ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

Kexin Wu; Luonan Chen; Renxiao Wang

arxiv: 2606.17668 · v1 · pith:HYRWLJOXnew · submitted 2026-06-16 · 💻 cs.LG · cs.AI· q-bio.QM

ASTEROID: A Spatiotemporal Information Transformer for Forecasting Multi-Step Time Series of Molecular Dynamics

Kexin Wu , Luonan Chen , Renxiao Wang This is my paper

Pith reviewed 2026-06-27 01:27 UTC · model grok-4.3

classification 💻 cs.LG cs.AIq-bio.QM

keywords molecular dynamicstransformerspatiotemporal sequencesmulti-step forecastingatomic coordinatesdata-driven simulationquantum mechanics datasets

0 comments

The pith

A spatiotemporal Transformer directly predicts multi-step atomic coordinates in molecular dynamics simulations.

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

The paper develops a model called ASTEROID that forecasts the outcomes of molecular dynamics simulations over multiple time steps without performing conventional iterative integration. It treats MD trajectories as high-dimensional spatiotemporal sequences and incorporates a Spatiotemporal Information Transformation equation into a Transformer that separately handles short- and long-range spatial interactions plus global and autoregressive temporal patterns. The goal is to deliver accurate long-horizon predictions on quantum-mechanics-derived datasets while lowering the computational expense of standard MD. A sympathetic reader would care because full MD runs become prohibitively expensive for large molecules over long times, so a reliable shortcut would open new studies that are currently impractical.

Core claim

ASTEROID reformulates MD trajectories as high-dimensional spatiotemporal sequences and integrates the Spatiotemporal Information Transformation equation into a Transformer architecture whose local-global self-attention captures both short- and long-range spatial interactions while its encoder-decoder structure combines global context with autoregressive forecasting, yielding higher multi-step accuracy than existing methods on quantum-mechanics-derived molecular datasets together with substantially reduced computational cost.

What carries the argument

The Spatiotemporal Information Transformation equation embedded in a Transformer that uses local-global self-attention for spatial dependencies and an encoder-decoder structure for temporal dependencies.

If this is right

Multi-step atomic coordinates can be obtained directly instead of through repeated integration steps.
Accuracy exceeds that of prior methods across the tested quantum-mechanics-derived benchmarks.
Overall computational cost drops compared with conventional MD simulation.
Iterative application of the model supports forecasting over extended time scales.

Where Pith is reading between the lines

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

If the learned patterns prove general, the same architecture could be applied to molecular systems never seen in training.
Chaining predictions for very long horizons may require additional safeguards against drift that the paper does not test.
The spatiotemporal modeling approach could transfer to other domains that combine spatial structure with time evolution.

Load-bearing premise

Patterns learned from the training trajectories will generalize to unseen molecular systems or much longer time horizons without accumulating unphysical errors.

What would settle it

Apply the trained model to a molecular system or simulation length absent from the training set and check whether the predicted atomic coordinates stay consistent with physical constraints over dozens of steps.

read the original abstract

Molecular dynamics (MD) simulation is computationally demanding, particularly for large-scale systems requiring long-term analysis. Accurate forecast of the outcomes of a MD simulation is not only an attractive scientific challenge but also has substantial practical value. In this work, we developed a data-driven framework, termed ASTEROID (Advanced Spatiotemporal TransformER fOr Inferring Dynamics), that can directly predict multi-step atomic coordinates, avoiding conventional iterative integration. For this purpose, our ASTEROID reformulates MD trajectories as high-dimensional spatiotemporal sequences and integrates the Spatiotemporal Information (STI) Transformation equation into a Transformer architecture. The core innovation of ASTEROID lies in its ability to model multiscale spatiotemporal dependencies. In particular, for spatial dependencies, a local-global self-attention mechanism captures both short- and long-range interactions. For temporal dependencies, an encoder-decoder structure integrates global context with autoregressive forecasting. ASTEROID was evaluated on several quantum-mechanics derived molecular datasets. Our results indicate that ASTEROID achieved not only a higher level of accuracy in multi-step prediction than existing methods on various benchmarks, but also significantly reduced computational cost of conventional MD simulation. Moreover, the model supports iterative multi-step forecasting over an extended time scale. This work establishes a robust and generalizable data-driven paradigm for accelerating MD simulations.

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.

