A Very Deep Graph Convolutional Network for ¹³C NMR Chemical Shift Calculations with Density Functional Theory Level...

Area of Science:

• Computational Chemistry
• Spectroscopy
• Machine Learning

Background:

• Nuclear magnetic resonance (NMR) chemical shift calculations are vital for structure elucidation in chemistry.
• Traditional density functional theory (DFT) methods are accurate but slow, while fast data-driven methods lack reliability.
• This creates a need for accurate, fast, and reliable computational tools for NMR chemical shift prediction.

Purpose of the Study:

• To develop a highly accurate and efficient computational model for ¹³C NMR chemical shift prediction.
• To address the limitations of existing methods in terms of speed and reliability for complex systems.
• To provide a practical tool for structure assignment and validation in organic and natural product chemistry.

Main Methods:

• Construction of a 54-layer-deep graph convolutional network (GCN) for ¹³C NMR chemical shift calculations.
• Utilizing the semiempirical method GFN2-xTB as a basis for the GCN model.
• Benchmarking the model's performance against DFT calculations on structure assignment tasks.

Main Results:

• The GCN model achieved high accuracy in ¹³C NMR chemical shift calculations with significantly reduced time costs.
• The model demonstrated competitive performance compared to DFT methods in structure assignment benchmarks.
• The model successfully resolved the complex J/K ring junction problem of maitotoxin, the largest molecule assigned by NMR calculations to date.
• The model is compatible with a wide range of organic systems, including large molecules with diverse elements.

Conclusions:

• The developed GCN model offers a powerful and efficient solution for ¹³C NMR chemical shift calculations.
• This approach provides a valuable tool for routine structure validation and assignment, especially for large and complex molecules.
• The user-friendly software enables new possibilities for elucidating previously intractable molecular structures using NMR calculations.