Historical NMR shift predictions and bibliography
This is an overview of errors achieved by various nuclear magnetic resonance (NMR) shift prediction methods over time. This list is restricted to small organic molecules, e. g. protein NMR prediction papers are not considered. Previous versions of this table were published in Prediction of chemical shift in NMR: a review and Nuclear Magnetic Resonance and Artificial Intelligence. The table and the image have been compiled by Stefan Kuhn, with help from Nils Schlörer and others. Any additions (new papers, new nuclei) are welcome. This also serves as a shift prediction bibliography.
As explained in the papers, the figures need to be considered with care. In particular, the issue of training and test sets and their influence on the results needs to be considered. Also note:
- Some methods are represented several times, which does not mean they are identical. This is clear for neural net works (classic or CNN), where the architectures can be vastly different, but also e.g. HOSE code implementations tend to be vary the original publication. For increments, there are many different sets of rules and values.
- Data may vary over time. E.g. NMRShiftDB in 2007 had different data from 2008.
- Not all publications are the original publication for a method, since they do not necessarily contain quantitative information (e.g. the original HOSE code paper [37] does not quantitatively evaluate the method).
- Not all publications contain 13C and 1H prediction, or the other nuclei listed.
The following table shows Historically achieved MAEs for various methods and datasets. It contains publications known to us by 28/3/2025.
Date | 1H MAE (ppm) | 13C MAE (ppm) | 17O MAE (ppm) | 15N MAE (ppm) | 19F MAE (ppm) | 195Pt MAE (ppm) | Method | Training | Test Dataset | Literature |
1/1990 | 5.5** | - | - | - | - | - | Pretsch increments | C13Shift | CSEARCH | [1] |
4/1992 | - | 1.56 | - | - | - | - | Optimized Increments | OPSI | 3 organic molecules | [2] |
5/1994 | 0.19** | - | - | - | - | - | Pretsch increments | Custom | 200 molecules | [3] |
7/1997 | - | 1.217**** | - | - | - | - | Pyridine increments | - | disubst pyridines | [35] |
7/1997 | - | 2.42**** | - | - | - | - | Pyridine increments | - | trisubst pyridines | [35] |
7/1997 | - | 3.665**** | - | - | - | - | AROSIM with orthocorrection | - | disubst pyridines | [35] |
7/1997 | - | 2.951**** | - | - | - | - | AROSIM with orthocorrection | - | trisubst pyridines | [35] |
12/2000 | - | 2.4 | - | - | - | - | classic neural network | 57 sesquiterpene lactones | 11 organic molecules | [4] |
1/2002 | 0.25 | - | - | - | - | - | classic neural network | 120 organic molecules | 259 organic molecules | [5] |
8/2002 | - | 1.2 | - | - | - | - | classic neural network | SpecInfo | Taxol | [6] |
8/2002 | - | 2.4 | - | - | - | - | classic neural network | SpecInfo | Taxol | [6] |
8/2002 | - | 3.8 | - | - | - | - | Pretsch increments | ChemDraw Pro | Taxol | [6] |
8/2002 | - | 3.7 | - | - | - | - | Pretsch increments | SpecTool | Taxol | [6] |
8/2002 | - | 1.0 | - | - | - | - | HOSE | SpecInfo | Taxol | [6] |
8/2002 | - | 2.7 | - | - | - | - | HOSE | PredictIt | Taxol | [6] |
8/2002 | - | 1.7 | - | - | - | - | HOSE | CNMR 6.0 | Taxol | [6] |
8/2002 | - | 3.4 | - | - | - | - | Cosmas 4.5 | n/a | Taxol | [6] |
8/2002 | - | 4.0 | - | - | - | - | Gaussian 98 | n/a | Taxol | [6] |
