AIMNet2-rxn: A machine-learned potential for generalized reaction modeling on a millions-of-pathways scale
9 affiliations
- 1Department of Chemical Engineering and Materials Science, Michigan State University, East Lansing, MI 48824, USA
- 2Department of Chemistry, Carnegie Mellon University, Pittsburgh, PA 15213, USA
- 3Carnegie Mellon University
- 4Michigan State University
- 5Deep Principle Inc., HangZhou, Zhejiang, China
- 6Department of Chemical Engineering, Purdue University, West Lafayette, IN 47907, USA
- 7Purdue University West Lafayette
- 8Department of Chemical Engineering, University of Notre Dame, Notre Dame, IN 46556, USA
- 9University of Notre Dame
Abstract
Mechanistic modeling of chemical transformations offers a compelling basis for understanding reactivity and allows for prediction of reaction outcomes before attempting experiments. Despite progress in machine-learned interatomic potentials (MLIPs), we demonstrate that available models lack the accuracy for diverse reaction modeling. With this motivation, we developed a general MLIP for mechanistic modeling of closed-shell carbon, hydrogen, nitrogen, and oxygen reactions, AIMNet2-rxn, using a dataset of ∼4.7 × 10 6 range-separated density functional theory calculations. AIMNet2-rxn enables reaction modeling ∼10 6 faster than the reference quantum mechanical (QM) methods while substantially outperforming graph-based ML, reaffirming the value using three-dimensional chemical information for training. On a test suite of well-known reaction mechanisms—such as amide formation, proton transfers, and pericyclics—AIMNet2-rxn yields 1 to 2 kilocalories per mole accuracy across reaction coordinates without retraining or system-specific fine-tuning. To exploit graphics processing unit parallelism and AIMNet2-rxn efficiency, we introduce a batched nudged elastic band procedure that readily achieves minimum energy pathway search on a millions-of-reactions scale. To demonstrate complex reaction characterization, the thermodynamics of an 11-step pathway producing hydroxymethylfurfural, the experimentally observed major product of glucose pyrolysis, is evaluated. Overall, the accuracy and efficiency afforded by AIMNet2-rxn create opportunities in high-throughput reaction discovery and deep reaction network analysis that would be infeasible with QM methods.
Versions
Related
- Excited-state dynamics and photo-reaction pathways of 4-nitrothiophenol in solution investigated by time-resolved soft X-ray absorption spectroscopy
- Efficient Light‐Driven Multi‐Electron Accumulation on Viologens by Molecular Control Over Photochemical Reaction Pathways
- Drift-React: One-step Generation of Reaction Pathways via SE(3) Drifting Fields
- Multiscale Insights into CO2 Chemisorption and Proton Transfer in Choline-Imidazolate based Deep Eutectic solvents: Exploring the Effect of Primary, Secondary and Tertiary Alkanolamine on Reaction Pathways and Product Distribution
- Resolving Photochemical Reaction Pathways Using Field-Induced Surface Hopping Simulation of fs-EUV-TRPES: The Case of Furan
BibTeX
@article{anstine2026aimnet2rxn,
title = {{AIMNet2-rxn: A machine-learned potential for generalized reaction modeling on a millions-of-pathways scale}},
author = {Dylan M. Anstine and Qiyuan Zhao and R.I. Zubatyuk and Shuhao Zhang and Veerupaksh Singla and Filipp Nikitin and Brett Matthew Savoie and Olexandr Isayev},
journal = {Science Advances},
year = {2026},
month = oct,
doi = {10.1126/sciadv.aea1557},
url = {https://www.science.org/doi/10.1126/sciadv.aea1557},
note = {CC BY-NC},
}
RIS
TY - JOUR
TI - AIMNet2-rxn: A machine-learned potential for generalized reaction modeling on a millions-of-pathways scale
AU - Dylan M. Anstine
AU - Qiyuan Zhao
AU - R.I. Zubatyuk
AU - Shuhao Zhang
AU - Veerupaksh Singla
AU - Filipp Nikitin
AU - Brett Matthew Savoie
AU - Olexandr Isayev
PY - 2026
DA - 2026/10/07
JO - Science Advances
DO - 10.1126/sciadv.aea1557
UR - https://www.science.org/doi/10.1126/sciadv.aea1557
AB - Mechanistic modeling of chemical transformations offers a compelling basis for understanding reactivity and allows for prediction of reaction outcomes before attempting experiments. Despite progress in machine-learned interatomic potentials (MLIPs), we demonstrate that available models lack the accuracy for diverse reaction modeling. With this motivation, we developed a general MLIP for mechanistic modeling of closed-shell carbon, hydrogen, nitrogen, and oxygen reactions, AIMNet2-rxn, using a dataset of ∼4.7 × 10 6 range-separated density functional theory calculations. AIMNet2-rxn enables reaction modeling ∼10 6 faster than the reference quantum mechanical (QM) methods while substantially outperforming graph-based ML, reaffirming the value using three-dimensional chemical information for training. On a test suite of well-known reaction mechanisms—such as amide formation, proton transfers, and pericyclics—AIMNet2-rxn yields 1 to 2 kilocalories per mole accuracy across reaction coordinates without retraining or system-specific fine-tuning. To exploit graphics processing unit parallelism and AIMNet2-rxn efficiency, we introduce a batched nudged elastic band procedure that readily achieves minimum energy pathway search on a millions-of-reactions scale. To demonstrate complex reaction characterization, the thermodynamics of an 11-step pathway producing hydroxymethylfurfural, the experimentally observed major product of glucose pyrolysis, is evaluated. Overall, the accuracy and efficiency afforded by AIMNet2-rxn create opportunities in high-throughput reaction discovery and deep reaction network analysis that would be infeasible with QM methods.
SN - 2375-2548
ER -