Deconvolution of NMR spectra : a deep learning-based approach

Schmid, Nicolas; Bruderer, Simon; Fischetti, Giulia; Paruzzo, Federico; Toscano, Giuseppe; Graf, Dominik; Fey, Michael; Henrici, Andreas; Grabner, Helmut; Wegner, Jan Dirk; Sigel, Roland K. O.; Heitmann, Björn; Wilhelm, Dirk

doi:10.21256/zhaw-27336

Please use this identifier to cite or link to this item: https://doi.org/10.21256/zhaw-27336

Publication type:	Conference poster
Type of review:	Peer review (abstract)
Title:	Deconvolution of NMR spectra : a deep learning-based approach
Authors:	Schmid, Nicolas Bruderer, Simon Fischetti, Giulia Paruzzo, Federico Toscano, Giuseppe Graf, Dominik Fey, Michael Henrici, Andreas Grabner, Helmut Wegner, Jan Dirk Sigel, Roland K. O. Heitmann, Björn Wilhelm, Dirk
et. al:	No
DOI:	10.21256/zhaw-27336
Proceedings:	EUROMAR 2022 Abstractbook
Editors of the parent work:	Prisner, Thomas
Page(s):	242
Conference details:	European Conference on Magnetic Resonance (EUROMAR), Utrecht, The Netherlands, 10-14 July 2022
Issue Date:	13-Jul-2022
Publisher / Ed. Institution:	ZHAW Zürcher Hochschule für Angewandte Wissenschaften
Language:	English
Subjects:	NMR spectroscopy; Deconvolution; Machine learning; Deep learning
Subject (DDC):	006: Special computer methods 530: Physics
Abstract:	We introduce a deep learning-based deconvolution approach for 1H NMR spectra, developed by leveraging concepts from the field of physics informed-learning, intelligent labeling, and tailored high dynamic range (HDR) spectral preprocessing. Since automation and faster workflows are major concerns in NMR spectroscopy, the algorithm handles uncorrected spectra without strict assumptions on phase and baseline correction as well as line shape. Due to the lack of high quality and consistently labeled experimental spectra in quantities needed to train modern deep learning models, we relied on synthetic spectra creation. Moreover, instead of training with synthetic spectra consisting of single lines, we created synthetic multiples that further supported a realistic deconvolution. We achieved super-human performance on corrected and uncorrected synthetic spectra. Finally, and most importantly, the results on synthetic data translate well to experimental spectra despite the covariate shift. Thus, this tool is a promising candidate for automated expert-level deconvolution of experimental HDR 1H NMR spectra.
URI:	https://euromar2022.org/wp-content/uploads/2022/07/Euromar-AbstractBook_2022-A4_22jul_new.pdf https://digitalcollection.zhaw.ch/handle/11475/27336
Fulltext version:	Published version
License (according to publishing contract):	Licence according to publishing contract
Departement:	School of Engineering
Organisational Unit:	Institute of Applied Mathematics and Physics (IAMP) Institute of Data Analysis and Process Design (IDP)
Published as part of the ZHAW project:	Maschinelles Lernen für NMR-Spektroskopie
Appears in collections:	Publikationen School of Engineering

Files in This Item:

File	Description	Size	Format
2022_Schmid-etal_Deconvolution-of-NMR-spectra_EUROMAR-Poster.pdf		3 MB	Adobe PDF	View/Open

Show full item record

Schmid, N., Bruderer, S., Fischetti, G., Paruzzo, F., Toscano, G., Graf, D., Fey, M., Henrici, A., Grabner, H., Wegner, J. D., Sigel, R. K. O., Heitmann, B., & Wilhelm, D. (2022). Deconvolution of NMR spectra : a deep learning-based approach [Conference poster]. In T. Prisner (Ed.), EUROMAR 2022 Abstractbook (p. 242). ZHAW Zürcher Hochschule für Angewandte Wissenschaften. https://doi.org/10.21256/zhaw-27336

Schmid, N. et al. (2022) ‘Deconvolution of NMR spectra : a deep learning-based approach’, in T. Prisner (ed.) EUROMAR 2022 Abstractbook. ZHAW Zürcher Hochschule für Angewandte Wissenschaften, p. 242. Available at: https://doi.org/10.21256/zhaw-27336.

N. Schmid et al., “Deconvolution of NMR spectra : a deep learning-based approach,” in EUROMAR 2022 Abstractbook, Jul. 2022, p. 242. doi: 10.21256/zhaw-27336.

SCHMID, Nicolas, Simon BRUDERER, Giulia FISCHETTI, Federico PARUZZO, Giuseppe TOSCANO, Dominik GRAF, Michael FEY, Andreas HENRICI, Helmut GRABNER, Jan Dirk WEGNER, Roland K. O. SIGEL, Björn HEITMANN und Dirk WILHELM, 2022. Deconvolution of NMR spectra : a deep learning-based approach. In: Thomas PRISNER (Hrsg.), EUROMAR 2022 Abstractbook [online]. Conference poster. ZHAW Zürcher Hochschule für Angewandte Wissenschaften. 13 Juli 2022. S. 242. Verfügbar unter: https://euromar2022.org/wp-content/uploads/2022/07/Euromar-AbstractBook_2022-A4_22jul_new.pdf

Schmid, Nicolas, Simon Bruderer, Giulia Fischetti, Federico Paruzzo, Giuseppe Toscano, Dominik Graf, Michael Fey, et al. 2022. “Deconvolution of NMR Spectra : A Deep Learning-Based Approach.” Conference poster. In EUROMAR 2022 Abstractbook, edited by Thomas Prisner, 242. ZHAW Zürcher Hochschule für Angewandte Wissenschaften. https://doi.org/10.21256/zhaw-27336.

Schmid, Nicolas, et al. “Deconvolution of NMR Spectra : A Deep Learning-Based Approach.” EUROMAR 2022 Abstractbook, edited by Thomas Prisner, ZHAW Zürcher Hochschule für Angewandte Wissenschaften, 2022, p. 242, https://doi.org/10.21256/zhaw-27336.