Model Predicts Molecular Energy Gaps From NMR Data

The research-stage tool could give chemists a faster read on electronic behavior, though it is weaker for sulfur- and boron-containing molecules.

Updated on Nov. 2, 2026 in Chemistry

Isometric editorial illustration of a ball-and-stick molecular model beside a stepped ladder of glowing energy levels, evoking molecular electronic structure.
Yale researchers unveiled a machine learning model that predicts molecular Homo-Lumo energy gaps from routine NMR spectral data, though it performs less well on sulfur- and boron-containing molecules. AI Illustration. Upload story photo >

Yale University researchers developed a machine learning model that predicts the Homo-Lumo energy gap of molecules from NMR spectral data. The model is research-stage and has not been released as a tool; it showed weaker performance for molecules containing sulfur and boron and for zwitterions.

Why it matters

The Homo-Lumo energy gap governs how molecules absorb light and conduct electricity, so predicting it from routine NMR readouts could speed up screening in chemistry. The approach reads local electron density from chemical shift data, a measurement chemists already collect.

The model uses eight features derived from chemical shift data, such as number of peaks and mean, maximum, or minimum shift values, against about 400 features in some comparison models. It was trained on over 100,000 small molecules, though mostly on computationally generated NMR data rather than measured spectra.

The players

Yale University

A private research university whose chemistry and data-science groups conducted the NMR-based machine learning study.

Kevin Jablonka

A researcher at Friedrich Schiller University Jena in Germany who commented on the work and specializes in machine learning for chemistry.

The details

NMR, or nuclear magnetic resonance spectroscopy, measures the chemical shift of atomic nuclei, a value that reflects the local electronic environment around each atom. The researchers condensed each molecule's chemical shift data into eight summary features, including the number of peaks and their mean, maximum, and minimum values. These features were fed into a mathematical model that estimates the Homo-Lumo gap, the energy difference between a molecule's highest occupied and lowest unoccupied orbitals.

Timeline

  1. No specific dates or timeframes were reported for the study.

The Tech Race

The work sits alongside machine learning efforts that predict molecular properties from structure, and its distinctive contribution is using NMR readouts as the input. The team says it is considering mechanism-specific descriptors for reactivity that could be localized to NMR data.

The model is a research prototype with no announced release, pricing, or availability timeline. Chemists are unlikely to use it directly until its limits with sulfur, boron, and zwitterionic molecules are addressed.

The takeaway

Watch for follow-up work that extends the model to experimental NMR data and to the sulfur, boron, and zwitterion cases where it currently underperforms. The team's stated plan to pursue user-friendly, chemistry-adjacent applications will indicate whether the approach reaches working chemists.

Further reading

For related background, explore recent Chemistry coverage across the United States.

Source note: This article includes information reported by Chemistry World.