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

Gogulan Karunanithy,D. Flemming Hansen

doi:10.1007/s10858-021-00366-w

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.

Highlights

Artificial intelligence (AI) and deep learning (DL) have led to huge advances in many fields, including computer vision and natural language processing, and is a methodology that is embedded in many everyday technologies (Lecun et al 2015)
We have previously shown that an architecture based on reconstructing the points in the time domain using a modified long short-term memory (LSTM) architecture is able to reconstruct lowly-sampled 2D spectra (12.5%), with lower error than either the SMILE or hmsIST algorithms (Hansen 2019)
We present below free induction decays (FIDs)-Net: a versatile deep neural networks (DNNs) architecture that is able to reconstruct the time domain of a diverse set of 2D NMR spectra with low error, matching or exceeding the performance of leading non-DL algorithms

Summary

Introduction

Artificial intelligence (AI) and deep learning (DL) have led to huge advances in many fields, including computer vision and natural language processing, and is a methodology that is embedded in many everyday technologies (Lecun et al 2015). The network must first extract the relevant features from the input data to produce the required output This flexibility has made deep learning successful at performing tasks that are often intuitively straightforward for human beings to perform, but difficult to formalise into an algorithm (Goodfellow et al 2016). NMR researchers have sought to automate different aspects of NMR data analysis, speeding up the process and lowering the requirement for extensive training. Many of these methods currently struggle to match human performance or can only do so in cases of data with sharp well-resolved peaks and minimal noise. The current state of deep learning within NMR and potential future directions have been reviewed recently by Chen et al (Chen et al 2020)

Methods

Results

Conclusion