Neural Networks for Conversion of Simulated NMR Spectra from Low-Field to High-Field for Quantitative Metabolomics

Area of Science:

• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Computational Chemistry
• Metabolomics

Background:

• Benchtop NMR instruments offer accessible and affordable NMR spectroscopy for research and industry.
• Low magnetic field NMR spectrometers present challenges in spectral resolution and signal-to-noise ratio (SNR), complicating quantitative analysis.
• Improving the quality of low-field NMR spectra is crucial for accurate analyte quantification.

Purpose of the Study:

• To evaluate neural network architectures for converting low-field (100 MHz) NMR spectra to high-field (400 MHz) spectra.
• To enhance spectral quality for improved quantitative analysis of metabolites.
• To compare the performance of direct metabolite quantification using low-field and high-field converted spectra.

Main Methods:

• Simulated NMR spectra at 100 MHz were processed using various neural network architectures, including transformers and multi-layered perceptrons (MLPs).
• The primary task involved converting low-field spectra to high-field spectra.
• MLPs were also employed for direct metabolite quantification from both simulated low-field (100 MHz) and high-field (400 MHz) spectra.

Main Results:

• The transformer network demonstrated capability in reliably converting low-field NMR spectra to high-field spectra, even in complex mixtures (21 and 87 metabolites).
• Direct metabolite quantification using MLPs on low-field spectra yielded slightly higher accuracy than using spectra converted to high-field.
• The findings suggest that direct quantification from low-field spectra may be sufficient, though further validation is required.

Conclusions:

• The transformer-based approach is effective for converting low-field NMR spectra to high-field spectra in metabolomic applications.
• This method holds potential for automating data processing in various NMR spectroscopy fields.
• Further research and experimental validation are necessary to fully establish the benefits of low-field versus high-field spectral conversion for quantification.