Although pure shift nuclear magnetic resonance (NMR) spectroscopy has enabled ultrahigh-resolution measurements in complex systems such as reaction mixtures and biomacromolecules, time-consuming data acquisition poses a limitation for further applications. Here, the authors propose a protocol combining non-uniform chunk sampling and physics-informed deep learning reconstruction that enables sampling artifact suppression, weak signal recovery and high-fidelity reconstruction of peak intensities for the fast implementation of one-, two- and multidimensional pure shift NMR.
- Jianfeng Bao
- Yang Ni
- Haolin Zhan