Pith. sign in

REVIEW 5 major objections 6 minor 56 references

A potassium ion channel simulated with a universal neural network potential

T0 review · 5 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A hydrogen bond between threonine T75 and water in the selectivity filter stabilizes water and enables soft knock-on potassium transport, according to neural-network-potential simulations.

desk verdict A new, reproducible T75-water H-bond in a KcsA selectivity filter simulation, with an honest but unproven claim that it enables physiological soft knock-on. read the letter →

arxiv 2411.18931 v1 pith:VFGNN4AA submitted 2024-11-28 q-bio.BM cond-mat.soft

Timothy T. Duignan This is my paper · ORCID
classification q-bio.BMcond-mat.soft MSC 92C4092C05
keywords potassiumchannelKcsAselectivityfilterneuralnetworkpotentialmoleculardynamicssoftknock-onwaterco-transporthydrogenbond
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 argues that a specific, previously overlooked hydrogen bond is what lets potassium ions flow through the KcsA channel at near-physiological speed. Simulating the selectivity filter with a universal neural network potential, the author observes water molecules being carried through the filter alongside potassium ions — a 'soft knock-on' mechanism that classical force-field simulations have failed to reproduce. The key actor is the hydroxyl side group of the threonine T75 residue, which reaches into the channel entrance and holds a water molecule in place long enough for the next ion to push the whole column forward. If this is right, it would resolve a long-standing debate about whether water co-transport is part of the conduction mechanism, and it would explain why standard simulations underestimate conductance by an order of magnitude. The simulated conductance of 80 ± 20 pS falls inside the experimental range of 40–250 pS.

What carries the argument

The central objects are the Orb-D3 universal neural network potential — trained on crystal-structure DFT, used here to supply forces for a nanosecond-scale simulation — and the hydrogen bond between the T75 side-chain hydroxyl and a water molecule at the S4 entrance of the selectivity filter. This hydrogen bond is the load-bearing interaction: it anchors water long enough to be co-transported, and its absence in classical force fields is what the paper proposes explains their hard knock-on behavior and low conductances.

What would settle it

Run the same simulation setup at a physiological driving force (0.2 V or less) with longer simulation time, or with the T75 hydroxyl removed via a T75C mutation: if the soft knock-on water co-transport and the T75–water hydrogen bond no longer appear and conductance drops to the underestimated values of classical force fields, the claim that this bond is the physiological mechanism collapses.

Watch

Extended reading notes

Core claim

Using the Orb-D3 neural network potential on an 8,450-atom cylinder containing the KcsA selectivity filter, the paper reports a conduction cycle in which a water molecule enters the S4 site, is stabilized by a hydrogen bond between the T75 side-chain hydroxyl and the water, then hops to S3 as potassium ions advance, completing a soft knock-on event. The same mechanism appeared in six independent trajectories, giving a conductance of 80 ± 20 pS that falls inside the experimental 40–250 pS range. The paper further argues that the T75 hydroxyl is causally important, citing the known T75C mutation experiment in which removing this hydroxyl lowers potassium conductance to rubidium-like levels. It also reports water-induced carbonyl flips at G77, V76 and T74 sites that have not all been seen before.

Load-bearing premise

The 0.1 eV/Å driving force (about 5 V across the membrane) plus the frozen outer cylinder of protein is assumed to reproduce the physiological conduction mechanism; the paper itself notes this force is far above physiological voltages and may distort the mechanism.

Editorial extensions

If this is right

  • If water co-transport is real, the selectivity filter is not a purely ionic pore; water moves with ions, affecting osmotic and energetic balances across the membrane.
  • The T75 hydroxyl becomes a concrete molecular target: mutations that remove or alter this hydrogen bond should reduce conductance, as the T75C mutation already does experimentally.
  • The observed carbonyl flips at G77, V76 and T74 may explain low S2 ion occupancy and could be linked to c-type inactivation, connecting the conduction mechanism to channel gating.
  • Universal neural network potentials trained on crystal data can capture flexible protein dynamics that classical force fields miss, opening membrane-protein simulation to these tools.
  • A conductance within the experimental range supports the soft knock-on mechanism as the physiological one, rather than the hard knock-on seen in classical molecular dynamics.

