ThTxGNN: AI-Powered Drug Repurposing for Thailand NHI Drugs

Executive Summary

ThTxGNN is an open-source drug repurposing platform that leverages Harvard’s TxGNN deep learning model to predict potential new indications for Thailand National Health Insurance (NHI) approved drugs. Unlike prediction-only tools, ThTxGNN provides a complete evidence validation pipeline, integrating clinical trials, literature, and safety data to generate actionable reports.

Key Differentiators

Feature Description
Evidence-Based Predictions Every prediction is validated against ClinicalTrials.gov, PubMed, and other authoritative sources
Five-Level Evidence Classification L1 (multiple Phase 3 RCTs) to L5 (AI prediction only)
SMART on FHIR Integration Ready-to-use app for EHR integration
FHIR R4 API Static FHIR resources for external system integration
Comprehensive Safety Data 222,391 DDI records, 8,359 DDSI records

Data Overview

Metric Count
Drug Reports 191
Drug Repurposing Candidates 142,328
Predicted Indications 1,268
Drug-Drug Interactions (DDI) 222,391
Drug-Disease Contraindications (DDSI) 8,359

Evidence Distribution

Level Definition Drug Count
L1 Multiple Phase 3 RCTs 6
L2 Single RCT or Phase 2 trials 12
L3 Observational studies 16
L4 Preclinical / mechanistic 19
L5 AI prediction only 138

Technology Stack

Prediction Model

  • TxGNN: Published in Nature Medicine (2023) by Harvard Zitnik Lab
  • Method: Knowledge graph + deep learning for drug-disease relationship prediction
  • Source: TxGNN Paper

Evidence Sources

Source Data Type
ClinicalTrials.gov Clinical trial data (NCT IDs)
PubMed Biomedical literature (PMIDs)
DrugBank Drug properties and interactions
TFDA Thailand FDA approval status
DDInter 2.0 Drug-drug interactions
Disease Ontology Disease classification (DOIDs)

Technical Implementation

  • Frontend: Static site (Jekyll) + JavaScript
  • FHIR Version: R4 (4.0.1)
  • Data Format: JSON, CSV
  • Hosting: GitHub Pages
  • Open Source: GitHub Repository

SMART on FHIR Integration

ThTxGNN provides a fully functional SMART on FHIR application that can be integrated into any FHIR R4-compliant EHR system.

Capabilities

  1. Patient Medication Retrieval: Reads MedicationRequest/MedicationStatement from EHR
  2. Drug Mapping: Maps RxNorm codes to ThTxGNN database via RxNorm API
  3. Repurposing Lookup: Displays predicted new indications for each medication
  4. Evidence Display: Shows evidence level and supporting references

Technical Specifications

Item Value
FHIR Version R4
OAuth OAuth 2.0 with PKCE
Scopes launch, patient/MedicationRequest.read, patient/MedicationStatement.read, openid, fhirUser
Launch URI https://thtxgnn.yao.care/smart/launch.html
Standalone URI https://thtxgnn.yao.care/smart/standalone.html

FHIR API Endpoints

Endpoint Description
/fhir/metadata CapabilityStatement
/fhir/MedicationKnowledge/{drug-slug}.json Drug resource (189 drugs)
/fhir/ClinicalUseDefinition/{drug-indication}.json Prediction resource (1,268 predictions)
/fhir/Bundle/all-predictions.json Complete dataset

Collaboration Opportunities

We are seeking collaboration with:

1. EHR Vendors & Healthcare Institutions

  • Pilot deployment of SMART App
  • Integration with clinical decision support workflows
  • Feedback for usability improvements

2. Clinical Trial Platforms

  • Enrichment of clinical trial data with AI predictions
  • Patient cohort identification

3. Drug-Drug Interaction Systems

  • Integration of ThTxGNN predictions as supplementary data layer
  • Combined DDI + repurposing alerts

4. Disease-Specific Research Teams

  • Custom analysis for specific therapeutic areas
  • Collaboration on validation studies

Disease Coverage

ThTxGNN covers predictions across multiple therapeutic areas:

Therapeutic Area Example Indications
Oncology Breast carcinoma, colorectal cancer, lung cancer
Neurology Epilepsy, migraine, Parkinson’s disease
Cardiology Hypertension, heart failure, arrhythmia
Infectious Disease Viral infections, bacterial infections
Autoimmune Rheumatoid arthritis, lupus, psoriasis
Metabolic Diabetes, obesity, dyslipidemia
Rare Diseases Various orphan diseases

How to Get Started

For Developers 

# Clone the repository
git clone https://github.com/yao-care/ThTxGNN.git

# Install dependencies
uv sync

# Run knowledge graph prediction
uv run python scripts/run_kg_prediction.py

For Healthcare Institutions

  1. Test the SMART App: Use SMART Launcher with our launch URL
  2. Explore the Data: Visit thtxgnn.yao.care to browse drug reports
  3. Contact Us: Reach out for integration discussions

For Researchers

  • Download Data: CSV/JSON exports available
  • API Access: Use FHIR endpoints for programmatic access
  • Cite Us: Reference our methodology page

Contact Information

Disclaimer

This platform is intended for research purposes only and does not constitute medical advice. All drug repurposing predictions require clinical validation before application. Healthcare decisions should be made in consultation with qualified medical professionals.

Last updated: 2026-03-01