Toward fast ultrahigh-resolution NMR spectroscopy with deep learning

Nuclear magnetic resonance (NMR) spectroscopy serves as a robust tool for the non-invasive probing of molecular structure, dynamics, and interactions. Nevertheless, its fundamental application scenarios remain confined by the requirements of fast data acquisition and high spectral resolution. Herein, this study combines artificial intelligence and pure shift NMR to implement fast ultrahigh-resolution 1D and 2D NMR spectroscopy. This protocol delivers pure shift NMR spectra with highly accelerated data acquisition along with high-fidelity peak reconstruction, for ultrahigh-resolution access to overlapped signals. Two types of fast ultrahigh-resolution 2D NMR, including pure shift DOSY and 2D J experiments, are implemented for efficient dynamics analysis and molecular structure determination. Further, this protocol's application is tested on the in situ observation of electrocatalysis reactions of 1-butanol. As a result, this study affords a fundamental innovation to artificial intelligence-enabled NMR spectroscopy and may boost the application landscapes of pure shift NMR techniques.