Practical Machine Learning Strategies. 5. Rapid, Accurate Prediction of 1H and 13C NMR Shifts in Conformationally Flexible Organic Molecules
Year: 2026
Type: preprint
Abstract: Accurate prediction of NMR chemical shifts for conformationally flexible natural products remains a central challenge in structure elucidation. Traditional quantum‑chemical (QM) approaches typically achieve ≈2.0–3.0 ppm 13 C accuracy for flexible natural products but often require hours of computation per molecule. MLHF‑NMR reproduces high‑level QM 13 C accuracy for flexible molecules while reducing computation times by factors of hundreds to more than a thousand, enabling routine, high‑throughput NMR prediction for conformationally flexible natural products. Against the high‑accuracy DELTA50 experimental 1 H and 13C shift benchmark, MLHF‑NMR achieves RMS errors of 0.15 ppm ( 1 H) and 1.73 ppm ( 13 C). Across more than 8.2k 13 C chemical shifts from 420 predominantly flexible natural products, the protocol yields an average RMS error of 2.23 ppm. Practical applications include structure assignment for newly isolated natural products, structure confirmation in independently synthesized molecules, rapid detection and correction of misassignments, and strategies for improving predicted 13C shift accuracy. Validation sets and sample MLHF‑NMR 13 C shift predictions for more than 1900 synthetic drugs and drug‑like molecules are available at https://mlnmr.wavefun.com. For flexible molecules, MLHF‑NMR routinely delivers 13 C shift accuracy comparable to the best QM methods while reducing computation time by 2–3 orders of magnitude.
Source: ChemRxiv
Authors: William Ohlinger, Philip Klunzinger, Thomas Hehre, Andy Driessen, Bernard Deppmeier, Jeffrey Johnson, Connor Ohlinger, Warren Hehre, Jurgen Schnitker, Noritaka Uchida, Yuusuke Watanabe, Ryan D. Cohen, Ian J. Bruno, Masaru Hashimoto, Ryuhi Kanehara, Seth B. Herzon, Hannah J. Riseman, José L. López-Pérez
Language: English
Subfield: Spectroscopy
Field: Chemistry
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7144300865