Desk Editor's Note private letter to a colleague

ASTEROID is a straightforward spatiotemporal Transformer for direct multi-step MD coordinate prediction that targets a real pain point in simulation cost, but the abstract supplies no numbers, ablations, or dataset details to judge whether the accuracy and speedup claims hold.

read the letter

The core of this paper is a Transformer that takes MD trajectories, treats them as spatiotemporal sequences, and uses local-global attention plus an encoder-decoder to forecast several steps of atomic positions at once instead of stepping through the integrator. That reformulation is the main technical move, and it lines up with the goal of cutting the expense of long runs.

What the work does cleanly is identify a practical target: conventional MD is slow for large systems or long times, and learning to jump ahead on seen patterns could help if the errors stay manageable. The architecture choices (local-global attention for space, autoregressive decoder for time) are reasonable extensions of existing time-series Transformers and match the multiscale nature of molecular motion.

The soft spots are all in the evidence. The abstract states higher accuracy and lower cost on quantum-mechanics datasets but gives no sample sizes, no error bars, no ablation tables, and no direct runtime comparisons that include training overhead. Without those, the central performance claim cannot be checked for selection effects or whether the gains survive on out-of-distribution molecules or longer horizons. The generalization worry the reader flagged is real; nothing in the summary rules out accumulating unphysical drift once the model leaves the training distribution.

This is the kind of paper that belongs in a computational chemistry or ML-for-science reading group if the full experiments turn out to be reproducible. A serious editor should send it to referees so the numbers and baselines can be examined; the idea is not exotic, but the application is concrete enough that solid empirical work would be useful to the subfield.

Referee Report

0 major / 1 minor

Summary. The manuscript introduces ASTEROID, a Transformer-based framework that reformulates MD trajectories as high-dimensional spatiotemporal sequences and integrates the Spatiotemporal Information (STI) Transformation equation. It employs a local-global self-attention mechanism to capture short- and long-range spatial interactions and an encoder-decoder structure for temporal dependencies with autoregressive forecasting. The model is evaluated on quantum-mechanics derived molecular datasets and claims higher accuracy in multi-step atomic coordinate prediction than existing methods, significantly reduced computational cost relative to conventional MD, and support for iterative forecasting over extended timescales.

Significance. If the empirical benchmark results hold with appropriate controls, this could offer a practical data-driven route to accelerate MD by replacing iterative integration with direct multi-step prediction, enabling longer timescale analysis at lower cost. The architectural choices for multiscale spatiotemporal modeling are internally consistent with the goal. The stress-test concern about OOD generalization and unphysical error accumulation does not land as a load-bearing issue for the reported in-distribution results, as the central claim rests on benchmark performance rather than requiring bounded long-horizon extrapolation.

minor comments (1)

The abstract states performance claims without referencing specific quantitative metrics, baseline methods, dataset sizes, or error bars; moving these details to the abstract or adding a results summary table would improve immediate readability.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the recommendation for minor revision. We are encouraged that the potential for data-driven acceleration of MD simulations is recognized, along with the internal consistency of the architectural choices.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The provided abstract and description outline a standard data-driven Transformer architecture (STI-Transformer with local-global attention and encoder-decoder) trained to forecast multi-step atomic coordinates from MD trajectories. No equations are shown, no parameters are described as fitted then renamed as predictions, and no self-citation chain or uniqueness theorem is invoked to justify the core method. The central claims rest on empirical benchmark comparisons rather than any derivation that reduces outputs to inputs by construction. The logical structure is self-contained against external benchmarks with no load-bearing self-referential steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; all such elements are unknown.

pith-pipeline@v0.9.1-grok · 5776 in / 1032 out tokens · 25502 ms · 2026-06-27T01:27:17.021032+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

60 extracted references · 7 canonical work pages · 1 internal anchor

[1]

Salo -Ahen, O. M. H. et al. Molecular Dynamics Simulations in Drug Discovery and Pharmaceutical Development. Processes 9, 71 (2020)

2020
[2]

Liu, X. et al. Molecular dynamics simulations and novel drug discovery. Expert Opinion on Drug Discovery 13, 23–37 (2018)

2018
[3]

& Singh, D

Singh, S., Bani Baker, Q. & Singh, D. B. Chapter 18 - Molecular docking and molecular dynamics simulation. in Bioinformatics (eds. Singh, D. B. & Pathak, R. K.) 291 –304 (Academic Press, 2022). doi:10.1016/B978-0-323-89775-4.00014-6

work page doi:10.1016/b978-0-323-89775-4.00014-6 2022
[4]

van Gunsteren, W. F. & Oostenbrink, C. Methods for Classical-Mechanical Molecular Simulation in Chemistry: Achievements, Limitations, Perspectives. J. Chem. Inf. Model. 64, 6281–6304 (2024)