Reading between the lines

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

  • A testable prediction follows: if the T75–water hydrogen bond is the stabilizer, then a T75S mutation, which keeps a hydroxyl but changes geometry, should alter but not abolish water co-transport, whereas T75A or T75C should abolish it — distinguishing hydrogen-bond geometry from mere steric effects.
  • The 5 V driving force may bias the observed mechanism; a stronger test would be enhanced-sampling calculations of the potential of mean force for water entry into S4 with and without the T75 hydroxyl, at zero applied voltage.
  • The paper's validation of Orb-D3 against a custom NNP for bulk KCl suggests the model's ion-water interactions are reasonable, but the selectivity filter is far outside training data; comparing predicted water occupancy and ion distributions against crystallographic or 2D-IR data would be a more stringent check.
  • If water co-transport is confirmed, it implies that K+/Na+ selectivity may partly operate through the energy of dehydrating and rehydrating water in the filter, not only through carbonyl coordination of the ions.
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

5 major / 6 minor

Summary. The manuscript reports ~1 ns molecular dynamics simulations of the selectivity filter (SF) of the KcsA potassium channel using the universal neural network potential Orb-D3-v2. The author observes a hydrogen bond between the T75 side-chain hydroxyl and a water molecule at the SF entrance, and reports a 'soft knock-on' transport mechanism in which water is co-transported with K+ through the SF, with an estimated conductance of 80±20 pS. Additional observations include carbonyl flipping of G77, V76, and T74 residues, and the absence of soft knock-on when the SF is initialized with only K+ ions. The paper argues that the T75 hydroxyl stabilizes water in the SF, enabling selective rapid K+ conduction, and that the T75C mutation experiment provides supporting evidence. The claimed implication is that universal NNPs can reveal biological mechanisms inaccessible to classical force fields.

Significance. If the central claim were established, the paper would be significant in three respects: it would identify a previously unnoticed structural element (the T75–water hydrogen bond) as a key stabilizer of water in the selectivity filter; it would provide a concrete mechanism for water co-transport during K+ conduction, a long-debated question; and it would demonstrate that a universal neural network potential trained on crystals can produce stable, mechanistically informative simulations of a membrane protein. The paper also makes beneficial contributions in sharing input files and analysis scripts, and in including a bulk-electrolyte validation of the potential. However, the current evidence does not yet support the causal claim: the simulations are driven at 0.1 eV/Å (~5 V), the transport mechanism is shown to be force-dependent, the conductance estimate is based on non-equilibrium forced transit times, and the T75C experiment provides only circumstantial support. The central claim therefore remains a plausible hypothesis rather than an established finding, and the paper's abstract and conclusions should be tempered accordingly.

