Noise peak filtering in multi-dimensional NMR spectra using convolutional neural networks

Area of Science:

• Structural Biology
• Computational Chemistry
• Biophysics

Background:

• Multi-dimensional Nuclear Magnetic Resonance (NMR) spectroscopy is crucial for molecular structure determination and signal assignment.
• Automated NMR analysis tools exist but struggle with pervasive noise peaks, hindering accuracy.
• Eliminating noise is essential for reliable NMR data interpretation.

Purpose of the Study:

• To develop an automated method for effectively removing noise peaks from multi-dimensional NMR spectra.
• To enhance the accuracy and efficiency of automated NMR signal assignment and structure analysis.

Main Methods:

• Development of a novel noise elimination technique utilizing convolutional neural networks (CNNs).
• Implementation of the CNN-based method into a user-friendly program package named Filt_Robot.
• Validation of Filt_Robot's performance on 2D and 3D NMR spectra.

Main Results:

• Filt_Robot demonstrated high filtering accuracy, achieving 90-95% effectiveness in noise peak removal.
• The number of non-noise peaks identified by Filt_Robot closely matched manually curated peak lists.
• The developed method significantly enhances the reliability of automated NMR spectral analysis.

Conclusions:

• Convolutional neural networks provide a powerful approach for noise reduction in NMR spectroscopy.
• Filt_Robot offers a robust solution for improving automated NMR data processing.
• This method facilitates more accurate and efficient molecular structure analysis using NMR data.