2024
[5]

Adelusi, T. I. et al. Molecular modeling in drug discovery. Informatics in Medicine Unlocked 29, 100880 (2022)

2022
[6]

R., Valsson, O

Hé nin, J., Leliè vre, T., Shirts, M. R., Valsson, O. & Delemotte, L. Enhanced sampling methods for molecular dynamics simulations. LiveCoMS 4, (2022)

2022
[7]

Bai, Q. et al. Application advances of deep learning methods for de novo drug design and molecular dynamics simulation. WIREs Comput Mol Sci 12, e1581 (2022)

2022
[8]

Wang, H., Zhang, L., Han, J. & E, W. DeePMD-Kit: A deep learning package for many-body potential energy representation and molecular dynamics. Comput. Phys. Commun. 228, 178–184 (2018)

2018
[9]

Jia, W. et al. Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning. Preprint at arXiv https://arxiv.org/abs/2005.00223 (2020)

arXiv 2005
[10]

Zheng, P. et al. Artificial intelligence-enhanced quantum chemical method with broad applicability. Nat. Commun. 12, 7022 (2021)

2021
[11]

& Voth, G

Li, C. & Voth, G. A. Using machine learning to greatly accelerate path integral ab initio molecular dynamics. J. Chem. Theory Comput. 18, 599–604 (2022)

2022
[12]

Zhang, D. et al. DPA-2: towards a universal large atomic model for molecular and material simulation. Preprint at arXiv https://arxiv.org/abs/2312.15492 (2023)

arXiv 2023
[13]

Wang, Y. et al. Enhancing geometric representations for molecules with equivariant vector- scalar interactive message passing. Nat. Commun. 15, 313 (2024)

2024
[14]

Doerr, S. et al. TorchMD: a deep learning framework for molecular simulations. J. Chem. Theory Comput. 17, 2355–2363 (2021)

2021
[15]

Nassar, R. et al. Accelerating protein folding molecular dynamics using inter-residue distances from machine learning servers. J. Chem. Theory Comput. 18, 1929–1935 (2022)

1929
[16]

& Elber, R

Di Pierro, M. & Elber, R. Automated optimization of potential parameters. J. Chem. Theory Comput. 9, 3311–3320 (2013). 30

2013
[17]

& Chen, H.-F

Ji, X., Liu, H., Zhang, Y., Chen, J. & Chen, H.-F. Personal precise force field for intrinsically disordered and ordered proteins based on deep learning. J. Chem. Inf. Model. 63, 362–374 (2023)

2023
[18]

& Riniker, S

Thü rlemann, M., Bö selt, L. & Riniker, S. Regularized by physics: graph neural network parametrized potentials for the description of intermolecular interactions. J. Chem. Theory Comput. 19, 562–579 (2023)

2023
[19]

Ji, X. et al. Research and evaluation of the allosteric protein-specific force field based on a pre- training deep learning model. J. Chem. Inf. Model. 63, 2456–2468 (2023)

2023
[20]

& Yasuoka, K

Kawada, R., Endo, K., Yuhara, D. & Yasuoka, K. MD -GAN with multi -particle input: the machine learning of long -time molecular behavior from short -time MD data. Preprint at https://doi.org/10.48550/arXiv.2202.00995 (2022)

work page doi:10.48550/arxiv.2202.00995 2022
[21]

Gupta, C. et al. Mind reading of the proteins: Deep -learning to forecast molecular dynamics. Preprint at https://doi.org/10.1101/2020.07.28.225490 (2020)

work page doi:10.1101/2020.07.28.225490 2020
[22]

& Yasuoka, K

Endo, K., Tomobe, K. & Yasuoka, K. Multi-Step Time Series Generator for Molecular Dynamics. The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) (February 2-7, 2018; New Orleans, Louisiana, USA). AAAI Press (2018)

2018
[23]

Wang, D. et al. Efficient sampling of high-dimensional free energy landscapes using adaptive reinforced dynamics. Nat. Comput. Sci. 2, 20–29 (2021)

2021
[24]

Zhang, J. et al. Deep reinforcement learning of transition states. Phys. Chem. Chem. Phys. 23, 6888–6895 (2021)