major comments (5)
  1. [Sections 3.1 and 3.2] The central claim that the T75–water hydrogen bond enables soft knock-on transport as the physiological mechanism is inferred from simulations driven at 0.1 eV/Å, which the paper itself equates to ~5 V across a 50 Å membrane in Section 3.2. As the author notes in Section 4, this force 'may induce distortions in the transport mechanism.' More importantly, Section 3.1 shows that at forces greater than 0.1 eV/Å a hard knock-on mechanism occurs, while at exactly 0.1 eV/Å soft knock-on appears. This demonstrates that the observed mechanism is force-dependent, and there is no evidence that the 0.1 eV/Å branch is the physiologically relevant one rather than a force-induced crossover. To support the claim, the paper would need either a systematic force-dependence study showing a stable soft-knock-on regime at lower forces, or a free-energy calculation (e.g., PMF) that does not rely on an external bias.
  2. [Section 3.2] The conductance estimate of 80±20 pS is derived from the average transit time of a full conduction cycle under a 5 V applied force, with the membrane thickness assumed to be 50 Å, and only the six soft-knock-on runs are averaged while the one hard-knock-on run is excluded. This is not a measure of the physiological single-channel conductance, since the applied force drives the system far from equilibrium, and the error bar reflects only the spread between six short trajectories, not statistical or systematic uncertainty. The claim that this value 'falls within the experimental range' in Section 3.2 is therefore misleading; the conductance should be reported as a rough consistency check under non-physiological drive, or removed.
  3. [Section 3.3] The T75C mutation data from Ref. [8] are presented as strong evidence that the T75 hydroxyl group is causally important for potassium conduction. However, that experiment demonstrates only that removing the hydroxyl reduces conductance; it does not show that the hydroxyl acts by hydrogen-bonding to a water molecule in the SF, nor that water co-transport is part of the conductive pathway. The connection between the mutation phenotype and the specific soft knock-on mechanism observed here is an untested inference. The paper should phrase this as a consistency argument rather than as validation of the proposed mechanism.
  4. [Sections 2 and 4] The simulation uses a 20 Å radius cylinder with atoms beyond 15 Å frozen, and Section 4 notes that the SF entrance gradually dehydrates over the course of the simulation due to movements of the surrounding protein and the frozen boundary. Such dehydration could either create or destroy the T75–water hydrogen bond that is central to the proposed mechanism, and no control simulation with a larger or flexible boundary is provided to rule this out. The RDF in Figure 5 is a static structural observable of the T75–T74 interaction, not a direct measure of the water hydrogen bond; the paper would be strengthened by quantifying the lifetime and occupancy of the water–T75 hydrogen bond over the trajectory and correlating it with the conduction events.
  5. [Section 3.4 and Conclusion] The observation in Section 3.4 that initializing the SF with four potassium ions and no water produces a hard knock-on mechanism, with the T75 hydroxyl instead hydrogen-bonding to a water molecule outside the SF, shows that the soft knock-on mechanism is not a unique consequence of the T75–water interaction but depends on the initial occupancy and the driving conditions. This is not necessarily a problem, but it should be acknowledged that the soft knock-on mechanism may be one of several conductive states, and the paper's conclusion that T75 'plays a crucial role' in the 'rapid, specific transport' (Section 3.3) is too strong given this variability.
minor comments (6)
  1. [Section 3.5] There is a stray 's' at the end of the section; please correct. Additionally, the terminology 'Orb-D3' and 'Orb-D3-v2' is used inconsistently; please settle on one name.
  2. [Section 2] The Langevin friction coefficient of 0.01 is given without units; please specify whether this is in units of 1/fs or another unit, and briefly justify the choice.
  3. [Figure 4] The caption refers to 'dark blue' for the hydrogen-bond indicator, but the color legend is not defined. Please add a legend or explicitly describe the color coding.
  4. [Section 1.2] The statement that NNPs 'cannot handle potassium ions' is too broad; there are NNP models that include K+ in electrolytes. Please qualify the statement to refer to the specific models intended for biological simulations at the time.
  5. [Section 2] The average temperature is reported as 307 K despite the thermostat set to 300 K, attributed to 'a small amount of noise on the forces.' A 7 K drift may indicate a thermostat or integration issue; please investigate and report the temperature profile over time, and assess whether this affects the structural conclusions.
  6. [Section 3.4] The observation of H25 deprotonation is intriguing, but the paper does not discuss whether this could be an artifact of the frozen boundaries or the large applied force. Please add a note on possible artifacts or a control analysis.

Circularity Check

1 steps flagged · score 2.0 of 10

No significant circularity: the T75–water H-bond and 80 ± 20 pS conductance are emergent simulation outputs confirmed by an external mutation experiment; only a minor non-load-bearing self-citation in the model-validation section flags the score at 2.

  1. other [Section 3.5, Bulk electrolyte (model validation)]
    "One objection to the use of the Orb-D3 model could be that it is trained on both neutral and charged potassium ions and hence might not reliably simulate potassium ions, as it cannot distinguish their charge state. To test this, a 2.4 M KCl aqueous solution was simulated and compared the K–O and K-K RDFs with another recently reported neural network potential simulation,[52] which was custom-trained on higher-quality and directly relevant training data."

