Accelerated Pure Shift NMR Spectroscopy with Deep Learning.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Computational Chemistry

Background:

• Pure shift nuclear magnetic resonance (NMR) spectroscopy offers high spectral resolution crucial for chemical analysis.
• Conventional pure shift NMR methods often require lengthy experimental durations, limiting their practical application.
• Accelerating acquisition in NMR is essential for increasing throughput in chemical research.

Purpose of the Study:

• To demonstrate the feasibility of deep learning (DL) for fast and high-quality reconstruction of pure shift NMR spectra.
• To develop a DL protocol capable of mitigating undersampling artifacts and resolving closely spaced peaks.
• To enable accelerated acquisition of pure shift NMR data without compromising spectral quality.

Main Methods:

• Implementation of a deep learning (DL) protocol for pure shift NMR spectral reconstruction.
• Training a lightweight neural network on simulated data to minimize computational resource requirements.
• Utilizing an M-to-S strategy to enhance the generalization capability of the DL model.

Main Results:

• Successful elimination of undersampling artifacts in reconstructed pure shift NMR spectra.
• Improved resolution enabling distinction between peaks with close chemical shifts.
• Demonstration of satisfactory reconstruction performance on standard personal computers, reducing the need for extensive training data and high-performance hardware.

Conclusions:

• Deep learning provides a viable approach for accelerating pure shift NMR acquisition and enhancing spectral quality.
• The developed lightweight DL model offers a practical solution for routine chemical analysis, accessible even with limited computational resources.
• This study represents a significant advancement towards the widespread adoption of DL in NMR spectroscopy for diverse chemical applications.