CABSEL/glycobert

Model Details

Model Name: GlycoBERT
Model Type: Transformer-based sequence classifier for glycan structure prediction
Architecture: BERT (Bidirectional Encoder Representations from Transformers)
Version: 1.0
Date: 6/25/2025

Model Description

GlycoBERT is a transformer-based deep learning model designed to predict glycan structures from tandem mass spectrometry (MS/MS) data. The model treats mass spectra as tokenized sequences ("MS sentences") and performs multi-class classification to assign spectra to one of 3,590 possible glycan structure classes.

Architecture Details

• Base Architecture: BertForSequenceClassification
• Parameters: 96 million
• Layers: 12 transformer layers
• Attention Heads: 12 per layer
• Hidden Size: 768 dimensions
• Max Sequence Length: 512 tokens
• Vocabulary Size: 10,010 tokens

Intended Use

Primary Use Cases

• Glycan structure prediction from MS/MS spectra
• High-throughput glycomics analysis
• Structural annotation of mass spectrometry data
• Research applications in glycobiology and glycoinformatics

Ethical Considerations

Responsible Use

• Model predictions should be validated experimentally
• Not intended for direct clinical decision-making without proper validation
• Users should understand the model's limitations and scope

Potential Risks

• Over-reliance on computational predictions without experimental validation
• Misinterpretation of confidence scores as absolute certainty
• Application to data significantly different from training distribution

Model Version

• GlycoBERT-F: Version trained on full dataset

Code and Data Availability

• Repository: GitHub (glycotrans)
• Training Data:
• Example Inference: Google Colab