    Ref [52] is the author's own prior work (Zhang, Pagotto, Gould, Duignan), so Orb-D3's potassium-in-water behavior is validated partly against the author's own custom-trained NNP; agreement between two models from the same group could reflect shared systematic biases rather than an independent check. However, this is not the central claim: the T75-hydroxyl–water hydrogen bond, the soft knock-on mechanism, and the 80 ± 20 pS conductance all arise from the SF trajectories, and the causal role of the T75 hydroxyl is supported by the external T75C mutation experiment (Ref [8], Zhou & MacKinnon 2004), not by a self-citation. The self-citation is therefore non-load-bearing and, per the rubric, contributes only a minimal score of 2 rather than actual circularity.

full rationale

The paper's central derivation chain is: (1) Orb-D3 trajectories of the KcsA selectivity filter spontaneously form a T75-hydroxyl–water hydrogen bond at the S4 entrance; (2) this H-bond is associated with a soft knock-on conduction cycle with water co-transport; (3) the measured cycle time under the applied 0.1 eV/Å force yields 80 ± 20 pS, inside the 40–250 pS experimental range; (4) the prior T75C mutation experiment (Zhou & MacKinnon, Ref [8]) independently shows that removing the T75 hydroxyl sharply reduces K+ conductance. Step (1) is an emergent trajectory observation, not an input constraint: the runs are initialized with K+ at S4/S2 and water at S1/S3, and the paper's potassium-only initialization produces hard knock-on, so water co-transport is not forced by construction. Step (3) is a computed output, not a fitted parameter: the conductance follows from the transit time divided by the applied voltage, and the paper does not tune the force or cycle time to match the experimental range. Step (4) is a pre-existing external experimental result cited as confirmation; it involves no overlap with the present author, so it is independent evidence. The only self-citations appear in Section 3.5 (Refs [52], [53], both including T. T. Duignan), where Orb-D3's K+–water RDFs are compared with the author's own DC-R2SCAN-NNP and DFT simulations; these are supporting model checks, not the load-bearing mechanism claim, and the DFT comparison provides an independent first-principles anchor. The paper's own acknowledged limitations—the 5 V (0.1 eV/Å) drive being 25–50 times physiological voltage, the frozen boundary truncation, the exclusion of the one hard knock-on run from the conductance average, and the absence of enhanced-sampling free-energy calculations—are external-validity and robustness concerns, not circularity: they do not make any reported quantity equivalent to the simulation inputs. Overall, the derivation is self-contained and the score reflects only the minor non-load-bearing self-citation.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the unvalidated accuracy of Orb-D3 for this protein, the equivalence of a 5 V forced drive to physiological conditions, and the choice of a truncated frozen-boundary system. The free parameters are hand-set simulation choices, not fitted to experimental data, but the force and membrane thickness directly determine the reported conductance.

free parameters (3)
  • Applied force on K+ ions = 0.1 eV/Angstrom (about 5 V transmembrane)
    Chosen to accelerate transport to observable timescales. The conductance estimate is computed from the resulting cycle time divided by this voltage, so the reported 80 plus or minus 20 pS value depends directly on this hand-set parameter.
  • Membrane thickness for conductance conversion = 50 Angstrom
    Used to convert the 0.1 eV/Angstrom force into a transmembrane voltage of 5 V and to scale the cycle time to conductance. A different assumed thickness changes the reported pS value.
  • Langevin friction coefficient = 0.01
    Thermostat parameter chosen for the simulation. It affects the dynamics and could influence hopping rates, though it is a standard choice for this kind of simulation.
assumptions (5)
  • domain assumption Orb-D3-v2 provides a sufficiently accurate potential energy surface for the KcsA selectivity filter, including hydrogen bonding and carbonyl flips.
    The model is trained on crystal DFT and is not equivariant or conservative. Bulk KCl RDFs are checked in Section 3.5, but the protein SF itself is not validated against quantum chemistry or experiment.
  • domain assumption The truncated 20 Angstrom radius cylinder with frozen outer atoms preserves the relevant dynamics of the selectivity filter.
    The paper notes gradual dehydration of the SF entrance and gap formation due to restraints in Sections 3.2 and 4, so boundary effects are acknowledged but assumed not to change the transport mechanism qualitatively.
  • domain assumption The 5 V applied force does not qualitatively change the conduction mechanism relative to physiological voltages.
    The author states in Section 3.2 that the force corresponds to a transmembrane voltage of 5 V, much higher than physiological, and in Section 4 that the force may induce distortions. The central claim assumes this is not the case.
  • domain assumption Initial structures taken from Ref [19] represent a valid starting state for the SF simulations.
    The simulations start from coordinates of an earlier classical MD paper, and no relaxation or validation of these coordinates with Orb-D3 is described before applying the driving force.
  • domain assumption PBE-D3 level of DFT used to train Orb-D3 is adequate for the relevant hydrogen-bond energetics in the filter.
    The paper notes in Section 3.5 that there are quantitative discrepancies in bulk electrolyte RDFs likely due to the PBE-D3 level of theory, yet the SF mechanism is interpreted without correction for this potential error.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A potassium ion channel simulated with a universal neural network potential." pith.science (2026). https://pith.science/paper/VFGNN4AA

