Predicting glycan structure from tandem mass spectrometry via deep learning

Area of Science:

• Biochemistry
• Computational Biology
• Glycoscience

Background:

• Glycans are crucial post-translational modifications impacting protein function in various biological processes.
• Structural annotation of glycans using tandem mass spectrometry (MS/MS) is a significant bottleneck in glycomics research.
• Current methods limit high-throughput glycomics and require specialized expertise.

Purpose of the Study:

• To develop an automated and rapid method for glycan structure prediction from MS/MS data.
• To overcome the limitations of manual annotation in glycomics.
• To enable high-throughput glycomics and democratize the field.

Main Methods:

• Development of CandyCrunch, a dilated residual neural network trained on 500,000 annotated MS/MS spectra.
• Creation of an open-access Python-based workflow for raw data conversion, prediction, and automated curation.
• Integration with the glycowork platform for enhanced usability and accessibility.

Main Results:

• CandyCrunch achieves high-top-1 accuracy (90.3%) in predicting glycan structures from raw liquid chromatography-MS/MS data within seconds.
• The workflow successfully recapitulates and extends expert annotations.
• Demonstrated utility in de novo annotation, diagnostic fragment identification, and high-throughput glycomics.

Conclusions:

• CandyCrunch significantly accelerates glycan structure elucidation from MS/MS data.
• The developed workflow democratizes structural glycomics, making it accessible to a broader research community.
• This tool facilitates a deeper understanding of the biological roles of glycans in health and disease.