Sensitivity-Enhanced Pure Shift Spectroscopy Empowered by Deep Learning and PSYCHE

Proton nuclear magnetic resonance (NMR) is the most used NMR technique. However, the limited range of chemical shifts and the complicated multiplet splitting often lead to severe spectral overlapping, making spectral analysis complex. Pure shift methods convert multiplets into singlets to greatly improve the spectral resolution but lead to severe sensitivity loss, aggravating the issue of low sensitivity of NMR compared to that of other spectroscopic techniques. Pure shift yielded by the chirp excitation (PSYCHE) method is currently the most popular pure shift method. But relatively small flip angles of chirp pulses are used to achieve a balance between the sensitivity and purity, leading to relatively low spectral sensitivity. In this work, the flip angle of 60° is utilized in the PSYCHE experiment to increase the sensitivity by four times but with strong recoupling artifacts. Then, a deep neural network model is employed to remove recoupling artifacts to obtain a clean spectrum. The model can correctly recognize all peaks, remove recoupling artifacts and chunking sidebands, and retain the desired pure shift peaks. The processed spectra are clean and free of recoupling artifacts. This method is applicable for semiquantitative analysis with high accuracy, allowing the monitoring of concentration changes of different substances in the mixture. This method will promote wider NMR applications, especially on low-concentration samples.