@misc{pith2026241118931,
  author       = {Pith},
  title        = {Pith review of: A potassium ion channel simulated with a universal neural network potential},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VFGNN4AA}},
  note         = {Machine review of arXiv:2411.18931}
}
abstract

Potassium ion channels are critical components of biology. They conduct potassium ions across the cell membrane with remarkable speed and selectivity. Understanding how they do this is crucially important for applications in neuroscience, medicine, and materials science. However, many fundamental questions about the mechanism they use remain unresolved, partly because it is extremely difficult to computationally model due to the scale and complexity of the necessary simulations. Here, the selectivity filter (SF) of the KcsA potassium ion channel is simulated using Orb-D3, a recently released universal neural network potential. A previously unreported hydrogen bond between water in the SF and the T75 hydroxyl side group at the entrance to the SF is observed. This hydrogen bond appears to stabilize water in the SF, enabling a soft knock-on transport mechanism where water is co-transported through the SF with a reasonable conductivity (80 $\pm$ 20 pS). Carbonyl backbone flipping is also observed at new sites in the SF. This work demonstrates the potential of universal neural network potentials to provide insights into previously intractable questions about complex systems far outside their training data distribution.

Figures

Figures reproduced from arXiv: 2411.18931 by the authors.

Figure 1
Figure 1. The KcsA potassium ion channel (PDB ID: 1K4C) and its selectivity filter [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Cross section of the full cylindrical system simulated. The SF and its [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Snapshots from a simulation trajectory showing rapid water co-transport [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Positions of potassium ions and water molecules through the SF as a [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Radial distribution function (RDF) of the O-H distance between the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Example snapshots from the simulations. (a) The T75 hydroxyl group [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Comparison of RDFs computed with Orb-D3 compared with a custom [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 49 canonical work pages

  1. [8]

    M. Zhou, R. MacKinnon, Journal of Molecular Biology2004, 338, 839–846

    work page

  2. [1]

    D. A. Doyle, J. M. Cabral, R. A. Pfuetzner, A. Kuo, J. M. Gulbis, S. L. Cohen, B. T. Chait, R. MacKinnon, Science 1998, 280, 69–77

    work page 1998

  3. [2]

    Y. Zhou, J. H. Morais-Cabral, A. Kaufman, R. MacKinnon, Nature 2001, 414, 43–48

    work page 2001

  4. [3]

    Roux, Essays Biochem

    B. Roux, Essays Biochem. 2017, 61, (Eds.: S. J. Lippard, J. M. Berg), 201– 209

    work page 2017

  5. [4]

    W. Shi, J. Shen, L. Shen, W. Hu, P. Xu, J. A. Baucom, S. Ma, S. Yang, X.-M. Chen, Y. Lu, Nano Lett.2020, 20, 5435–5442

    work page 2020

  6. [5]

    Miller, Biophysical Journal 1982, 38, 227–230

    C. Miller, Biophysical Journal 1982, 38, 227–230

    work page 1982

  7. [6]

    Alcayaga, X

    C. Alcayaga, X. Cecchi, O. Alvarez, R. Latorre, Biophysical Journal 1989, 55, 367–371

    work page 1989

  8. [7]