2021
[25]

Chen, H. et al. MLCV: bridging machine-learning-based dimensionality reduction and free- energy calculation. J. Chem. Inf. Model. 62, 1–8 (2022)

2022
[26]

Wang, Y., Lamim Ribeiro, J. M. & Tiwary, P. Machine learning approaches for analyzing and enhancing molecular dynamics simulations. Curr. Opin. Struct. Biol. 61, 139–145 (2020)

2020
[27]

N., Wang, J., Bhattarai, A

Do, H. N., Wang, J., Bhattarai, A. & Miao, Y. GLOW: a workflow integrating Gaussian- accelerated molecular dynamics and deep learning for free energy profiling. J. Chem. Theory Comput. 18, 1423–1436 (2022)

2022
[28]

& Wolf, S

Bray, S., Tä nzel, V. & Wolf, S. Ligand unbinding pathway and mechanism analysis assisted by machine learning and graph methods. J. Chem. Inf. Model. 62, 4591–4604 (2022)

2022
[29]

Aranganathan, A., Gu, X., Wang, D., Vani, B. P. & Tiwary, P. Modeling Boltzmann -weighted structural ensembles of proteins using artificial intelligence –based methods. Current Opinion in Structural Biology 91, 103000 (2025)

2025
[30]

& Tang, J

Lu, J., Zhong, B., Zhang, Z. & Tang, J. Str2Str: a score -based framework for zero -shot protein conformation sampling. Preprint at arXiv https://doi.org/10.48550/arXiv.2306.03117 (2024)

work page doi:10.48550/arxiv.2306.03117 2024
[31]

& Jaakkola, T

Jing, B., Berger, B. & Jaakkola, T. AlphaFold meets flow matching for generating protein ensembles. In International Conference on Machine Learning. 896, 22277 - 22303 (2024)

2024
[32]

& Berger, B

Jing, B., Stä rk, H., Jaakkola, T. & Berger, B. Generative modeling of molecular dynamics 31 trajectories. In 38th Conference on Neural Information Processing Systems (2024)

2024
[33]

Klein, L. et al. Timewarp: Transferable Acceleration of Molecular Dynamics by Learning Time- Coarsened Dynamics. in Advances in Neural Information Processing Systems (eds. Oh, A. et al.) 36 52863–52883 (2023)

2023
[34]

A Critical Review of Recurrent Neural Networks for Sequence Learning

Lipton, Z. C., Berkowitz, J. & Elkan, C. A Critical Review of Recurrent Neural Networks for Sequence Learning. Preprint at arXiv https://doi.org/10.48550/arXiv.1506.00019 (2015)

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1506.00019 2015
[35]

Hochreiter & J

S. Hochreiter & J. Schmidhuber. Long Short -Term Memory. Neural Computation 9, 1735–1780 (1997)

1997
[36]

D., Swart, T

Mienye, I. D., Swart, T. G. & Obaido, G. Recurrent Neural Networks: A Comprehensive Review of Architectures, Variants, and Applications. Information 15, 517 (2024)

2024
[37]

Vaswani, A. et al. Attention is all you need. In Advances in Neural Information Processing Systems 30 6000–6010 (2017)

2017
[38]

Wen, Q. et al . Transformers in time series: a survey. Preprint at arXiv https://arxiv.org/abs/2202.07125 (2023)

arXiv 2023
[39]

& Chen, L

Ma, H., Zhou, T., Aihara, K. & Chen, L. Predicting time series from short-term high-dimensional data. Int. J. Bifurc. Chaos 24, 1430033 (2014)

2014
[40]

& Chen, L

Ma, H., Leng, S., Aihara, K., Lin, W. & Chen, L. Randomly distributed embedding making short- term high-dimensional data predictable. Proc. Natl Acad. Sci. USA 115, E9994–E10002 (2018)

2018
[41]

Dynamical Systems And Turbulence, Warwick 1980 366–381 (Springer, Berlin, 1981)

Takens, F. Dynamical Systems And Turbulence, Warwick 1980 366–381 (Springer, Berlin, 1981)

1980
[42]

Sauer, T., Yorke, J. A. & Casdagli, M. Embedology. J. Stat. Phys. 65, 579–616 (1991)

1991
[43]

