DeepSAT: Learning Molecular Structures from Nuclear Magnetic Resonance Data

Abstract

The identification of molecular structure is essential for understanding chemical diversity and for developing drug leads from small molecules. Nevertheless, the structure elucidation of small molecules by Nuclear Magnetic Resonance (NMR) experiments is often a long and non-trivial process that relies on years of training. To achieve this process efficiently, several spectral databases have been established to retrieve reference NMR spectra. However, the number of reference NMR spectra available is limited and has mostly facilitated annotation of commercially available derivatives. Here, we introduce DeepSAT, a neural network-based structure annotation and scaffold prediction system that directly extracts the chemical features associated with molecular structures from their NMR spectra. Using only the 1H-13C HSQC spectrum, DeepSAT identifies related known compounds and thus efficiently assists in the identification of molecular structures. DeepSAT is expected to accelerate chemical and biomedical research by accelerating the identification of molecular structures.

Advancing Soil Health: Challenges and Opportunities in Integrating Digital Imaging, Spectroscopy, and Machine Learning for Bioindicator Analysis

“…106 Technological advancements, such as the application of CNN methods for the classification of 1 H NMR signals and the development of software such as DeepSAT, have been referenced for their improvements in NMR data interpretation. 107 However, there is a noticeable scarcity of literature on the application of machine learning for SOM analysis using NMR spectroscopy. SOM analysis using NMR spectroscopy is challenging, despite being a nondestructive technique, presenting challenges in sample preparation.…”

“…The evaluation of SOM stemming from hydrochar and biochar, through 1 H NMR, has been acknowledged for its role in identifying a range of organic compounds . Technological advancements, such as the application of CNN methods for the classification of 1 H NMR signals and the development of software such as DeepSAT, have been referenced for their improvements in NMR data interpretation . However, there is a noticeable scarcity of literature on the application of machine learning for SOM analysis using NMR spectroscopy.…”

HSQC Spectra Simulation and Matching for Molecular Identification

“…There have been some attempts to tackle these challenges. For instance, Cottrell et al introduced SMART, a technique that leverages a neural network trained on HSQC spectra, to enhance natural products deduplication by identifying clusters of similar compounds. , DeepSAT developed from the same group employs a neural network-based scaffold prediction system based on chemical features from HSQC spectra . Another 2D method, developed by Zanardi and Sarotti, is based on an artificial neural network (ANN-PRA) trained on pattern recognition descriptors of DFT-calculated and experimental HSQC data, which could correctly identify mischaracterized structures from the literature.…”

“…22,23 DeepSAT developed from the same group employs a neural network-based scaffold prediction system based on chemical features from HSQC spectra. 24 Another 2D method, developed by Zanardi and Sarotti, is based on an artificial neural network (ANN-PRA) 25 trained on pattern recognition descriptors of DFT-calculated and experimental HSQC data, which could correctly identify mischaracterized structures from the literature. These approaches, although novel and effective to a certain extent, do bring their own limitations, especially as the performance of a neural network can depend heavily on the diversity of structures included in the training set.…”

Bioassay-Guided and DeepSAT-Driven Precise Mining of Monoterpenoid Coumarin Derivatives with Antifeedant Effects from the Leaves of Ailanthus altissima

“…22,24−27 In this work, bioactivity-guided isolation was performed to discover the bioactive components. The chemical profile of the most active fraction followed was analyzed using the DeepSAT technology, 28 which is an effective tool to directly extract the chemical features associated with molecular structures from the HSQC spectrum, and the primary components were determined to be monoterpenoid coumarins. Further targeting and isolation led to 16 coumarin derivatives being targeted and isolated, including seven undescribed monoterpenoid coumarins and three undescribed monoterpenoid phenylpropanoids, opening up new possibilities for the development of botanical pesticides.…”