    ˚Aqvist, V

    J. ˚Aqvist, V. Luzhkov, Nature 2000, 404, 881–884

    work page 2000

Show all 56 references
  1. [9]

    H. Ando, M. Kuno, H. Shimizu, I. Muramatsu, S. Oiki, J. Gen. Physiol.2005, 126, 529–538

    2005

  2. [10]

    Iwamoto, S

    M. Iwamoto, S. Oiki, J. Neurosci. 2011, 31, 12180–12188

    2011

  3. [11]

    Mironenko, U

    A. Mironenko, U. Zachariae, B. L. De Groot, W. Kopec, Journal of Molecular Biology 2021, 433, 167002

    2021

  4. [12]

    H. T. Kratochvil, J. K. Carr, K. Matulef, A. W. Annen, H. Li, M. Maj, J. Ostmeyer, A. L. Serrano, H. Raghuraman, S. D. Moran, J. L. Skinner, E. Perozo, B. Roux, F. I. Valiyaveetil, M. T. Zanni, Science 2016, 353, 1040– 1044

    2016

  5. [13]

    M. J. Ryan, L. Gao, F. I. Valiyaveetil, M. T. Zanni, A. A. Kananenka, J. Am. Chem. Soc.2023, 145, 18529–18537

    2023

  6. [14]

    P. S. Langan, V. G. Vandavasi, K. L. Weiss, P. V. Afonine, K. El Omari, R. Duman, A. Wagner, L. Coates, Nat Commun 2018, 9, 4540

    2018

  7. [15]

    Furini, C

    S. Furini, C. Domene, Proc. Natl. Acad. Sci. U.S.A.2009, 106, 16074–16077

    2009

  8. [16]

    D. A. K¨ opfer, C. Song, T. Gruene, G. M. Sheldrick, U. Zachariae, B. L. De Groot, Science 2014, 346, 352–355

    2014

  9. [17]

    Kopec, D

    W. Kopec, D. A. K¨ opfer, O. N. Vickery, A. S. Bondarenko, T. L. C. Jansen, B. L. De Groot, U. Zachariae, Nature Chem 2018, 10, 813–820

    2018

  10. [18]

    P. W. Fowler, E. Abad, O. Beckstein, M. S. P. Sansom, J. Chem. Theory Comput. 2013, 9, 5176–5189

    2013

  11. [19]

    Z. Jing, J. A. Rackers, L. R. Pratt, C. Liu, S. B. Rempe, P. Ren, Chem. Sci. 2021, 12, 8920–8930

    2021

  12. [20]

    Furini, C

    S. Furini, C. Domene, J. Chem. Theory Comput.2020, 16, 7148–7159. 13

    2020

  13. [21]

    Treptow, Y

    W. Treptow, Y. Liu, C. A. Z. Bassetto, B. I. Pinto, J. Alves, C. Chipot, F. Bezanilla, B. Roux, Isoleucine Gate Blocks K+ Conduction in C-type Inac- tivation, version 1, 2024, https://arxiv.org/abs/2403.16179 (visited on 09/25/2024), pre-published

    2024 arXiv

  14. [22]

    Behler, M

    J. Behler, M. Parrinello, Phys. Rev. Lett.2007, 98, 1–4

    2007

  15. [23]

    Behler, Chem

    J. Behler, Chem. Rev.2021, 121, 10037–10072

    2021

  16. [24]

    Kocer, T

    E. Kocer, T. W. Ko, J. Behler, Annu. Rev. Phys. Chem.2022, 73, 163–186

    2022

  17. [25]

    Batzner, A

    S. Batzner, A. Musaelian, B. Kozinsky, Nat Rev Phys2023, 5, 437–438

  18. [26]

    Omranpour, P

    A. Omranpour, P. M. De Hijes, J. Behler, C. Dellago, Perspective: Atomistic Simulations of Water and Aqueous Systems with Machine Learning Potentials, 2024, https://arxiv.org/abs/2401.17875 (visited on 01/31/2024), pre- published

    2024 arXiv

  19. [27]