Calvo, F.; Galindez, J.; Gadé a, F. X. Sampling the Configuration Space of Finite Atomic Systems: How Ergodic Is Molecular Dynamics? J. Phys. Chem. A 106, 4145 –4152. https://doi.org/10.1021/jp013691+ (2002)

work page doi:10.1021/jp013691 2002
[44]

& Chen, L

Chen, P., Liu, R., Aihara, K. & Chen, L. Autoreservoir computing for multistep ahead prediction based on the spatiotemporal information transformation. Nat. Commun. 11, 4568 (2020)

2020
[45]

& Chen, L

Peng, H., Chen, P., Liu, R. & Chen, L. Spatiotemporal information conversion machine for time- series forecasting. Fundam. Res. 2, S2667325822004538 (2022)

2022
[46]

& Chen, L

Tao, P., Hao, X., Cheng, J. & Chen, L. Predicting time series by data -driven spatiotemporal information transformation. Inf. Sci. 622, 859–872 (2023)

2023
[47]

Chmiela, S. et al. Accurate global machine learning force fields for molecules with hundreds of atoms. Sci. Adv. 9, eadf0873 (2023)

2023
[48]

Short molecular dynamics of a peptide inside a pure DMPC membrane [Dataset]

Cruz, D. Short molecular dynamics of a peptide inside a pure DMPC membrane [Dataset]. figshare https://doi.org/10.6084/m9.figshare.8046437.v1 (2019)

work page doi:10.6084/m9.figshare.8046437.v1 2019
[49]

R. W. Schafer. What Is a Savitzky -Golay Filter? IEEE Signal Processing Magazine 28, 111–117 (2011)

2011
[50]

& Bengio, Y

Glorot, X. & Bengio, Y. Understanding the difficulty of training deep feedforward neural 32 networks. In International Conference on Artificial Intelligence and Statistics (2010)

2010
[51]

Jiang, J., Han, C., Zhao, W. X. & Wang, J. PDFormer : propagation delay-aware dynamic long- range transformer for traffic flow prediction. In Proceedings of the AAAI Conference on Artificial Intelligence 37, 4365-4373 (AAAI Press, 2023)

2023
[52]

Huo, G. et al . Hierarchical spatio -temporal graph convolutional networks and transformer network for traffic flow forecasting. IEEE Trans. Intell. Transp. Syst. 24, 3855–3867 (2023)

2023
[53]

Liu, Y. et al . iTransformer: inverted transformers are effective for time series forecasting. In International Conference on Learning Representations (2024)

2024
[54]

& Rosenberg, A

Myers, C., Rabiner, L. & Rosenberg, A. Performance tradeoffs in dynamic time warping algorithms for isolated word recognition. IEEE Trans. Acoust. Speech Signal Process. 28, 623–635 (1980)

1980
[55]

Smola, A. J. & Schö lkopf, B. A tutorial on support vector regression. Stat. Comput. 14, 199-222 (2004)

2004
[56]

Elman, J. L. Finding structure in time. Cogn. Sci. 14, 179–211 (1990)

1990
[57]

Holt, C. C. Forecasting seasonals and trends by exponentially weighted moving averages. Int. J. Forecast. 20, 5–10 (2004)

2004
[58]

& Chen, L

You, Y., Zhang, L., Tao, P., Liu, S. & Chen, L. Spatiotemporal transformer neural network for time-series forecasting. Entropy 24, 1651 (2022)

2022
[59]

Zeng, A., Chen, M., Zhang, L. & Xu, Q. Are transformers effective for time series forecasting? In Proceedings of the 37th AAAI Conference on Artificial Intelligence 37, 11121–11128 (2023)

2023
[60]

H., Sinthong, P

Nie, Y., Nguyen, N. H., Sinthong, P. & Kalagnanam, J. A time series is worth 64 words: long - term forecasting with transformers. In International Conference on Learning Representations (2023)

2023

[1] [1]

Salo -Ahen, O. M. H. et al. Molecular Dynamics Simulations in Drug Discovery and Pharmaceutical Development. Processes 9, 71 (2020)

2020

[2] [2]

Liu, X. et al. Molecular dynamics simulations and novel drug discovery. Expert Opinion on Drug Discovery 13, 23–37 (2018)

2018

[3] [3]

