JTF-Net: A joint time-frequency domain deep learning network for accurate reconstruction and reference-free assessment of NMR Spectra

Yanqin Lin,Yao Luo,Zhenhua Su,Wenhan Chen,Xiaoqi Shi,Jie Luo,Zhong Chen

DOI: https://doi.org/10.26434/chemrxiv-2024-rdv0l

2024-04-02

Abstract:High-quality nuclear magnetic resonance (NMR) spectra can be rapidly acquired by combining non-uniform sampling techniques (NUS) with reconstruction algorithms. However, current deep learning (DL) methods are constrained to single-domain reconstruction (either in the time domain or frequency domain), leading to limitations like peak loss and artifact peaks, ultimately failing to achieve optimal performance. More importantly, the lack of fully sampled spectra presents challenges in assessing the quality of reconstructed spectra, making it difficult, even impossible, to determine their quality. In this study, we propose a joint Time-Frequency domain deep learning network, referred to as JTF-Net. JTF-Net effectively combines time domain and frequency domain features, exhibiting better reconstruction performance on protein spectra across various dimensions compared to traditional algorithms and single-domain DL methods. As the most significant breakthrough of this study, JTF-Net established a confidence lattice, and based on this confidence lattice, we propose the first reference-free quality assessment metric in the field of NMR spectrum reconstruction, denoted as Lattice Confidence Rate (LCR). LCR is a pioneering achievement, capable of evaluating the quality of reconstructed NMR spectra without the fully sampled spectra, making it more suitable for practical applications.

Chemistry

What problem does this paper attempt to address?

### Problems Addressed by the Paper

The paper aims to address two main issues in Nuclear Magnetic Resonance (NMR) spectroscopy:

1. **Rapid Acquisition of High-Quality NMR Spectra**:
- Current methods combine Non-Uniform Sampling (NUS) techniques with reconstruction algorithms to quickly acquire high-quality NMR spectra. However, existing deep learning methods are limited to reconstruction in a single domain (time domain or frequency domain), leading to peak loss and artifact peaks, and thus failing to achieve optimal performance.

2. **Quality Assessment of Reconstructed Spectra**:
- In the absence of fully sampled spectra, assessing the quality of reconstructed spectra becomes very challenging. Commonly used evaluation metrics such as Pearson Correlation Coefficient (PCC) and Root Mean Square Deviation (RMSD) require fully sampled spectra as a reference, which are often difficult to obtain in practical applications.

To address the above issues, the paper proposes a Joint Time-Frequency Domain Deep Learning Network (JTF-Net), which effectively combines information from both the time and frequency domains, thereby demonstrating better performance in the reconstruction of multidimensional protein NMR spectra. Additionally, JTF-Net establishes a confidence grid and, based on this, proposes the first quality assessment metric without the need for reference spectra—the Local Confidence Rate (LCR), which can evaluate the quality of reconstructed spectra without fully sampled spectra.