Accuracy and Efficiency Benchmarks of Pretrained Machine Learning Potentials for Molecular Simulations

Abstract

The rapid development of pretrained Machine Learning Interatomic Potentials (MLIPs) that cover a wide range of molecular species has made it challenging to select the best model for a given application. We benchmark 15 pretrained MLIPs, evaluating each one on accuracy, speed, memory use, and ability to produce stable simulations. This provides an objective basis for practitioners to select the most appropriate MLIP for their own simulations and offers insight into which factors most strongly influence model accuracy. We find that the number of model parameters and the size of the training set are both strongly correlated with accuracy, but observe no benefit from including explicit Coulomb energy terms. Speed and memory use are determined as much by the model architecture as by the size of the model.

Modelling Energy Transfer Efficiency in Collision Induced Dissociation and Unfolding Using Molecular Dynamics Simulations
Austin W. Green, James S. Prell
Journal of the American Society for Mass Spectrometry7 Oct 2026Peer-reviewedLicence not statedFull textPhysical chemistryPMC
Finite-temperature dynamic stability separates foundation machine-learning interatomic potentials that harmonic benchmarks rank equally
Frank Cai
ChemRxiv3 Sep 2026PreprintCC BYFull textMaterials & polymersChemRxiv
LM2KM: feature extraction from pretrained protein and molecular language models combined with quantum mechanics (QM) features for KM predictions
Jingchen Zhai, Lianjin Cai, Taoyu Niu +2
Digital discovery20 Oct 2026Peer-reviewedCC BYFull textGeneral chemistryEurope PMCEurope PMCPMC
Toward Chemical Accuracy in Biomolecular Simulations through Data-Driven Many-Body Potentials: III. A Library of Amino Acid Building Blocks
Xinmao Lyu, Ethan F. Bull-Vulpe, Jiwon Huh +1
ChemRxiv6 Nov 2026PreprintLicence not statedGeneral chemistryChemRxiv
Modified Ewald Summation Method for Slab Geometries: Application to Donnan Potentials, Confinement Electrostatics, and Titration Simulations.
Yan Levin, Amin Bakhshandeh
The journal of physical chemistry. B1 Oct 2026Peer-reviewedCC BYFull textPhysical chemistryEurope PMC

Cite

BibTeX RIS

BibTeX

@article{eastman2026accuracy,
  title = {{Accuracy and Efficiency Benchmarks of Pretrained Machine Learning Potentials for Molecular Simulations}},
  author = {Peter K. Eastman and Evan Pretti and Thomas E. Markland},
  journal = {Journal of Chemical Theory and Computation},
  year = {2026},
  month = jun,
  doi = {10.1021/acs.jctc.6c00130},
  url = {https://doi.org/10.1021/acs.jctc.6c00130},
  note = {CC BY-NC-ND},
}

RIS

TY  - JOUR
TI  - Accuracy and Efficiency Benchmarks of Pretrained Machine Learning Potentials for Molecular Simulations
AU  - Peter K. Eastman
AU  - Evan Pretti
AU  - Thomas E. Markland
PY  - 2026
DA  - 2026/06/05
JO  - Journal of Chemical Theory and Computation
DO  - 10.1021/acs.jctc.6c00130
UR  - https://doi.org/10.1021/acs.jctc.6c00130
AB  - The rapid development of pretrained Machine Learning Interatomic Potentials (MLIPs) that cover a wide range of molecular species has made it challenging to select the best model for a given application. We benchmark 15 pretrained MLIPs, evaluating each one on accuracy, speed, memory use, and ability to produce stable simulations. This provides an objective basis for practitioners to select the most appropriate MLIP for their own simulations and offers insight into which factors most strongly influence model accuracy. We find that the number of model parameters and the size of the training set are both strongly correlated with accuracy, but observe no benefit from including explicit Coulomb energy terms. Speed and memory use are determined as much by the model architecture as by the size of the model.
SN  - 1549-9618
SN  - 1549-9626
ER  -

Abstract

Related

Cite