FID-Net: A versatile deep neural network architecture for NMR spectral reconstruction and virtual decoupling

Abstract

In recent years, the transformative potential of deep neural networks (DNNs) for analysing and interpreting NMR data has clearly been recognised. However, most applications of DNNs in NMR to date either struggle to outperform existing methodologies or are limited in scope to a narrow range of data that closely resemble the data that the network was trained on. These limitations have prevented a widescale uptake of DNNs in NMR. Addressing this, we introduce FID-Net, a deep neural network architecture inspired by WaveNet, for performing analyses on time domain NMR data. We first demonstrate the effectiveness of this architecture in reconstructing non-uniformly sampled (NUS) biomolecular NMR spectra. It is shown that a single network is able to reconstruct a diverse range of 2D NUS spectra that have been obtained with arbitrary sampling schedules, with a range of sweep widths, and a variety of other acquisition parameters. The performance of the trained FID-Net in this case exceeds or matches existing methods currently used for the reconstruction of NUS NMR spectra. Secondly, we present a network based on the FID-Net architecture that can efficiently virtually decouple 13Cα-13Cβ couplings in HNCA protein NMR spectra in a single shot analysis, while at the same time leaving glycine residues unmodulated. The ability for these DNNs to work effectively in a wide range of scenarios, without retraining, paves the way for their widespread usage in analysing NMR data.

13C Direct Detected NMR for Challenging Systems

“…Initially proposed for indirect dimensions of triple resonance experiments 7 and implemented for 13 C detected experiments, 8,9 this approach has been recently revived by the incorporation of AI methods. 10 The other possibility, provided the chemical shifts of the two nuclear spins involved are sufficiently different to allow for their selective irradiation, consists of band-selective homonuclear decoupling in which the acquisition time is shared between acquisition and decoupling mode in alternating time intervals (Figure 1). 11−13 This approach requires an additional radio frequency channel; decoupling sidebands are observed depending on the frequency of acquisition and decoupling periods.…”

Virtual Homonuclear Decoupling in Direct Detection Nuclear Magnetic Resonance Experiments Using Deep Neural Networks

“…Deep Neural Network Architecture. We employed the FID-Net architecture 26 for our DNNs and here only provide a brief outline of its features. When a DNN is used to model time-domain NMR data (FIDs; free induction decays), a key challenge to overcome is that information about the resonating nuclei is not localized to a specific part of an FID but rather contained over its entire length.…”

“…Recently, we demonstrated that a deep neural network (DNN) based on dilated convolutional layers, FID-Net, could be trained to perform a variety of transformations on time domain NMR data including reconstructing nonuniformly sampled (NUS) spectra. 26 Building on this idea, we show here that FID-Net can be trained to decouple directly detected spectra using a single spectrum (Figure 1). We apply this methodology to 13 C-detected protein CON spectra and 13 C− 13 C spectra of protein side chains to decouple spectra on the basis of the in-phase component of the experiment alone.…”

“…Recently, we demonstrated that a deep neural network (DNN) based on dilated convolutional layers, FID-Net, could be trained to perform a variety of transformations on time domain NMR data including reconstructing nonuniformly sampled (NUS) spectra . Building on this idea, we show here that FID-Net can be trained to decouple directly detected spectra using a single spectrum (Figure ).…”

SpecDB: A Relational Database for Archiving Biomolecular NMR Spectra Data

“…NMR spectroscopy has a rich history of developing and applying machine learning at all stages in the experimental pipeline 7,8 . High impact examples include predicting protein torsion angles 9 , chemical shift prediction 10 , NMR spectral peak picking 11,12 , and reconstruction of non-uniformly sampled free induction decays (FIDs) [13][14][15] . Additionally, there have been efforts to organize NMR data into datasets suitable for machine learning, like the RefDB dataset with re-referenced chemical shifts 16 .…”