    O. T. Unke, M. St¨ ohr, S. Ganscha, T. Unterthiner, H. Maennel, S. Kashu- bin, D. Ahlin, M. Gastegger, L. Medrano Sandonas, J. T. Berryman, A. Tkatchenko, K.-R. M¨ uller,Sci. Adv.2024, 10, eadn4397

    2024

  20. [28]

    C. Chen, S. P. Ong, Nat Comput Sci2022, 2, 718–728

  21. [29]

    Takamoto, C

    S. Takamoto, C. Shinagawa, D. Motoki, K. Nakago, W. Li, I. Kurata, T. Watanabe, Y. Yayama, H. Iriguchi, Y. Asano, T. Onodera, T. Ishii, T. Kudo, H. Ono, R. Sawada, R. Ishitani, M. Ong, T. Yamaguchi, T. Kataoka, A. Hayashi, N. Charoenphakdee, T. Ibuka, Nat Commun 2022, 13, 2991

    2022

  22. [30]

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, G. Ceder, Nat Mach Intell 2023, 5, 1031–1041

    2023

  23. [31]

    Batatia, P

    I. Batatia, P. Benner, Y. Chiang, A. M. Elena, D. P. Kov´ acs, J. Riebesell, X. R. Advincula, M. Asta, W. J. Baldwin, N. Bernstein, A. Bhowmik, S. M. Blau, V. C˘ arare, J. P. Darby, S. De, F. Della Pia, V. L. Deringer, R. Elijoˇ sius, Z. El- Machachi, E. Fako, A. C. Ferrari, A...

    2023 arXiv

  24. [32]

    D. P. Kov´ acs, J. H. Moore, N. J. Browning, I. Batatia, J. T. Horton, V. Kapil, W. C. Witt, I.-B. Magd˘ au, D. J. Cole, G. Cs´ anyi, MACE-OFF23: Transferable Machine Learning Force Fields for Organic Molecules, 2023, http://arxiv.org/abs/2312.15211 (visited on 01/17/2024), pr...

    2023 arXiv

  25. [33]

    Merchant, S

    A. Merchant, S. Batzner, S. S. Schoenholz, M. Aykol, G. Cheon, E. D. Cubuk, Nature 2023, 624, 80–85. 14

    2023

  26. [34]

    H. Yang, C. Hu, Y. Zhou, X. Liu, Y. Shi, J. Li, G. Li, Z. Chen, S. Chen, C. Zeni, M. Horton, R. Pinsler, A. Fowler, D. Z¨ ugner, T. Xie, J. Smith, L. Sun, Q. Wang, L. Kong, C. Liu, H. Hao, Z. Lu, MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pres...

    2024 arXiv

  27. [35]

    Neumann, J

    M. Neumann, J. Gin, B. Rhodes, S. Bennett, Z. Li, H. Choubisa, A. Hussey, J. Godwin, Orb: A Fast, Scalable Neural Network Potential, 2024, http : //arxiv.org/abs/2410.22570 (visited on 10/31/2024), pre-published

    2024 arXiv

  28. [36]

    A. Jain, S. P. Ong, G. Hautier, W. Chen, W. D. Richards, S. Dacek, S. Cholia, D. Gunter, D. Skinner, G. Ceder, K. A. Persson,APL Mater.2013, 1, 011002

    2013

  29. [37]

    Schmidt, N

    J. Schmidt, N. Hoffmann, H.-C. Wang, P. Borlido, P. J. M. A. Carri¸ co, T. F. T. Cerqueira, S. Botti, M. A. L. Marques,Advanced Materials2023, 35, 2210788

  30. [38]

    S. N. Pozdnyakov, M. Ceriotti, Smooth, Exact Rotational Symmetrization for Deep Learning on Point Clouds, 2023, http://arxiv.org/abs/2305.19302 (visited on 06/05/2023), pre-published

    2023 arXiv

  31. [39]

    M. F. Langer, S. N. Pozdnyakov, M. Ceriotti, Probing the Effects of Broken Symmetries in Machine Learning, 2024, http : / / arxiv . org / abs / 2406 . 17747 (visited on 07/05/2024), pre-published

    2024

  32. [40]

    Hjorth Larsen, J

    A. Hjorth Larsen, J. Jørgen Mortensen, J. Blomqvist, I. E. Castelli, R. Chris- tensen, M. Du lak, J. Friis, M. N. Groves, B. Hammer, C. Hargus, E. D. Her- mes, P. C. Jennings, P. Bjerre Jensen, J. Kermode, J. R. Kitchin, E. Leonhard Kolsbjerg, J. Kubal, K. Kaasbjerg, S. Lysgaa...