8/2002 | - | 1.9 | - | - | - | - | classic neural network | SpecInfo | 1547 shifts/100 molecules | [6] |
8/2002 | - | 2.5 | - | - | - | - | classic neural network | SpecInfo | 1547 shifts/100 molecules | [6] |
8/2002 | - | 1.4 | - | - | - | - | SpecInfo | SpecInfo | 1547 shifts/100 molecules | [6] |
5/2007 | - | 1.59 | - | - | - | - | ACD/Labs predictor | ACD/Labs | NMRShiftD | [7] |
5/2007 | - | 2.22 | - | - | - | - | CSEARCH | CSEARCH | NMRShiftDB | [7] |
1/2008 | 0.28 | - | - | - | - | - | CHARGE | n/a | Wiley 1H NMR database | [8] |
1/2008 | 0.28 | - | - | - | - | - | Pretsch Increments | n/a | Wiley 1H NMR database | [8] |
1/2008 | 0.28 | - | - | - | - | - | Pretsch 3D increments | n/a | Wiley 1H NMR database | [8] |
1/2008 | 0.28 | - | - | - | - | - | Combined | n/a | Wiley 1H NMR database | [8] |
9/2008 | 0.154 | - | - | - | - | - | HOSE | NMRShiftDB | * | [9] |
9/2008 | 0.154 | - | - | - | - | - | J48 | NMRShiftDB | * | [9] |
9/2008 | 0.154 | - | - | - | - | - | RF | NMRShiftDB | * | [9] |
9/2008 | 0.154 | - | - | - | - | - | SVM | NMRShiftDB | * | [9] |
5/2009 | - | 1.85 | - | - | - | - | custom encoding partial least squares regression | 190,000 structures | 16,000 structures | [10] |
1/2010 | - | 1.58 | - | - | - | - | HOSE | ACD/Labs | 205 molecules/2531 shifts | [11] |
1/2010 | - | 1.91 | - | - | - | - | NN | ACD/Labs | 205 molecules/2531 shifts | [11] |
1/2010 | - | 2.15 | - | - | - | - | increment | n/a | 205 molecules/2531 shifts | [11] |
1/2010 | - | 3.29 | - | - | - | - | DFT-GIAO | n/a | 205 molecules/2531 shifts | [11] |
1/2010 | 0.1081 | - | - | - | - | - | improved HOSE | in house | 282 | [12] |
4/2019 | 0.25 | 2.82 | - | - | - | - | stereo HOSE | nmrshitdb2 | * | [13] |
4/2019 | 0.29 | 3.52 | - | - | - | - | HOSE | nmrshitdb2 | * | [13] |
5/2019 | - | 1.63 | - | - | - | - | modgraph (HOSE + NN) | modgraph | 13 molecules | [14] |
5/2019 | - | 2.2 | - | - | - | - | ML | MestreLab | 13 molecules | [14] |
5/2019 | - | 1.7 | - | - | - | - | ML | MestreLab | 13 molecules | [14] |
5/2019 | - | 1.3 | - | - | - | - | Ensemble of previous three | as per method | 13 molecules | [14] |
8/2019 | 0.28 | 1.43 | - | - | - | - | CNN | subset of nmrshitdb2 | * | [15] |
11/2019 | 0.23 | 2.45 | - | - | - | - | ML | DFT calculations | 410 structures | [16] |
4/2020 | 0.224 | 1.355 | - | - | - | - | MPNN | subset of nmrshitdb2 | * | [17] |
2/2021 | 0.11*** | 0.70*** | 1.69*** | 2.47*** | - | - | Δ-machine learning | 57456 DFT calculations | 3780 DFT calculations | [18] |
8/2021 | - | - | - | 6.12 | - | - | kernel ridge regression | DFT 623 molecule | DFT 205 molecules | [19] |
12/2021 | 0.10 | 1.26 | - | - | - | - | GNN | Subset of nmrshitdb2 | * | [32] |
8/2022 | - | - | - | - | 52.63 | - | CNN | 19F subset of nmrshitdb2 | * | [20] |
8/2022 | - | - | - | - | 8.88 | - | improved HOSE | 19F subset of nmrshitdb2 | * | [20] |
8/2022 | 0.191 | 1.228 | - | - | - | - | GNN | Subset of nmrshitdb2 | * | [31] |
2/2023 | - | - | - | - | - | 170.07 | Laplacian kernel ridge regression | Own | * | [34] |
9/2023 | 0.209 | 2.18 | - | - | - | - | FullSSPrUCe (GNN) | nmrshiftdb2 | * | [21] |
11/2023 | 0.168 | 2.938 | - | - | - | - | ComENet | GlycoNMR 80% | GlycoNMR 10% | [22] |
11/2023 | 0.145 | 2.550 | - | - | - | - | DimeNet++ | GlycoNMR 80% | GlycoNMR 10% | [22] |
11/2023 | 0.140 | 2.492 | - | - | - | - | SchNet | GlycoNMR 80% | GlycoNMR 10% | [22] |
