DEEP Phaser: A Deep Learning Tandem Vision Transformer for Fully Automated NMR Phase Correction

Area of Science:

• Analytical Chemistry
• Biochemistry
• Computational Chemistry

Background:

• Phase correction is a critical step in Nuclear Magnetic Resonance (NMR) data processing.
• Current automated methods often necessitate manual intervention by experts for optimal results.

Purpose of the Study:

• To develop an advanced, automated phase correction algorithm for NMR spectra.
• To achieve high phasing accuracy across diverse experimental NMR datasets without manual adjustments.

Main Methods:

• Implementation of a deep learning algorithm utilizing a tandem vision transformer artificial neural network.
• Training the model on a comprehensive dataset of synthetic solution-NMR spectra.
• The algorithm determines zeroth- and first-order phase correction based on the entire spectrum.

Main Results:

• DEEP Phaser demonstrates very high phasing accuracy on a wide array of experimental solution 1D NMR spectra.
• The method is effective for various sample types, including small molecules, complex mixtures, and biomacromolecules.
• No further manual adjustments were required for the tested spectra.

Conclusions:

• DEEP Phaser offers a robust and automated solution for NMR spectra phase correction.
• The deep learning approach significantly improves efficiency and accuracy in NMR data processing.
• The software is publicly available as free software and a web server for broad accessibility.