DeepNMR: Unleashing the full potential of NMR spectroscopy with artificial intelligence and deep learning
Lead Research Organisation:
UNIVERSITY COLLEGE LONDON
Department Name: Structural Molecular Biology
Abstract
Nuclear Magnetic Resonance (NMR) spectroscopy is ubiquitous in material science, chemistry, structural biology, and clinical
diagnosis. In chemical synthesis, the identification and characterisation of compounds hinge on NMR and in bioscience NMR provides
unprecedented insight into functional motions and on non-covalent interactions with atomic resolution. However, the analysis of
NMR spectra, in particular biomolecular NMR spectra, still largely depend on interpretations by specialists with years of training. Even
more so, the development of NMR methods to allow for new applications relies on specialists with decades of training and excellent
intuition. These constraints have meant that the full potential of NMR as a tool in chemistry, biochemistry, and medicine, is far from
being reached. The proposed research will address this inhibitory constrain of biomolecular NMR by fully integrating artificial
intelligence (AI) with the analysis of NMR data and with the development of new NMR methods. Using supervised deep learning,
deep neural networks (DNNs) will be developed to analyse complex biomolecular NMR spectra. Analysis with DNNs is robust and
once the DNN is trained, it does not require an optimisation of processing parameters. The DNNs can therefore easily be integrated
into automated data-processing pipelines. Reinforcement deep learning will be employed to design intelligent machines that provide
the next generations of NMR methods. With these tools, the scientist can simply request the intelligent machine to derive a method
and an analysis tool to characterise a specific set of parameters or functions of the macromolecule in question. Being able to fully
integrate AI with NMR, and concomitantly develop NMR and AI as one tool, is high-risk, but once successful will unleash the immense
potential of current and future NMR hardware to provide unprecedented insights into a broad range of molecules, in material science,
in biochemistry, and in medicine.
diagnosis. In chemical synthesis, the identification and characterisation of compounds hinge on NMR and in bioscience NMR provides
unprecedented insight into functional motions and on non-covalent interactions with atomic resolution. However, the analysis of
NMR spectra, in particular biomolecular NMR spectra, still largely depend on interpretations by specialists with years of training. Even
more so, the development of NMR methods to allow for new applications relies on specialists with decades of training and excellent
intuition. These constraints have meant that the full potential of NMR as a tool in chemistry, biochemistry, and medicine, is far from
being reached. The proposed research will address this inhibitory constrain of biomolecular NMR by fully integrating artificial
intelligence (AI) with the analysis of NMR data and with the development of new NMR methods. Using supervised deep learning,
deep neural networks (DNNs) will be developed to analyse complex biomolecular NMR spectra. Analysis with DNNs is robust and
once the DNN is trained, it does not require an optimisation of processing parameters. The DNNs can therefore easily be integrated
into automated data-processing pipelines. Reinforcement deep learning will be employed to design intelligent machines that provide
the next generations of NMR methods. With these tools, the scientist can simply request the intelligent machine to derive a method
and an analysis tool to characterise a specific set of parameters or functions of the macromolecule in question. Being able to fully
integrate AI with NMR, and concomitantly develop NMR and AI as one tool, is high-risk, but once successful will unleash the immense
potential of current and future NMR hardware to provide unprecedented insights into a broad range of molecules, in material science,
in biochemistry, and in medicine.
People |
ORCID iD |
| Flemming Hansen (Principal Investigator) |
Publications
Shukla VK
(2023)
Biomolecular NMR spectroscopy in the era of artificial intelligence.
in Structure (London, England : 1993)
Shukla VK
(2023)
Intrinsic structural dynamics dictate enzymatic activity and inhibition.
in Proceedings of the National Academy of Sciences of the United States of America
Bolik-Coulon N
(2023)
Less is more: A simple methyl-TROSY based pulse scheme offers improved sensitivity in applications to high molecular weight complexes.
in Journal of magnetic resonance (San Diego, Calif. : 1997)
Karunanithy G
(2024)
Solution-state methyl NMR spectroscopy of large non-deuterated proteins enabled by deep neural networks.