  33. [41]

    Vanden-Eijnden, G

    E. Vanden-Eijnden, G. Ciccotti, Chemical Physics Letters 2006, 429, 310– 316

    2006

  34. [42]

    Spivak, J

    M. Spivak, J. E. Stone, J. Ribeiro, J. Saam, P. L. Freddolino, R. C. Bernardi, E. Tajkhorshid, J. Chem. Inf. Model.2023, 63, 4664–4678

    2023

  35. [43]

    C. Hui, R. De Vries, W. Kopec, B. De Groot, Charge Scaling in Potassium Channel Simulations: Conductance, Ion Occupancy, Voltage Response, and Selectivity, 2024, http://biorxiv.org/lookup/doi/10.1101/2024.10. 07.616972 (visited on 10/08/2024), pre-published

    2024 doi

  36. [44]

    LeMasurier, L

    M. LeMasurier, L. Heginbotham, C. Miller, J. Gen. Physiol.2001, 118, 303– 314

    2001

  37. [45]

    J. H. Morais-Cabral, Y. Zhou, R. MacKinnon, Nature 2001, 414, 37–42

    2001

  38. [46]

    Boiteux, D

    C. Boiteux, D. J. Posson, T. W. Allen, C. M. Nimigean, Proc. Natl. Acad. Sci. U.S.A. 2020, 117, 29968–29978

    2020

  39. [47]

    Bern` eche, B

    S. Bern` eche, B. Roux, Structure 2005, 13, 591–600

    2005

  40. [48]

    S. M. Saparov, P. Pohl, Proc. Natl. Acad. Sci. U.S.A.2004, 101, 4805–4809

    2004

  41. [49]

    Hoomann, N

    T. Hoomann, N. Jahnke, A. Horner, S. Keller, P. Pohl, Proc. Natl. Acad. Sci. U.S.A. 2013, 110, 10842–10847. 15

    2013

  42. [50]

    K. Mita, T. Sumikama, M. Iwamoto, Y. Matsuki, K. Shigemi, S. Oiki, Proc. Natl. Acad. Sci. U.S.A.2021, 118, e2017168118

    2021

  43. [51]

    R. P. Swenson, C. M. Armstrong, Nature 1981, 291, 427–429

    1981

  44. [52]

    Zhang, J

    J. Zhang, J. Pagotto, T. Gould, T. T. Duignan, Scalable Molecular Simulation of Electrolyte Solutions with Quantum Chemical Accuracy. 2023, http:// arxiv.org/abs/2310.12535 (visited on 11/09/2023), pre-published

    2023 arXiv

  45. [53]

    T. T. Duignan, G. K. Schenter, J. L. Fulton, T. Huthwelker, M. Balasubrama- nian, M. Galib, M. D. Baer, J. Wilhelm, J. Hutter, M. Del Ben, X. S. Zhao, C. J. Mundy, Phys. Chem. Chem. Phys.2020, 22, 10641–10652

    2020

  46. [54]

    Ball, How Life Works: A User’s Guide to the New Biology, Picador, Lon- don, 2024

    P. Ball, How Life Works: A User’s Guide to the New Biology, Picador, Lon- don, 2024

    2024

  47. [55]

    T. T. Duignan, ACS Phys. Chem Au2024, 4, 232–241

  48. [56]

    G. De Fabritiis, Machine Learning Potentials: A Roadmap Toward Next- Generation Biomolecular Simulations, version 1, 2024, https://arxiv.org/ abs/2408.12625 (visited on 10/28/2024), pre-published. 16

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.