Aug 2026· Journal of Chemical Information and Modeling· 0 citations· 35 references
TL;DR
A machine-learning-assisted framework to improve quantum-chemical prediction of 19F NMR chemical shifts by using machine learning to diagnose and correct subset-dependent limitations in the shielding-shift relationship within a practical quantum-chemical workflow is developed.
Abstract
19F nuclear magnetic resonance (NMR) spectroscopy is widely used for structural elucidation of fluorinated molecules, but reliable prediction of 19F chemical shifts remains challenging because quantum-chemical calculations are computationally demanding and their accuracy can vary across diverse molecular environments. In this work, we develop a machine-learning-assisted framework to improve quantum-chemical prediction of 19F NMR chemical shifts. A data set of 2605 experimental shifts was compiled from the literature, and isotropic shielding constants were calculated using a density functional theory (DFT)/gauge-including atomic orbital (GIAO) calculation protocol. Machine learning was then used to analyze the relationship between calculated shielding values and experimental chemical shifts. The analysis indicates that the data set can be partitioned into operationally defined, structure-associated regimes in which the mapping between calculated shielding and experimental shift differs systematically. By identifying the structural characteristics of these regimes and constructing prediction models separately for each region, the overall predictive accuracy of the quantum-chemical framework is significantly improved. The resulting models achieve mean absolute errors below 4 ppm and show practical promise under the tested benchtop 60 MHz conditions after simple linear calibration. Application to a fluorinated reaction mixture further demonstrates the utility of the approach for assisting spectral interpretation and prioritizing candidate structures. These results show that the main contribution of the present work is not simply applying machine learning to 19F NMR prediction, but using machine learning to diagnose and correct subset-dependent limitations in the shielding-shift relationship within a practical quantum-chemical workflow.
Fast and accurate chemical shielding estimators are essential for shielding-driven Nuclear Magnetic Resonance (NMR) crystallography. Machine-learning models for shielding predictions have matured significantly and today are primarily limited by the electronic structure reference data they are trained on. Here, we introduce ShiftML4, a shielding-tensor model trained directly on monomer-corrected calculations that approximate PBE0, rather than the PBE reference targeted by earlier ShiftML models. ShiftML4 is trained on a diverse set of structures containing 12 of the most common NMR nuclei in molecular organic solids. On experimental benchmark sets, the 13C isotropic RMSE against experiment is 1.67 ppm, compared with 2.34 ppm for GIPAW-PBE on the same geometries. ShiftML4 gives a similar 1H prediction RMSE to ShiftML3 (0.5 ppm) and improves the 15N RMSE from 7.24 to 6.08 ppm. The model also reduces errors in the shielding-tensor anisotropy, with an RMSE of 4.63 ppm on 13C CSA principal components against 5.85 ppm for GIPAW. The improvements in prediction accuracy are retained on better geometries. Basing shift predictions on structures relaxed with PET-MOLS, a recent machine-learned interatomic potential that reaches approximate hybrid-DFT geometries in seconds, lowers the ShiftML4 errors further to 0.48 ppm (1H), 1.49 ppm (13C) and 3.66 ppm (15N).
Matthias Kellner, Ruben Rodriguez-Madrid, Jacob B. Holmes et al.· 0 citations
We evaluated a fully machine-learning-assisted workflow for NMR crystallography by combining the Universal Model for Atoms (UMA) interatomic potential for crystal structure optimization with the ShiftML3 prediction of solid-state NMR shieldings. Benchmarking against conventional periodic density functional theory (DFT) calculations for 1H, 13C, and 15N chemical shifts demonstrates that ML-based geometry optimization consistently improves the accuracy of 13C and 15N predictions relative to standard PBE optimization, highlighting the dominant role of structural refinement. ShiftML3 achieves DFT-level accuracy for shielding prediction and, when combined with UMA-optimized geometries, matches or surpasses periodic DFT for 13C and 15N while reducing the computational cost by orders of magnitude. We further show that hybrid PBE0 single-molecule corrections remain effective for both DFT- and ShiftML3-derived shieldings, extending their applicability to modern machine-learning models. These results establish a new computational paradigm for NMR crystallography by replacing both computational bottlenecks of the conventional DFT workflow with modern machine-learning models.
E.v.a. Chaloupecká, O. Socha, Martin Dračínský· Journal of Physical Chemistr...· 0 citations
Accurately predicting NMR chemical shifts of exchangeable protons in solution remains challenging because of the combined influence of solute–solvent interactions and molecular dynamics. We introduce a framework that integrates machine-learning molecular dynamics (ML-MD) with the ShiftML3 machine-learning shielding model for rapid and accurate prediction of NMR spectra in solvated molecules. Although originally developed for solids, ShiftML3 effectively captures intermolecular contributions to shielding in solution. We validate the method across a range of chemically diverse systems, including water in organic solvents, solvated alcohols, hydrogen-bonded nucleobases, glucose anomers, and alkylated acetamides. The ML-MD + ShiftML3 framework reproduces experimentally observed chemical shifts of exchangeable protons with near-quantitative accuracy, resolving subtle hydrogen-bonding and conformational effects that implicit-solvent DFT fails to capture. These results establish ML-MD + ShiftML3 as a transferable and computationally efficient way of incorporating solvation and dynamics into NMR spectroscopy, enabling realistic chemical shift predictions for flexible, hydrogen-bonded, and complex molecular systems. The authors develop a framework integrating machine-learning molecular dynamics with the ShiftML3 machine-learning shielding mode to accurately predict NMR chemical shifts of exchangeable protons in solution, outperforming DFT approaches in capturing solvation effects.
O. Socha, Jana Pavlišová, Debashree Manna et al.· Nature Communications· 2 citations
MolDeTr addresses the spectrum-conditioned inverse problem and extracts spin-system parameters directly from measured 1D 1H NMR spectra, thereby substantially improving chemical-shift prediction precision by one to 2 orders of magnitude compared to existing structure-conditioned approaches.
N. Schmid, Marc Wanner, G. Fischetti et al.· Analytical Chemistry· 1 citation
A manually verified, solvent-annotated 11B NMR data set constructed via a large language model (LLM)-assisted workflow provides a form of virtual spectral resolution, enabling the discrimination of chemically inequivalent boron sites that are difficult to resolve experimentally.
Penghui Li, Ben Gao, Shiyang Wang et al.· JACS Au· 0 citations
Accurate prediction of physicochemical properties is increasingly limited by an information ceiling of structure-only molecular descriptors. Here, predicted 1H|13C NMR chemical shifts are transformed into fixed-length NMR vectors and concatenated with ECFP4 to form the hybrid spectral–structural representation SpectraPRINTS, enabling direct evaluation of representational complementarity across logP, logS, and logD (pH 2.6, 7.4, and 10.5), as well as the most acidic and most basic pKas. With a fixed learning protocol, SpectraPRINT reduces error for lipophilicity- and solubility-related end points (up to 39% lower RMSE vs ECFP4), while no systematic gain is observed for the most acidic and most basic macroscopic pK a end points. The workflow is released as NMR-AI, a freely accessible web platform integrating NMR spectra prediction, descriptor construction, and property prediction, enabling interactive use and independent validation. The NMR-AI platform is accessible at https://cheminformaticsportal.if-pan.krakow.pl/.
Wojciech Pietruś, Arkadiusz Leniak, R. Kurczab· Journal of Chemical Informat...· 0 citations
