NMRNet

NMRNet performs automated peak picking in multidimensional NMR spectra to enable accurate resonance assignment and facilitate structure calculation and analysis of macromolecular dynamics and interactions.

Key Features:

• Deep Learning Integration: Employs a convolutional neural network (CNN) to perform visual analysis of multidimensional NMR spectra for peak recognition.
• High Precision Performance: Validated on 31 manually annotated spectra with average precision (AP) scores of 0.9596 for backbone, 0.9058 for side-chain, and 0.8271 for NOESY spectra.
• Integration with Automated Assignment: Produces extracted peak lists that integrate with automated assignment routines such as FLYA, yielding correct assignment rates of 90.40%, 89.90%, and 90.20% for three benchmark proteins.
• Efficiency in Structure Calculation: Accurately and precisely picks peaks to accelerate structure calculation and analysis of macromolecular dynamics and interactions while maintaining accuracy comparable to human experts.

Scientific Applications:

• Structural biology: Automating peak picking in NMR spectroscopy to accelerate resonance assignment and enable efficient analysis of macromolecular dynamics and interactions.

Methodology:

Applies a convolutional neural network trained on manually annotated multidimensional NMR spectra (31 spectra) to learn patterns associated with signal peaks and distinguish them from noise.