Combined NMR and Deep Neural Network approach enhances resolution of protein aromatic side chains

Scientists have developed a deep learning tool that significantly improves the analysis of protein aromatic side chains using Nuclear Magnetic Resonance (NMR) spectroscopy. This breakthrough, led by D. Flemming Hansen and colleagues, addresses current challenges in obtaining high-resolution NMR spectra for aromatic residues, which are critical for understanding protein structure and dynamics.
 
Traditional NMR experiments primarily focus on the protein backbone and aliphatic side chains, such as methyl groups, while aromatic side chains remain difficult to resolve due to spectral complexity and overlap. To solve this, the team developed a deep neural network (DNN) called FID-Net-2. By integrating specialised NMR pulse sequences with advanced deep learning, FID-Net-2 enhances the quality of two-dimensional ¹H-¹³C correlation spectra for aromatic side chains.
 
A key innovation is the design of complementary NMR experiments that generate unique spectral features, enabling FID-Net-2 to reconstruct high-resolution spectra with accurate intensity measurements. Notably, the model also predicts uncertainties, on a point-by-point basis, allowing for reliable quantitative analysis.
 
The method was validated on protein samples ranging from 7 to 40 kDa, demonstrating its effectiveness in reconstructing high-resolution aromatic ¹H-¹³C correlation maps and three-dimensional aromatic-methyl NOESY spectra. These enhanced spectra facilitate the assignment of aromatic signals and provide valuable insights into kinetic processes. By analysing peak intensities and their uncertainties, the method quantifies dynamic processes such as folding rates with high confidence.
 
FID-Net-2’s ability to estimate uncertainties is crucial for distinguishing valid cross-peaks in NOESY spectra and ensuring accurate kinetic measurements. Moreover, the flexibility of the deep learning model means it could likely be retrained for other types of NMR data, broadening its potential applications.
 
This AI-driven approach represents a significant step forward in structural biology, enabling researchers to gain clearer insights into protein structure and dynamics. By overcoming the challenges of aromatic side chain analysis, FID-Net-2 has the potential to enhance NMR-based investigations of complex biomolecules and biological processes. 

Source: Read the full paper in Science Advances: ‘A combined NMR and deep neural network approach for enhancing the spectral resolution of aromatic side chains in proteins’ DOI: 10.1126/sciadv.adr215

Links:

Hansen Lab 

Bluesky @dflemminghansen.bsky.social