& Singh, D

Singh, S., Bani Baker, Q. & Singh, D. B. Chapter 18 - Molecular docking and molecular dynamics simulation. in Bioinformatics (eds. Singh, D. B. & Pathak, R. K.) 291 –304 (Academic Press, 2022). doi:10.1016/B978-0-323-89775-4.00014-6

work page doi:10.1016/b978-0-323-89775-4.00014-6 2022

[4] [4]

van Gunsteren, W. F. & Oostenbrink, C. Methods for Classical-Mechanical Molecular Simulation in Chemistry: Achievements, Limitations, Perspectives. J. Chem. Inf. Model. 64, 6281–6304 (2024)

2024

[5] [5]

Adelusi, T. I. et al. Molecular modeling in drug discovery. Informatics in Medicine Unlocked 29, 100880 (2022)

2022

[6] [6]

R., Valsson, O

Hé nin, J., Leliè vre, T., Shirts, M. R., Valsson, O. & Delemotte, L. Enhanced sampling methods for molecular dynamics simulations. LiveCoMS 4, (2022)

2022

[7] [7]

Bai, Q. et al. Application advances of deep learning methods for de novo drug design and molecular dynamics simulation. WIREs Comput Mol Sci 12, e1581 (2022)

2022

[8] [8]

Wang, H., Zhang, L., Han, J. & E, W. DeePMD-Kit: A deep learning package for many-body potential energy representation and molecular dynamics. Comput. Phys. Commun. 228, 178–184 (2018)

2018

[9] [9]

Jia, W. et al. Pushing the limit of molecular dynamics with ab initio accuracy to 100 million atoms with machine learning. Preprint at arXiv https://arxiv.org/abs/2005.00223 (2020)

arXiv 2005

[10] [10]

Zheng, P. et al. Artificial intelligence-enhanced quantum chemical method with broad applicability. Nat. Commun. 12, 7022 (2021)

2021

[11] [11]

& Voth, G

Li, C. & Voth, G. A. Using machine learning to greatly accelerate path integral ab initio molecular dynamics. J. Chem. Theory Comput. 18, 599–604 (2022)

2022

[12] [12]

Zhang, D. et al. DPA-2: towards a universal large atomic model for molecular and material simulation. Preprint at arXiv https://arxiv.org/abs/2312.15492 (2023)

arXiv 2023

[13] [13]

Wang, Y. et al. Enhancing geometric representations for molecules with equivariant vector- scalar interactive message passing. Nat. Commun. 15, 313 (2024)

2024

[14] [14]

Doerr, S. et al. TorchMD: a deep learning framework for molecular simulations. J. Chem. Theory Comput. 17, 2355–2363 (2021)

2021

[15] [15]

Nassar, R. et al. Accelerating protein folding molecular dynamics using inter-residue distances from machine learning servers. J. Chem. Theory Comput. 18, 1929–1935 (2022)

1929

[16] [16]

& Elber, R

Di Pierro, M. & Elber, R. Automated optimization of potential parameters. J. Chem. Theory Comput. 9, 3311–3320 (2013). 30

2013

[17] [17]

& Chen, H.-F

Ji, X., Liu, H., Zhang, Y., Chen, J. & Chen, H.-F. Personal precise force field for intrinsically disordered and ordered proteins based on deep learning. J. Chem. Inf. Model. 63, 362–374 (2023)

2023

[18] [18]

& Riniker, S

Thü rlemann, M., Bö selt, L. & Riniker, S. Regularized by physics: graph neural network parametrized potentials for the description of intermolecular interactions. J. Chem. Theory Comput. 19, 562–579 (2023)

2023

[19] [19]

Ji, X. et al. Research and evaluation of the allosteric protein-specific force field based on a pre- training deep learning model. J. Chem. Inf. Model. 63, 2456–2468 (2023)

2023

[20] [20]

& Yasuoka, K

Kawada, R., Endo, K., Yuhara, D. & Yasuoka, K. MD -GAN with multi -particle input: the machine learning of long -time molecular behavior from short -time MD data. Preprint at https://doi.org/10.48550/arXiv.2202.00995 (2022)

work page doi:10.48550/arxiv.2202.00995 2022

[21] [21]

Gupta, C. et al. Mind reading of the proteins: Deep -learning to forecast molecular dynamics. Preprint at https://doi.org/10.1101/2020.07.28.225490 (2020)

work page doi:10.1101/2020.07.28.225490 2020

[22] [22]

& Yasuoka, K

Endo, K., Tomobe, K. & Yasuoka, K. Multi-Step Time Series Generator for Molecular Dynamics. The Thirty-Second AAAI Conference on Artificial Intelligence (AAAI-18) (February 2-7, 2018; New Orleans, Louisiana, USA). AAAI Press (2018)

