Neural Message Passing for NMR Chemical Shift Prediction

Area of Science:

• Computational chemistry
• Machine learning
• Spectroscopy

Background:

• Accurate prediction of Nuclear Magnetic Resonance (NMR) spectra is crucial for large-scale molecular structure validation and elucidation.
• Existing methods may lack the precision required for complex molecular analysis.

Purpose of the Study:

• To develop an improved machine learning method for predicting NMR chemical shifts of atoms in novel molecules.
• To enhance the accuracy and efficiency of NMR spectral prediction for automated structure analysis.

Main Methods:

• Utilized a message passing neural network operating on a graph representation of molecules.
• Enhanced graph representation by implicitly treating hydrogen atoms and incorporating diverse node and edge features.
• Trained the model on an NMR database for chemical shift prediction.

Main Results:

• Achieved higher prediction performance for chemical shifts in both proton (¹H) NMR and carbon-13 (¹³C) NMR spectra of small molecules.
• Demonstrated the method's effectiveness in determining correct molecular structures by searching candidate molecules against new NMR spectra.

Conclusions:

• The proposed method offers a significant advancement in predicting NMR chemical shifts.
• This approach facilitates more reliable and scalable automated molecular structure validation and elucidation using NMR data.