A pilot study for fragment identification using 2D NMR and deep learning

Abstract

This paper presents a proof of concept of a method to identify substructures in 2D NMR spectra of mixtures using a bespoke image‐based convolutional neural network application. This is done using HSQC and HMBC spectra separately and in combination. The application can reliably detect substructures in pure compounds, using a simple network. Results indicate that it can work for mixtures when trained on pure compounds only. HMBC data and the combination of HMBC and HSQC show better results than HSQC alone in this pilot study.

Direct deduction of chemical class from NMR spectra

“…HMBC alone has better results than HSQC alone, and both together are similar to HSQC. For single networks, this is in line with the results reported in [1] for substructure classification. It should be noted that, as explained in Section 3.5, the results could most likely be optimised, and this represents a proof of concept.…”

“…However, a full elucidation may not always be possible or even necessary since some properties might be achievable directly from the spectra. If this is the case, a prioritisation of substances to be closely investigated for compound assignment can be done in the early stages of a study.A previous example for this idea was demonstrated in [1], where the authors showed that the existence of certain substructures can be concluded from profiles in the spectra. Another potentially useful application is chemical classification.…”

“…A previous example for this idea was demonstrated in [1], where the authors showed that the existence of certain substructures can be concluded from profiles in the spectra. Another potentially useful application is chemical classification.…”

Deep Learning-Based Method for Compound Identification in NMR Spectra of Mixtures

“…Third, the deep learning methods have high expressive power and model capacity because of the depth efficiency, which can take full advantage of big data [55]. Due to these advantages, deep learningbased methods have achieved a state-of-the-art performance in numerous related fields of NMR spectroscopy [56,57], ranging from spectral reconstruction [58][59][60], denoising [61], peak picking [62,63], chemical shift prediction [64][65][66][67][68] and molecular recognition (the SMART method proposed by Zhang et al in 2017) [69] to molecule identification [70][71][72]. It has shown unprecedented capabilities in solving difficult problems in NMR spectroscopy.…”

SMART‐Miner: A convolutional neural network‐based metabolite identification from 1H‐13C HSQC spectra

“…In NMR metabolomics, CNNs have been used to identify molecular structures, [ 15 ] peak deconvolution, [ 16 ] and fragment identification. [ 17 ] The neural network learns the features of the input data based on the back propagation algorithm, which avoids problems arising from hand‐designed or rule‐based approaches. Therefore, we focused on the development of a CNN algorithm as an alternative automation tool for metabolite identification to be used in metabolomic research.…”