Methodology

Knowledge Graph Prediction

ThTxGNN uses the TxGNN (Therapeutic Target Graph Neural Network) knowledge graph to predict potential drug-disease relationships.

TxGNN Knowledge Graph

The TxGNN knowledge graph contains:

  • 17,081 diseases mapped to DOID (Disease Ontology)
  • 7,958 drugs mapped to DrugBank
  • 80,127 drug-disease relationships including indications and contraindications

Prediction Process

  1. Drug Mapping: Thai FDA-approved drugs are mapped to DrugBank IDs using:
    • Generic name matching
    • Brand name normalization
    • Active ingredient extraction
  2. Disease Mapping: Thai indications are mapped to English disease terms using:
    • Direct translation
    • Medical terminology lookup
    • DISEASE_DICT mapping table
  3. KG Query: For each mapped drug, query the knowledge graph for:
    • Known indications (existing approvals)
    • Potential new indications (repurposing candidates)
  4. Evidence Collection: For promising candidates, collect supporting evidence from:
    • PubMed literature
    • ClinicalTrials.gov
    • Thai Clinical Trial Registry (TCTR)

Scoring

Predictions include confidence scores based on:

  • Graph structure (node connectivity)
  • Relationship strength in knowledge graph
  • Supporting evidence count

Limitations

  • Predictions are computational hypotheses requiring clinical validation
  • Knowledge graph data may not include recent drug approvals
  • Thai-specific drug formulations may not have DrugBank mappings
  • Disease terminology translation may lose clinical nuance

Disclaimer

This methodology is for research purposes only. Drug repurposing candidates identified through this system require rigorous clinical validation before any therapeutic application.