CandyCrunch

CandyCrunch predicts glycan structures directly from liquid chromatography-tandem mass spectrometry (LC-MS/MS) data to enable structural glycomics analyses.

Key Features:

• Dilated residual neural network: Uses a dilated residual neural network trained on 300,000 annotated MS/MS spectra.
• Predictive performance: Achieves a top-1 accuracy of 87.7% on glycan structure prediction from MS/MS data.
• Raw data processing: Converts raw LC-MS/MS data into glycan structure predictions in seconds.
• Automated curation and fragment annotation: Performs automated curation and fragment annotation that replicate and extend expert annotations.
• Workflow implementation: Implemented as a Python-based workflow for processing MS/MS spectra and generating predictions.
• Platform integration: Integrates with the glycowork platform for downstream glycomics analyses.

Scientific Applications:

• De novo glycan annotation: Enables de novo structural annotation of glycans directly from MS/MS spectra.
• Diagnostic fragment identification: Identifies diagnostic fragments for glycan structural elucidation from LC-MS/MS data.
• High-throughput glycomics: Facilitates large-scale glycomics studies by rapid prediction of glycan structures from MS/MS datasets.
• Biological role elucidation: Supports studies linking glycan structures to protein function and disease-related glycosylation changes.

Methodology:

Modeling uses a dilated residual neural network trained on 300,000 annotated MS/MS spectra; the pipeline performs automated curation and fragment annotation and converts raw LC-MS/MS data into glycan structure predictions within a Python-based workflow.