2018

[23] [23]

Wang, D. et al. Efficient sampling of high-dimensional free energy landscapes using adaptive reinforced dynamics. Nat. Comput. Sci. 2, 20–29 (2021)

2021

[24] [24]

Zhang, J. et al. Deep reinforcement learning of transition states. Phys. Chem. Chem. Phys. 23, 6888–6895 (2021)

2021

[25] [25]

Chen, H. et al. MLCV: bridging machine-learning-based dimensionality reduction and free- energy calculation. J. Chem. Inf. Model. 62, 1–8 (2022)

2022

[26] [26]

Wang, Y., Lamim Ribeiro, J. M. & Tiwary, P. Machine learning approaches for analyzing and enhancing molecular dynamics simulations. Curr. Opin. Struct. Biol. 61, 139–145 (2020)

2020

[27] [27]

N., Wang, J., Bhattarai, A

Do, H. N., Wang, J., Bhattarai, A. & Miao, Y. GLOW: a workflow integrating Gaussian- accelerated molecular dynamics and deep learning for free energy profiling. J. Chem. Theory Comput. 18, 1423–1436 (2022)

2022

[28] [28]

& Wolf, S

Bray, S., Tä nzel, V. & Wolf, S. Ligand unbinding pathway and mechanism analysis assisted by machine learning and graph methods. J. Chem. Inf. Model. 62, 4591–4604 (2022)

2022

[29] [29]

Aranganathan, A., Gu, X., Wang, D., Vani, B. P. & Tiwary, P. Modeling Boltzmann -weighted structural ensembles of proteins using artificial intelligence –based methods. Current Opinion in Structural Biology 91, 103000 (2025)

2025

[30] [30]

& Tang, J

Lu, J., Zhong, B., Zhang, Z. & Tang, J. Str2Str: a score -based framework for zero -shot protein conformation sampling. Preprint at arXiv https://doi.org/10.48550/arXiv.2306.03117 (2024)

work page doi:10.48550/arxiv.2306.03117 2024

[31] [31]

& Jaakkola, T

Jing, B., Berger, B. & Jaakkola, T. AlphaFold meets flow matching for generating protein ensembles. In International Conference on Machine Learning. 896, 22277 - 22303 (2024)

2024

[32] [32]

& Berger, B

Jing, B., Stä rk, H., Jaakkola, T. & Berger, B. Generative modeling of molecular dynamics 31 trajectories. In 38th Conference on Neural Information Processing Systems (2024)

2024

[33] [33]

Klein, L. et al. Timewarp: Transferable Acceleration of Molecular Dynamics by Learning Time- Coarsened Dynamics. in Advances in Neural Information Processing Systems (eds. Oh, A. et al.) 36 52863–52883 (2023)

2023

[34] [34]

A Critical Review of Recurrent Neural Networks for Sequence Learning

Lipton, Z. C., Berkowitz, J. & Elkan, C. A Critical Review of Recurrent Neural Networks for Sequence Learning. Preprint at arXiv https://doi.org/10.48550/arXiv.1506.00019 (2015)

work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1506.00019 2015

[35] [35]

Hochreiter & J

S. Hochreiter & J. Schmidhuber. Long Short -Term Memory. Neural Computation 9, 1735–1780 (1997)

1997

[36] [36]

D., Swart, T

Mienye, I. D., Swart, T. G. & Obaido, G. Recurrent Neural Networks: A Comprehensive Review of Architectures, Variants, and Applications. Information 15, 517 (2024)

2024

[37] [37]

Vaswani, A. et al. Attention is all you need. In Advances in Neural Information Processing Systems 30 6000–6010 (2017)

2017

[38] [38]

Wen, Q. et al . Transformers in time series: a survey. Preprint at arXiv https://arxiv.org/abs/2202.07125 (2023)

arXiv 2023

[39] [39]

& Chen, L

Ma, H., Zhou, T., Aihara, K. & Chen, L. Predicting time series from short-term high-dimensional data. Int. J. Bifurc. Chaos 24, 1430033 (2014)

2014

[40] [40]

& Chen, L

Ma, H., Leng, S., Aihara, K., Lin, W. & Chen, L. Randomly distributed embedding making short- term high-dimensional data predictable. Proc. Natl Acad. Sci. USA 115, E9994–E10002 (2018)

2018

[41] [41]

