2D-JCOG: Transforming 1D ¹H NMR Spectra into J-δ Correlation Maps via the Shared Splitting Theorem and Graph Neural...

Area of Science:

• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Computational Chemistry
• Machine Learning

Background:

• Scalar coupling constants in 1H NMR spectra provide crucial structural information but are challenging to extract automatically.
• Classical methods like multiplet analysis fail under spectral congestion or strong coupling.
• Iterative quantum-mechanical (QM) methods require accurate initial estimates and can face convergence problems.

Purpose of the Study:

• To develop a robust, automated method for extracting scalar coupling constants and visualizing coupling networks from 1H NMR data.
• To overcome limitations of traditional spectral analysis techniques, particularly in complex or moderately coupled systems.

Main Methods:

• Introduced 2D-JCOG, a deep learning framework utilizing Graph Neural Networks (GNNs).
• Leveraged the Shared Splitting Theorem (SST) as a physical invariant to identify coupled spin systems.
• Employed a hybrid message-passing strategy with attention and mean-aggregation layers for robust feature extraction and context aggregation.

Main Results:

• Achieved high performance on QM-simulated spectra, with 92-96% recall and 82-91% precision across varying complexities.
• Validated on experimental data (GISSMO database) with 91.9% recall and a mean J-value error of 0.113 Hz.
• Demonstrated applicability to moderately coupled systems where first-order approximations fail.

Conclusions:

• 2D-JCOG offers a powerful, automated alternative for J-coupling extraction and visualization in 1H NMR spectral analysis.
• The method enhances structural elucidation by providing direct visualization of coupling connectivity.
• It extends automated analysis to systems previously requiring manual interpretation or advanced QM methods.