Two-Dimensional Laplace NMR Reconstruction through Deep Learning Enhancement.

Area of Science:

• Nuclear Magnetic Resonance (NMR) Spectroscopy
• Computational Chemistry
• Data Science

Background:

• Laplace NMR provides crucial diffusion and relaxation data for molecular dynamics and spin interactions.
• Traditional Fourier NMR excels at composition and structure determination.
• Inverse Laplace Transform (ILT) processing for Laplace NMR is complex and often ill-posed, especially in 2D.

Purpose of the Study:

• To develop an effective and user-friendly method for two-dimensional Laplace NMR reconstruction.
• To address the challenges associated with the ill-posed nature of ILT in multidimensional NMR.
• To integrate physics-based knowledge with deep learning for improved NMR data processing.

Main Methods:

• A physics-informed deep learning strategy was developed for 2D Laplace NMR reconstruction.
• A forward process model was constructed to simulate multidimensional decay signals based on physical laws.
• A noniterative neural network algorithm was employed, learning prior information from synthetic data.

Main Results:

• The proposed approach demonstrated practical effectiveness in experimental 2D Laplace NMR data.
• The method successfully reconstructed complex multidimensional decay signals.
• The deep learning model avoided tedious parameter tuning, enhancing usability.

Conclusions:

• The blended physics-informed and data-driven approach offers a significant advancement in 2D Laplace NMR.
• This technique provides a fresh perspective on multidimensional Laplace NMR inversion.
• The method enhances the accuracy and accessibility of molecular dynamics studies using NMR.