Dynamical Systems And Turbulence, Warwick 1980 366–381 (Springer, Berlin, 1981)

Takens, F. Dynamical Systems And Turbulence, Warwick 1980 366–381 (Springer, Berlin, 1981)

1980

[42] [42]

Sauer, T., Yorke, J. A. & Casdagli, M. Embedology. J. Stat. Phys. 65, 579–616 (1991)

1991

[43] [43]

Calvo, F.; Galindez, J.; Gadé a, F. X. Sampling the Configuration Space of Finite Atomic Systems: How Ergodic Is Molecular Dynamics? J. Phys. Chem. A 106, 4145 –4152. https://doi.org/10.1021/jp013691+ (2002)

work page doi:10.1021/jp013691 2002

[44] [44]

& Chen, L

Chen, P., Liu, R., Aihara, K. & Chen, L. Autoreservoir computing for multistep ahead prediction based on the spatiotemporal information transformation. Nat. Commun. 11, 4568 (2020)

2020

[45] [45]

& Chen, L

Peng, H., Chen, P., Liu, R. & Chen, L. Spatiotemporal information conversion machine for time- series forecasting. Fundam. Res. 2, S2667325822004538 (2022)

2022

[46] [46]

& Chen, L

Tao, P., Hao, X., Cheng, J. & Chen, L. Predicting time series by data -driven spatiotemporal information transformation. Inf. Sci. 622, 859–872 (2023)

2023

[47] [47]

Chmiela, S. et al. Accurate global machine learning force fields for molecules with hundreds of atoms. Sci. Adv. 9, eadf0873 (2023)

2023

[48] [48]

Short molecular dynamics of a peptide inside a pure DMPC membrane [Dataset]

Cruz, D. Short molecular dynamics of a peptide inside a pure DMPC membrane [Dataset]. figshare https://doi.org/10.6084/m9.figshare.8046437.v1 (2019)

work page doi:10.6084/m9.figshare.8046437.v1 2019

[49] [49]

R. W. Schafer. What Is a Savitzky -Golay Filter? IEEE Signal Processing Magazine 28, 111–117 (2011)

2011

[50] [50]

& Bengio, Y

Glorot, X. & Bengio, Y. Understanding the difficulty of training deep feedforward neural 32 networks. In International Conference on Artificial Intelligence and Statistics (2010)

2010

[51] [51]

Jiang, J., Han, C., Zhao, W. X. & Wang, J. PDFormer : propagation delay-aware dynamic long- range transformer for traffic flow prediction. In Proceedings of the AAAI Conference on Artificial Intelligence 37, 4365-4373 (AAAI Press, 2023)

2023

[52] [52]

Huo, G. et al . Hierarchical spatio -temporal graph convolutional networks and transformer network for traffic flow forecasting. IEEE Trans. Intell. Transp. Syst. 24, 3855–3867 (2023)

2023

[53] [53]

Liu, Y. et al . iTransformer: inverted transformers are effective for time series forecasting. In International Conference on Learning Representations (2024)

2024

[54] [54]

& Rosenberg, A

Myers, C., Rabiner, L. & Rosenberg, A. Performance tradeoffs in dynamic time warping algorithms for isolated word recognition. IEEE Trans. Acoust. Speech Signal Process. 28, 623–635 (1980)

1980

[55] [55]

Smola, A. J. & Schö lkopf, B. A tutorial on support vector regression. Stat. Comput. 14, 199-222 (2004)

2004

[56] [56]

Elman, J. L. Finding structure in time. Cogn. Sci. 14, 179–211 (1990)

1990

[57] [57]

Holt, C. C. Forecasting seasonals and trends by exponentially weighted moving averages. Int. J. Forecast. 20, 5–10 (2004)

2004

[58] [58]

& Chen, L

You, Y., Zhang, L., Tao, P., Liu, S. & Chen, L. Spatiotemporal transformer neural network for time-series forecasting. Entropy 24, 1651 (2022)

2022

[59] [59]

Zeng, A., Chen, M., Zhang, L. & Xu, Q. Are transformers effective for time series forecasting? In Proceedings of the 37th AAAI Conference on Artificial Intelligence 37, 11121–11128 (2023)

2023

[60] [60]

H., Sinthong, P

Nie, Y., Nguyen, N. H., Sinthong, P. & Kalagnanam, J. A time series is worth 64 words: long - term forecasting with transformers. In International Conference on Learning Representations (2023)

2023