Diffusion Ordered NMR Spectroscopy (DOSY) is a powerful technique for studying mixtures by probing the diffusion behavior of mixed compounds and enabling their identification and separation in mixture samples. The performance of DOSY generally relies on the adopted reconstruction algorithm to determine diffusion coefficients from diffusion-dependent signal decays, thus producing a 2D spectrum that resolves components by chemical shift and molecular diffusion. Although deep-learning provides an effective approach to DOSY reconstruction, existing deep-learning-based reconstruction methods generally face the limitation of inadequate feature extraction, which may lead to reconstruction artifacts in some scenarios. Here, we propose STELT (Spatiotemporal Extraction Laplace Transform), a lightweight deep-learning framework based on spatiotemporal feature extraction. STELT uses a dual-branch architecture that combines a temporal convolution module to capture dynamic patterns in decay signals with a self-attention module to extract spatial features along the chemical-shift dimension. Experimental results demonstrate the proposed method achieves superior reconstruction accuracy and noise suppression with significantly reduced computational overhead, thereby offering a practical and efficient solution for high-quality DOSY analysis.
Jingmin Lin, Bo Chen, Guolan Peng et al.· Journal of Chemical Physics· 0 citations
Multidimensional NMR spectroscopy provides rich molecular-level information on species and structures, with broad significance across chemistry, biology, and materials science. However, its widespread application is generally limited by prolonged acquisition times. Combining non-uniform sampling techniques with spectra reconstruction methods offers a promising solution to this acquisition bottleneck. Traditional reconstruction methods are robust but constrained by algorithmic assumptions and approximations, whereas deep learning approaches can potentially overcome these limitations and achieve higher reconstruction fidelity, though generalization to unseen data remains challenging. Here, we present an accelerated conditional diffusion model for multidimensional NMR spectra reconstruction, formulating the task as a probabilistic iterative denoising process that progressively refines undersampled spectra under physical constraints. Experiments demonstrate that this method outperforms both traditional and end-to-end deep learning algorithms in peak recovery, artifact suppression, and robustness across multiple sampling conditions and experimental datasets.
Bo Chen, Xun Guan, Zhuoran Rong et al.· National Science Review· 0 citations
Phase-sensitive NMR spectroscopy provides essential information for accurate component identification, quantitative analysis, and structural characterization, particularly in protein studies. However, the acquisition of high-quality phase-sensitive NMR spectra with absorptive line shapes typically requires complementary quadrature acquisition and elaborate phase correction, which often involves additional experimental repetitions and time-consuming manual operations. In this study, we present a diffusion-based deep-learning framework for automatic phase-sensitive NMR spectrum reconstruction directly from common NMR experimental data, free of quadrature acquisition and phase correction operation. The proposed method formulates the phasing problem as a conditional probabilistic generative process in which a denoising network iteratively refines noisy spectral estimates toward physically consistent absorption-mode spectra under the guidance of the observed magnitude-mode data. Comprehensive validation on a diverse set of protein samples demonstrates the effectiveness and robustness of the proposed method, thus providing an effective and automated solution for phase-sensitive NMR spectroscopy reconstruction.
Zhuoran Rong, Bo Chen, Jie Shao et al.· Analytical Chemistry· 0 citations