Researchers Develop AI Model for High-Quality Nuclear Magnetic Resonance Spectra
Researchers from Xiamen University, in collaboration with the Chinese Academy of Sciences Institute of Precise Measurement and Innovation Technology and the University of Manchester, have developed a novel artificial intelligence model that significantly advances nuclear magnetic resonance spectroscopy. Led by Professor Lin Yanqin, the team introduced the Spin Echo to Pure Shift Network, or SE2PSNet, a deep learning architecture designed to generate high-quality pure chemical shift spectra that overcome longstanding limitations in traditional NMR analysis. Conventional hydrogen NMR spectra frequently suffer from severe peak overlap, complicating compound identification and quantitative analysis. While pure chemical shift techniques simplify multiplet signals into singlets to alleviate spectral congestion, they typically incur sensitivity loss, artifact generation, and integration distortion. SE2PSNet addresses these challenges through a structured three-module framework comprising spectral correlation, quantitative attention, and pure shift recovery. By integrating residual connections with attention mechanisms, the model amplifies features critical for quantitative analysis while minimizing information loss during complex peak deconvolution. The network employs a joint loss function combining mean squared error and relative error, ensuring balanced training that preserves weak signal reconstruction alongside dominant peaks. In validation tests using the densely coupled antibiotic azithromycin, SE2PSNet successfully resolved overlapping signals that persisted in traditional PSYCHE spectra. The reconstructed spectra demonstrated exceptional quantitative fidelity, achieving a determination coefficient of 0.9975 when compared to reference standards. The methodology establishes a new paradigm for NMR data processing that simultaneously delivers high resolution, enhanced sensitivity, and reliable quantification. The breakthrough holds substantial implications for pharmaceutical development, drug metabolism studies, and materials science, where precise spectral analysis is critical. By streamlining complex data interpretation, the technology supports the accelerated design and optimization of novel therapeutics. The research was published in Nature Communications under the title High-quality pure shift NMR spectra by deep learning using multi-spectral input and joint loss functions. Xiamen University is listed as the corresponding and first-affiliated institution, with graduate students Cai Weigang and Li Yiyang as co-lead authors. The project received funding from the National Key R&D Program of China, the National Natural Science Foundation of China, and the UK Engineering and Physical Sciences Research Council. This publication follows a sustained research trajectory by the Lin group at the intersection of machine learning and magnetic resonance spectroscopy. Their ongoing work continues to shape the methodological standards for AI-driven analytical chemistry, with multiple prior publications recognized as ESI Highly Cited Papers. The SE2PSNet framework represents a measurable step forward in translational spectroscopic analysis, offering scalable tools for academic laboratories and industrial R&D pipelines alike.
