User Guide
Learn how to effectively use MyTxGNN for drug repurposing research.
Getting Started
What Can You Do with MyTxGNN?
- Browse Drug Predictions: View predicted new indications for 508 NPRA-approved drugs
- Access FHIR API: Integrate predictions into your clinical systems
- Download Data: Get prediction data in CSV format for analysis
- Monitor News: Track health news related to drug repurposing in Malaysia
Understanding Predictions
Two Types of Predictions
MyTxGNN provides two complementary prediction methods:
Knowledge Graph (KG) Predictions
- Based on existing relationships in biomedical knowledge graphs
- Identifies drugs with similar indication patterns
- 41,560 predictions for 508 drugs
Deep Learning (DL) Predictions
- Neural network confidence scores
- Higher scores indicate stronger predicted associations
- 176,021 high-confidence predictions (score ≥ 0.7)
Interpreting Results
| Score Range | Interpretation | Recommended Action |
|---|---|---|
| ≥ 0.9 | Very high confidence | Priority for investigation |
| 0.7 - 0.9 | High confidence | Worth investigating |
| 0.5 - 0.7 | Moderate confidence | Consider with caution |
| < 0.5 | Low confidence | Exploratory only |
Using the FHIR API
Capability Statement
Check API capabilities:
GET /fhir/metadata
MedicationKnowledge
Get drug information:
GET /fhir/MedicationKnowledge/{drugbank_id}
Example:
GET /fhir/MedicationKnowledge/db00860
ClinicalUseDefinition
Get drug-indication predictions:
GET /fhir/ClinicalUseDefinition/{drug}-{indication}
Example:
GET /fhir/ClinicalUseDefinition/db00860-allergic-rhinitis
Research Workflow
Recommended Steps
- Identify Target
- Start with a drug of interest
- Or start with a disease you want to find treatments for
- Review Predictions
- Check KG predictions for the drug/disease
- Review DL confidence scores
- Gather Evidence
- Use our evidence collectors (ClinicalTrials.gov, PubMed)
- Check for existing clinical trials
- Assess Feasibility
- Consider drug availability in Malaysia (NPRA status)
- Review safety profile
- Plan Next Steps
- Design validation studies
- Consult with clinical experts
Tips for Researchers
Best Practices
- Cross-reference predictions with existing literature
- Check multiple sources for clinical trial data
- Consider mechanism of action when evaluating predictions
- Note prediction confidence scores when prioritizing candidates
Common Pitfalls to Avoid
- Don’t treat predictions as validated findings
- Don’t ignore safety considerations
- Don’t rely solely on high scores - context matters
- Don’t skip regulatory considerations
Frequently Asked Questions
Q: How accurate are the predictions?
A: TxGNN has been validated against known drug-disease relationships with high accuracy. However, predictions should be validated through clinical evidence before application.
Q: Can I use this data for commercial purposes?
A: The data is provided for research purposes. Commercial use may be subject to individual data source licenses (DrugBank, etc.).
Q: How often is the data updated?
A: NPRA data is updated weekly. Predictions are regenerated periodically as the underlying models improve.
Q: Why are some drugs not included?
A: Only drugs that could be mapped to DrugBank IDs are included. Some local or combination products may not have DrugBank entries.
Getting Help
Resources
- GitHub Issues: Report bugs or request features
- Methodology Page: Understand our prediction approach
- Data Sources: Learn about our data origins
Contact
For research collaboration inquiries, please open a GitHub issue with the “collaboration” label.
MyTxGNN is designed for research purposes. Always validate predictions through proper clinical trials and consult healthcare professionals for medical decisions.