11/2023 | 0.146 | 3.044 | - | - | - | - | SphereNet | GlycoNMR 80% | GlycoNMR 10% | [22] |
12/2023 | 0.138 | 1.79 | - | - | - | - | fragment-based | COLMAR | 768 COLMAR Metabolites | [23] |
3/2024 | 0.210 | 2.228 | - | - | - | - | GNN | nmrshiftdb2 subset | HMDB and CH-NMR-NP | [24] |
5/2024 | 0.10 | - | - | - | - | - | GNN | PROSPRE 3755 compounds | PROSPRE 272 compounds | [25] |
5/2024 | - | 0.7 | - | - | - | - | DFT | - | 132 shifts | [26] |
6/2024 | 0.185 | 0.944 | - | - | - | - | DFT+3D GNN | nmrshiftdb2 80% | nmrshiftdb2 20% | [27] |
6/2024 | - | 0.9*** | - | - | - | - | GNN | 2026 organic molecules | 171 benzenic structures | [28] |
8/2024 | 0.035**** | 0.31**** | - | - | - | - | E(3) equivariant graph neural network | CASPER | Monosaccharides | [29] |
8/2024 | 0.026**** | 0.23**** | - | - | - | - | E(3) equivariant graph neural network | CASPER | Disaccharides | [29] |
8/2024 | 0.033**** | 0.30**** | - | - | - | - | E(3) equivariant graph neural network | CASPER | Trisaccharides | [29] |
11/2024 | 0.158 | 1.189 | - | - | - | - | GNN | nmrshiftdb2 | * | [30] |
12/2024 | - | - | - | - | 3.31 | - | Gradient Boosting Regression | - | 501 fluorinated compounds | [33] |
2/2025 | 0.167 | 2.025 | - | - | - | - | Multi-task pre-training and unsupervised learning | nmrshiftdb2 | 479 molecules | [36] |
*Employs crossvalidation with identical dataset
**Standard deviation instead of mean absolute error, error excludes atoms which are considered impossible to predict.
***Root mean squared error instead of mean absolute error.
****Not included in graph since a specialized subset for testing only.
Abbreviations: ML=machine learning, HOSE=Hierarchically Ordered Spherical Description of Environment, DFT-GIAO=density functional theory - independent atomic orbitals, CNN=Convolutional Neural Network, MPNN=message passing neural network
This figure uses the data from the above table, showing results for 1H and 13C NMR predictions from 1990 until now, ordered by publication time. The light lines give the least squares linear regression up to 6/2024 (this is the data from Nuclear Magnetic Resonance and Artificial Intelligence). The dashed light lines are the least squares linear regression for data up to 2/2001 (this is the data from Prediction of chemical shift in NMR: a review). Squares indicate ab initio calculations, triangles deep learning methods, pentagons combined methods. As before, conclusions require careful analysis.
Chart as pdf
Literature
- Fürst A, Pretsch E. A computer program for the prediction of 13-C-NMR chemical shifts of organic compounds. Analytica Chimica Acta 1990;229:17-25. https://www.sciencedirect.com/science/article/pii/S0003267000851053.
- Chen L, Robien W. Optimized prediction of 13C-NMR spectra using increments. Comparison with other methods. Fresenius' Journal of Analytical Chemistry 1992 Apr;344(4):214-216. https://doi.org/10.1007/BF00322714
- Bürgin Schaller R, Pretsch E. A computer program for the automatic estimation of 1H NMR chemical shifts. Analytica Chimica Acta 1994;290(3):295-302. https://www.sciencedirect.com/science/article/pii/0003267094801169.
- Magri Fátima M. M., Militão JSL, Ferreira MJP, Brant AJC, Emerenciano VP. A new program to 13C NMR spectrum prediction based on tridimensional models. Spectroscopy 2001;15:99-117.
- Aires-de Sousa J, Hemmer MC, Gasteiger J. Prediction of 1H NMR chemical shifts using neural networks. Anal Chem 2002 Jan;74(1):80-90.
