Deep Learning Assisted Proton Pure Shift NMR Spectroscopy.

Area of Science:

• Analytical Chemistry
• Organic Chemistry
• Spectroscopy

Background:

• Proton Nuclear Magnetic Resonance (¹H NMR) spectroscopy is vital for analyzing organic molecules.
• Complex ¹H NMR spectra often suffer from signal overlap and intricate multiplet patterns, hindering analysis.
• Existing pure-shift NMR techniques can struggle with low-sensitivity samples and exchangeable protons.

Purpose of the Study:

• To develop a deep-learning approach for simplifying complex ¹H NMR spectra.
• To generate virtual homonuclear decoupled pure shift spectra from spin-echo modulated ¹H NMR data.
• To improve the sensitivity, resolution, and quantifiability of NMR analyses for challenging samples.

Main Methods:

• A novel deep-learning algorithm was designed to process spin-echo modulated ¹H NMR spectra.
• The method transforms these spectra into high-resolution singlet NMR spectra (virtual pure shift spectra).
• Uncertainty prediction was integrated into the transformation process to enable quantification.

Main Results:

• The deep-learning approach successfully generated highly sensitive and high-resolution singlet NMR spectra.
• Experimental validation on complex organic compounds demonstrated superior performance compared to current methods.
• The method effectively handles signal overlaps and provides uncertainty estimations for quantitative analysis.

Conclusions:

• The developed deep-learning method offers significant advantages for analyzing complex organic molecules using ¹H NMR.
• It enhances the characterization of low-sensitivity samples and systems with exchangeable protons.
• This approach overcomes limitations of traditional pure-shift spectra and conventional NMR analysis.