in Nature communications
De D
(2024)
Mapping the FF domain folding pathway via structures of transiently populated folding intermediates
in Proceedings of the National Academy of Sciences
Kakita V
(2024)
Deep Learning Assisted Proton Pure Shift NMR Spectroscopy
| Title | A combined NMR and Deep Neural Network approach for enhancing the spectral resolution of aromatic side chains in proteins |
| Description | Nuclear magnetic resonance (NMR) spectroscopy has become an important technique in structural biology for characterising the structure, dynamics and interactions of macromolecules. While a plethora of NMR methods are now available to inform on backbone and methyl-bearing side-chains of proteins, a characterisation of aromatic side chains is more challenging and often requires specific labelling or 13C-detection. Here we present a deep neural network (DNN) named FID-Net-2, which transforms NMR spectra recorded on simple uniformly 13C labelled samples to yield high-quality 1H-13C correlation spectra of the aromatic side chains. Key to the success of the DNN is the design of a complementary set of NMR experiments that produce spectra with unique features to aid the DNN produce high-resolution aromatic 1H-13C correlation spectra with accurate intensities. The reconstructed spectra can be used for quantitative purposes as FID-Net-2 predicts uncertainties in the resulting spectra. We have validated the new methodology experimentally on protein samples ranging from 7 to 40 kDa in size. We demonstrate that the method can accurately reconstruct high resolution two-dimensional aromatic 1H-13C correlation maps, high resolution three-dimensional aromatic-methyl NOESY spectra to facilitate aromatic 1H-13C assignments, and that the intensities of peaks from the reconstructed aromatic 1H-13C correlation maps can be used to quantitatively characterise the kinetics of protein folding. More generally, we believe that this strategy of devising new NMR experiments specifically for analysis using customised DNNs represents a substantial advance that will have a major impact on the study of molecules using NMR in the years to come. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | Charactering aromatic side chains in proteins by NMR and deep learning |
| URL | https://zenodo.org/doi/10.5281/zenodo.13745297 |
| Title | Solution-State Methyl NMR Spectroscopy of Large Non-Deuterated Proteins Enabled by Deep Neural Networks |
| Description | Training data for Deep Neural Networks developed in the manuscript "Solution-State Methyl NMR Spectroscopy of Large Non-Deuterated Proteins Enabled by Deep Neural Networks" Experimental cross-validation data: 2D spectra of HDAC8, MSG, and a7a7 proteasome 3D NOESY spectra of MSG Please see GitHub (https://github.com/gogulan-k/FID-Net/tree/main/FID-Net_Non_deuterated_proteins_13C-1H) for additional scripts and details. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | A deep learning method to transform complex methyl NMR spectra of large proteins |
| URL | https://zenodo.org/doi/10.5281/zenodo.10022405 |
| Title | Solution-State Methyl NMR Spectroscopy of Large Non-Deuterated Proteins Enabled by Deep Neural Networks |
| Description | Training data for Deep Neural Networks developed in the manuscript "Solution-State Methyl NMR Spectroscopy of Large Non-Deuterated Proteins Enabled by Deep Neural Networks" Experimental cross-validation data: 2D spectra of HDAC8, MSG, and a7a7 proteasome 3D NOESY spectra of MSG Please see GitHub (https://github.com/gogulan-k/FID-Net/tree/main/FID-Net_Non_deuterated_proteins_13C-1H) for additional scripts and details. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | A method to characterise the methyl side chains of large proteins |
| URL | https://zenodo.org/doi/10.5281/zenodo.10022404 |
| Description | Collaboration with Bruker Biospin |
| Organisation | Bruker Corporation |
| Department | Bruker (United Kingdom) |
| Country | United Kingdom |
| Sector | Private |
| PI Contribution | This collaboration is about implementation of our recently developed Deep Learning tools for NMR spectroscopy into the standard software used by a majority of NMR spectroscopist worldwide (TopSpin). My group provides the trained deep neural networks and knowledge about the execution of these. |
| Collaborator Contribution | Sill being negotiated. |
| Impact | No output yet. |
| Start Year | 2023 |
| Description | Collaboration with Bruker Biospin |
| Organisation | Bruker Corporation |
| Department | Bruker BioSpin |
| Country | Germany |
| Sector | Private |
| PI Contribution | This collaboration is about implementation of our recently developed Deep Learning tools for NMR spectroscopy into the standard software used by a majority of NMR spectroscopist worldwide (TopSpin). My group provides the trained deep neural networks and knowledge about the execution of these. |
| Collaborator Contribution | Sill being negotiated. |
| Impact | No output yet. |
| Start Year | 2023 |
| Description | Collaboration with Francis Crick NMR centre. |
| Organisation | Francis Crick Institute |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | We have been developing deep learning networks and trained these. |
| Collaborator Contribution | The Francis Crick NMR centre has provided experimental NMR spectra, NMR allocations, and computational resources. |
| Impact | Deep neural network, published here: https://doi.org/10.26434/chemrxiv.13295888.v2 |
| Start Year | 2020 |
| Description | Collaboration with Prof. Rosenzweig (Weizmann Institute of Science) |
| Organisation | Weizmann Institute of Science |
| Country | Israel |
| Sector | Academic/University |
| PI Contribution | This is a collaboration with Prof. Rina Rosenzweig at the Weizmann Institute of Science. We are collaborating on characterising aromatic side chains in large proteins using a combination of deep learning (Hansen), new isotope labelling (Rosenzweig), new NMR methods (Hansen) and functional cellular assays (Rosenzweig). Funding obtained from Weizmann, UCL and British Council. |
| Collaborator Contribution | Development of deep learning technique and NMR pulse sequences |
| Impact | Funding from UCL: Global Engagement grant Funding from WIS: Mini symposium for both groups Funding from British Council: |
| Start Year | 2024 |
| Description | NMRBox |
| Organisation | University of Connecticut |
| Department | Health Center (Uconn Health) |
| Country | United States |
| Sector | Academic/University |
| PI Contribution | The deep neural networks developed during to process biomolecular NMR spectra have been made available to the NMRBox, a global resource for biomolecular NMR software. |
| Collaborator Contribution | The partner, NMRBox (nmrbox.org) is currently hosting our software and tools on their large computational resource, so researcher can easily and freely directly use our developed tools. |
| Impact | Our tools are easily available to a large group science, internationally. |
| Start Year | 2020 |
| Title | FID-Net: A Versatile Deep Neural Network Architecture for NMR Spectral Reconstruction and Virtual Decoupling |
| Description | Deep Neural Network for the reconstruction and homonuclear decoupling of NMR spectra |
| Type Of Technology | Software |
| Year Produced | 2024 |
| Open Source License? | Yes |
| Impact | Originally released in a simple version in 2020 - continuously updated. Used throughout the NMR community and has inspired a new generation of DNN architectures to analyse and transform complex NMR spectra |