- Meiler J, Maier W, Will M, Meusinger R. Using Neural Networks for 13C NMR Chemical Shift Prediction-Comparison with Traditional Methods. Journal of Magnetic Resonance 2002;157(2):242-252. https://www.sciencedirect.com/science/article/pii/S1090780702925991.
- Sasaki R, NMRShiftDB, ACD/Labs, and Modgraph; 2021 (accessed January 12, 2021). https://acdlabs.typepad.com/ my_weblog/2007/05/nmrshiftdb_acdl.html.
- Davies AN. Who has the best proton NMR crystal ball? Spectroscopy Europe 2008;20(1):20-23.
- Kuhn S, Egert B, Neumann S, Steinbeck C. Building blocks for automated elucidation of metabolites: Machine learning methods for NMR prediction. BMC Bioinformatics 2008 Sep;9(1):400. https://doi.org/10.1186/1471-2105-9-400.
- Blinov KA, Smurnyy YD, Churanova TS, Elyashberg ME, Williams AJ. Development of a fast and accurate method of 13C NMR chemical shift prediction. Chemometrics and Intelligent Laboratory Systems 2009;97(1):91-97. https://www.sciencedirect.com/science/article/pii/S0169743909000112, selected papers presented at the 6th Winter Symposium on Chemometrics Kazan, Russia 16-22 February 2008.
- Elyashberg M, Blinov K, Smurnyy Y, Churanova T, Williams A. Empirical and DFT GIAO quantum-mechanical methods of 13C chemical shifts prediction: competitors or collaborators? Magnetic Resonance in Chemistry 2010;48(3):219-229. https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/abs/10.1002/mrc.2571.
- Spanton SG, Whittern D. The development of an NMR chemical shift prediction application with the accuracy necessary to grade proton NMR spectra for identity. Magnetic Resonance in Chemistry 2009;47(12):1055-1061. https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/abs/10.1002/mrc.2512.
- Kuhn S, Johnson SR. Stereo-Aware Extension of HOSE Codes. ACS Omega 2019 Apr;4(4):7323-7329.
- Cobas C, Ensemble NMR Prediction; 2021 (accessed March 27, 2021). https://resources.mestrelab.com/ensemble-nmr-prediction/.
- Jonas E, Kuhn S. Rapid prediction of NMR spectral properties with quantified uncertainty. Journal of Cheminformatics 2019 Aug;11(1):50.
- Gerrard W, Bratholm LA, Packer MJ, Mulholland AJ, Glowacki DR, Butts CP. IMPRESSION - prediction of NMR parameters for 3-dimensional chemical structures using machine learning with near quantum chemical accuracy. Chem Sci 2020;11:508-515. http://dx.doi.org/10.1039/C9SC03854J.
- Kwon Y, Lee D, Choi YS, Kang M, Kang S. Neural Message Passing for NMR Chemical Shift Prediction. Journal of Chemical Information and Modeling 2020;60(4):2024-2030. https://doi.org/10.1021/acs.jcim.0c00195, pMID: 32250618.
- Unzueta PA, Greenwell CS, Beran GJO. Predicting Density Functional Theory-Quality Nuclear Magnetic Resonance Chemical Shifts via &Dealta;-Machine Learning. J Chem Theory Comput 2021 Feb;17(2):826-840.
- W. Gerrard, C. Yiu, C. P. Butts, Prediction of 15N chemical shifts by machine learning, Magn Reson Chem 2022, 60(11), 1087. https://doi.org/10.1002/mrc.5208.
- S. Kuhn, R. M. Borges, F. Venturini and M. Sansotera, "Dataset Size and Machine Learning - Open NMR Databases as a Case Study," 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), Los Alamitos, CA, USA, 2022, pp. 1632-1636, doi: 10.1109/COMPSAC54236.2022.00259.
- Williams, J.; Jonas, E. Rapid prediction of full spin systems using uncertainty-aware machine learning. Chem. Sci. 2023, 14, 10902-10913. https://doi.org/10.1039/D3SC01930F.
- Chen, Z.; Badman, R.P.; Foley, L.; Woods, R.; Hong, P. GlycoNMR: Dataset and benchmarks for NMR chemical shift prediction of carbohydrates with graph neural networks, 2023, [arXiv:cs.LG/2311.17134].
