Downloads

Download MyTxGNN prediction data for your research.

---

Available Datasets

Knowledge Graph Predictions

File	Description	Format	Size
repurposing_candidates.csv	KG-based predictions	CSV	~4.7 MB

Columns:

• license_id: NPRA registration number
• brand_name: Product name
• ingredient: Active ingredient
• drugbank_id: DrugBank identifier
• disease_name: Predicted indication
• disease_id: Disease identifier
• source: Prediction source
• is_new: New indication flag

Statistics:

• 41,560 predictions
• 508 unique drugs
• 940 unique diseases

---

DrugBank Mapping

File	Description	Format	Size
drugbank_mapping.csv	Drug ID mappings	CSV	~1.8 MB

Columns:

• ingredient: Normalized ingredient name
• drugbank_id: DrugBank identifier
• match_score: Mapping confidence

---

Deep Learning Predictions

File	Description	Format	Size
txgnn_checkpoint.csv	DL model predictions	CSV	~756 MB

Columns:

• drugbank_id: DrugBank identifier
• drug_name: Drug name
• disease_name: Predicted indication
• txgnn_score: Confidence score (0-1)

Statistics:

• 9,968,985 total predictions
• 176,021 high-confidence (score ≥ 0.7)
• 585 unique drugs
• 17,041 unique diseases

---

How to Download

Via GitHub

All data files are available in the project repository:

git clone https://github.com/yao-care/MyTxGNN.git
cd MyTxGNN/data/processed

Direct Links

• KG Predictions
• DrugBank Mapping

---

FHIR Resources

FHIR R4 formatted resources are also available:

Resource Type	Count	Location
MedicationKnowledge	508	/fhir/MedicationKnowledge/
ClinicalUseDefinition	41,560	/fhir/ClinicalUseDefinition/

Access via:

GET /fhir/MedicationKnowledge/{drugbank_id}.json
GET /fhir/ClinicalUseDefinition/{drug}-{indication}.json

---

Data Dictionary

Prediction Score Interpretation

Score Range	Meaning
0.95 - 1.00	Very high confidence
0.90 - 0.95	High confidence
0.70 - 0.90	Moderate-high confidence
0.50 - 0.70	Moderate confidence
< 0.50	Low confidence

Disease Naming Convention

Disease names follow the TxGNN knowledge graph ontology, which is derived from:

• MONDO Disease Ontology
• Disease Ontology (DO)
• Human Phenotype Ontology (HPO)

---

Usage Examples

Python

import pandas as pd

# Load KG predictions
kg = pd.read_csv('repurposing_candidates.csv')

# Filter for a specific drug
metformin = kg[kg['ingredient'] == 'METFORMIN']

# Get high-score DL predictions
dl = pd.read_csv('txgnn_checkpoint.csv')
high_conf = dl[dl['txgnn_score'] >= 0.7]

R

library(readr)

# Load KG predictions
kg <- read_csv('repurposing_candidates.csv')

# Filter by drug
prednisolone <- kg %>% filter(ingredient == 'PREDNISOLONE')

# Summary statistics
kg %>% group_by(disease_name) %>% summarise(n = n()) %>% arrange(desc(n))

---

Terms of Use

By downloading this data, you agree to:

1. Attribution: Cite MyTxGNN and TxGNN (Huang et al., 2023) in publications
2. Research Use: Data is provided for research purposes
3. No Clinical Decisions: Do not use predictions for clinical decisions without validation
4. Compliance: Follow applicable data protection regulations

---

Citation

When using this data, please cite:

@article{huang2023txgnn,
  title={A foundation model for clinician-centered drug repurposing},
  author={Huang, Kexin and others},
  journal={Nature Medicine},
  year={2023},
  doi={10.1038/s41591-023-02233-x}
}

---

Disclaimer
Prediction data is provided for research purposes only. Predictions have not been clinically validated and should not be used for medical decisions.