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

Area of Science:

• Biophysical Chemistry
• Structural Biology
• Computational Chemistry

Background:

• Homonuclear scalar couplings, particularly one-bond ¹³C-¹³C couplings, complicate biomolecular ¹³C-detected NMR experiments.
• These couplings significantly reduce spectral sensitivity and resolution, hindering detailed structural analysis.
• Conventional methods like virtual decoupling require multiple spectra and linear combinations, increasing experimental complexity.

Purpose of the Study:

• To introduce a deep neural network (DNN) based method for virtual decoupling in NMR spectroscopy.
• To demonstrate the effectiveness of DNNs in improving spectral resolution and sensitivity.
• To reduce the experimental complexity by requiring only a single spectrum.

Main Methods:

• Development and application of deep neural networks for processing NMR data.
• Virtual decoupling of homonuclear ¹³C-¹³C couplings using a single spectrum.
• Application to various ¹³C-detected NMR experiments, including CON, protein side chains, and ¹³C_α-¹³CO spectra.

Main Results:

• The DNN method successfully achieved virtual decoupling, significantly boosting spectral resolution.
• Sensitivity was enhanced, and the minimum effective phase cycles were reduced by at least a factor of 2.
• Effective decoupling was achieved even in challenging cases where traditional methods failed.

Conclusions:

• Deep neural networks provide a flexible and powerful alternative for virtual decoupling in NMR spectroscopy.
• This approach simplifies experimental procedures and enhances data quality for biomolecular studies.
• The DNN method facilitates the use of simpler pulse sequences, broadening the applicability of NMR.