Explore the GlycoDisease Candidate Relationship Resource
Search and reuse candidate protein–disease evidence extracted from a glycoscience-focused corpus,
with stable identifiers, verbatim source passages, explicit glycan-support status, and mapping uncertainty.
How We Extract Data
• Structured AI extraction with verbatim evidence
• Automated enrichment via UniProt, GlyGen & GlyTouCan APIs
• Candidate extraction for protein–disease evidence records
• Glycan structure mapping with detailed annotations
Database Access
• Evidence browser: Inspect passages, anchors, and PMCID provenance
• Entity browsers: Explore proteins, diseases, and glycan annotations
• Build datasets with transparent provenance and uncertainty
• Generate hypotheses for glycobiology studies
• Optionally rank sequences as an auxiliary research analysis
Disease Categories
Annotation Field Coverage
Relationship Types
Auxiliary Ranking Signal
Protein List
Name
UniProt ID
Gene
Glyco Sites
Glycans
Status
Action
Disease-Protein Associations
Glycan Structure Library
Protein-Disease Relationships
Protein
Relationship
Disease
Mapping
Glycan evidence
Verbatim evidence
Candidate Relationship Network
Scope:
Layout:
Protein (circle)
Disease (diamond)
Overview shows the first 200 retained evidence records; choose Full network for all records or click Network from a disease/protein row for a focused view.
Color intensity indicates connection degree (darker = more connections). Drag to rearrange nodes.
Sequence-Based Association Ranking
Enter a protein sequence of 10–1,022 residues to obtain exploratory category-specific
literature-derived candidate-P-versus-unlabeled ranking scores from the frozen sequence-only model.
Ranking Results
Enter a protein sequence and click "Rank Sequence" to see results.
Upload FASTA File
Upload a FASTA file containing multiple protein sequences (max 100 sequences).
Validated length: 10–1,022 residues Interpretation: Not a calibrated disease probability
Download Data
Download the complete GlycoDisease database for your research.
Data reuse terms are pending author confirmation. Downloaded records are
glycoscience-corpus-derived candidates and have not been independently adjudicated.
Loading evidence-quality scope...
Protein-Disease Relationships
Versioned candidate evidence records with source PMCIDs, paragraph anchors, and verbatim passages.
Access candidate data programmatically. The complete machine-readable contract is available at
/openapi.json.
GET
/api/proteins
List all proteins
GET
/api/protein/{id}
Get protein details
GET
/api/relationships
List relationships (paginated)
GET
/api/diseases
List diseases with stats
POST
/api/rank
Rank sequence against candidate-P and unlabeled cohorts
GET
/api/search?q={query}
Search database
GlycoDisease Candidate Relationship Resource
A resource of glycoscience-corpus-derived candidate protein–disease evidence with
explicit glycan-support strata, external annotations, and auxiliary sequence-based ranking.
Candidate rows have not been validated against an independently adjudicated gold standard.
Key Features
Versioned candidate evidence records with auditable source passages
Mapped and unresolved protein records with explicit normalization status
Explicit glycan-involvement support is reported separately from database inclusion
Auxiliary candidate-P-versus-unlabeled sequence ranking using ESM-2 embeddings
Mandatory human adjudication is not a release prerequisite. Authors may verify records during use. Report an evidence ID, source passage and discrepancy to zhangyang@fudan.edu.cn; confirmed issues inform pipeline fixes and versioned data corrections.
Fixed-cohort exploratory model comparisons: seven methods and four sensitivity populations; no independent external validation or automatic replacement of the deployed model.
Citation
GlycoDisease: an evidence-grounded resource for glycoscience-corpus-derived candidate protein-disease relationships.
Cong Wang, Yan Jiang, XiaoJun Zhao, Chuan-yu Sun, Shouling Mi, and Yang Zhang
Manuscript, 2026
DOI: pending
Contact
For questions, suggestions, or collaborations:
zhangyang@fudan.edu.cn
License
Platform source code uses the MIT License. Data, model weights, and literature excerpts have separate reuse terms; MIT does not grant rights to those materials.