- Rigel, N.; Li, D.W.; Brüschweiler, R. COLMARppm: A Web Server Tool for the Accurate and Rapid Prediction of 1H and 13C NMR Chemical Shifts of Organic Molecules and Metabolites. Analytical Chemistry 2024, 96, 701-709. https://doi.org/10.1021/ 352
acs.analchem.3c03677.
- Li, Y.; Xu, H.; Hong, P. AI-enabled prediction of NMR spectroscopy: Deducing 2-D NMR of carbohydrate, 2024, [arXiv:cs.LG/2403.11353].
- Sajed, T.; Sayeeda, Z.; Lee, B.L.; Berjanskii, M.; Wang, F.; Gautam, V.; Wishart, D.S. Accurate Prediction of 1H NMR Chemical Shifts of Small Molecules Using Machine Learning. Metabolites 2024, 14. https://doi.org/10.3390/metabo14050290.
- Ramos, S.A.; Mueller, L.J.; Beran, G.J.O. The interplay of density functional selection and crystal structure for accurate NMR chemical shift predictions. Faraday Discuss. 2024, pp. -. https://doi.org/10.1039/D4FD00072B.
- Han, C.; Zhang, D.; Xia, S.; Zhang, Y. Accurate Prediction of NMR Chemical Shifts: Integrating DFT Calculations with Three-Dimensional Graph Neural Networks. Journal of Chemical Theory and Computation 2024. https://doi.org/10.1021/acs.jctc.4c00422.
- Duprat, F.; Ploix, J.L.; Dreyfus, G. Can Graph Machines Accurately Estimate 13C NMR Chemical Shifts of Benzenic Compounds? Preprints 2024. https://doi.org/10.20944/preprints202406.0511.v1.
- Maria Bånkestad; Kevin M. Dorst; Göran Widmalm; Jerk Rönnols. Carbohydrate NMR chemical shift prediction by GeqShift employing E(3) equivariant graph neural networks. RSC Advances, 2024, 14, 26585-26595.
- Chen, H., Liang, T., Tan, K. et al. GT-NMR: a novel graph transformer-based approach for accurate prediction of NMR chemical shifts. J Cheminform 16, 132 (2024).
- Jongmin Han, Hyungu Kang, Seokho Kang, Youngchun Kwon, Dongseon Lee, Youn-Suk Choi. Scalable graph neural network for NMR chemical shift prediction. Phys. Chem. Chem. Phys. 2022 24, 26870-26878.
- Yanfei Guan, S. V. Shree Sowndarya, Liliana C. Gallegos, Peter C. St. John, Robert S. Paton. Real-time prediction of 1H and 13C chemical shifts with DFT accuracy using a 3D graph neural network. Chem. Sci., 2021, 12, 12012.
- Sophia Li, Emma Wang, Leia Pei, Sourodeep Deb, Prashanth Prabhala, Sai Hruday Reddy Nara, Raina Panda, Shiven Eltepu, Marx Akl, Larry McMahan, Edward Njoo, Evaluation of machine learning models for the accelerated prediction of density functional theory calculated 19F chemical shifts based on local atomic environments, Artificial Intelligence Chemistry,
Volume 2, Issue 2, 2024, 100078.
- E. E. Ondar, M. V. Polynski, V. P. Ananikov, Predicting 195Pt NMR Chemical Shifts in Water-Soluble Inorganic/Organometallic Complexes with a Fast and Simple Protocol Combining Semiempirical Modeling and Machine Learning. ChemPhysChem 2023, 24, e202200940.
- St. Thomas, I. Brüuhl, D. Heilmann, E. Kleinpeter, 13C NMR Chemical Shift Calculations for Some Substituted Pyridines: A Comparative Consideration. J. Chem. Inf. Comput. Sci. 1997, 37, 4, 726-730.
- Li, Y., Xu, H., Kumar, A. et al. TransPeakNet for solvent-aware 2D NMR prediction via multi-task pre-training and unsupervised learning. Commun Chem 8, 51 (2025).
- W. Bremser, Hose - a novel substructure code, Analytica Chimica Acta, Volume 103, Issue 4, 1978, Pages 355-365,