Accessible hybrid DFT-quality NMR crystallography via gas-phase machine learning interatomic potentials
- 1National High Magnetic Field Laboratory, Florida State University 1800 E. Paul Dirac Dr Tallahassee FL 32310 USA
- 2Department of Chemistry and Biochemistry, Florida State University Tallahassee FL 32306 USA
Abstract
Nuclear magnetic resonance (NMR) crystallography is a robust method for structure determination, but its reliance on density functional theory (DFT) calculations for geometry refinement limits its speed and accessibility. Recent machine-learning predictors such as ShiftML3 can evaluate magnetic shieldings in seconds, but still need high-quality crystal geometries that are normally obtained from slow DFT optimisations. Here, we demonstrate that machine learning interatomic potentials (MLIPs) can eliminate this bottleneck for organic crystals. Interestingly, models trained on gas-phase molecules at the hybrid-DFT ωB97M-V level (OMol25 dataset)—such as UMA-omol and MACE-Polar-1—deliver structural quality rivaling hybrid periodic-DFT, without requiring large computational resources. The resulting MLIP/ShiftML3 workflow reduces computational costs by at least three orders of magnitude, making high-accuracy NMR crystallography accessible without HPC infrastructure, and opening the door to fast molecular dynamics simulations at the hybrid DFT level. Combined with sensitivity enhancement via dynamic nuclear polarization (DNP), we demonstrate the approach on l-histidine, extracting 13C and 15N chemical shift tensors and using 1H chemical shifts at natural isotopic abundance to discriminate between its monoclinic and orthorhombic polymorphs.
Related
- Correcting the Liquid Densities of Universal Machine-Learning Interatomic Potentials Using Local-Number-Density Repulsion
- pdb2reaction: End-to-End Reaction-Path Elucidation from PDB Structures Using Machine-Learning Interatomic Potentials
- Applicability Domain of Universal Machine Learning Interatomic Potentials at Solid-State Battery Interfaces
- How accurate are foundational machine learning interatomic potentials for heterogeneous catalysis?
- Reference Conventions and Structure Types, Rather Than 4f Occupancy, Shape Lanthanide Errors in Universal Machine-Learning Interatomic Potentials
BibTeX
@article{gunaga2026accessible,
title = {{Accessible hybrid DFT-quality NMR crystallography via gas-phase machine learning interatomic potentials}},
author = {Shubha S. Gunaga and Robert W. Schurko and Sean T. Holmes and Frederic Mentink-Vigier},
journal = {Chemical Science},
year = {2026},
month = sep,
doi = {10.1039/d6sc04941a},
url = {https://doi.org/10.1039/d6sc04941a},
note = {CC BY-NC 3.0},
}
RIS
TY - JOUR
TI - Accessible hybrid DFT-quality NMR crystallography via gas-phase machine learning interatomic potentials
AU - Shubha S. Gunaga
AU - Robert W. Schurko
AU - Sean T. Holmes
AU - Frederic Mentink-Vigier
PY - 2026
DA - 2026/09/08
JO - Chemical Science
DO - 10.1039/d6sc04941a
UR - https://doi.org/10.1039/d6sc04941a
AB - Nuclear magnetic resonance (NMR) crystallography is a robust method for structure determination, but its reliance on density functional theory (DFT) calculations for geometry refinement limits its speed and accessibility. Recent machine-learning predictors such as ShiftML3 can evaluate magnetic shieldings in seconds, but still need high-quality crystal geometries that are normally obtained from slow DFT optimisations. Here, we demonstrate that machine learning interatomic potentials (MLIPs) can eliminate this bottleneck for organic crystals. Interestingly, models trained on gas-phase molecules at the hybrid-DFT ωB97M-V level (OMol25 dataset)—such as UMA-omol and MACE-Polar-1—deliver structural quality rivaling hybrid periodic-DFT, without requiring large computational resources. The resulting MLIP/ShiftML3 workflow reduces computational costs by at least three orders of magnitude, making high-accuracy NMR crystallography accessible without HPC infrastructure, and opening the door to fast molecular dynamics simulations at the hybrid DFT level. Combined with sensitivity enhancement via dynamic nuclear polarization (DNP), we demonstrate the approach on l-histidine, extracting 13C and 15N chemical shift tensors and using 1H chemical shifts at natural isotopic abundance to discriminate between its monoclinic and orthorhombic polymorphs.
SN - 2041-6520
SN - 2041-6539
ER -