]

    T., Unke, O

    Sch¨ utt, K. T., Unke, O. T. & Gastegger, M. Equivariant message passing for the prediction of tensorial properties and molecular spectra (2021). InProceedings of the 38th International Conference on Machine Learning, PMLR 139, pp. 9377–9388

    2021

  77. [77]

    & Parrinello, M

    Behler, J. & Parrinello, M. Generalized Neural-Network Representation of High- Dimensional Potential-Energy Surfaces.Phys. Rev. Lett.98, 146401 (2007). URL https://link.aps.org/doi/10.1103/PhysRevLett.98.146401

    work page doi:10.1103/physrevlett.98.146401 2007

  78. [78]

    & von Lilienfeld, O

    Rupp, M., Tkatchenko, A., M¨ uller, K.-R. & von Lilienfeld, O. A. Fast and Accurate Modeling of Molecular Atomization Energies with Machine Learning. Phys. Rev. Lett.108, 058301 (2012). URL https://link.aps.org/doi/10.1103/ PhysRevLett.108.058301

    2012

  79. [79]

    P., Cs´ anyi, G

    De, S., Bart´ ok, A. P., Cs´ anyi, G. & Ceriotti, M. Comparing molecules and solids across structural and alchemical space.Phys. Chem. Chem. Phys.18, 13754– 13769 (2016). URL https://pubs.rsc.org/en/content/articlelanding/2016/cp/ c6cp00415f

    2016

  80. [80]

    S., Bratholm, L

    Christensen, A. S., Bratholm, L. A., Faber, F. A. & Anatole von Lilienfeld, O. FCHL revisited: Faster and more accurate quantum machine learning.J. Chem. Phys.152, 044107 (2020). URL http://aip.scitation.org/doi/10.1063/1.5126701

    work page doi:10.1063/1.5126701 2020

Showing first 80 references.