Data Sources

MyTxGNN integrates multiple authoritative data sources for comprehensive drug repurposing predictions.

---

Primary Data Sources

TxGNN Knowledge Graph

Attribute	Value
Source	Harvard Zitnik Lab
Publication	Nature Medicine (2023)
Nodes	17,080 (drugs, diseases, genes, proteins)
Use	AI prediction model

The TxGNN knowledge graph forms the foundation of our predictions, integrating relationships from multiple biomedical databases.

Malaysia NPRA Drug Registry

Attribute	Value
Source	data.gov.my
Agency	National Pharmaceutical Regulatory Agency (NPRA)
Records	27,938 pharmaceutical products
License	CC BY 4.0
Update Frequency	Daily

Fields Used:

• Registration number (reg_no)
• Product name
• Active ingredients
• Registration status
• Holder information

DrugBank

Attribute	Value
Source	DrugBank Online
Version	Latest available
Drugs Mapped	508
Use	Drug standardization, mechanism data

Fields Used:

• DrugBank ID
• Drug name standardization
• Mechanism of action
• Known indications

---

Evidence Collection Sources

ClinicalTrials.gov

Attribute	Value
Source	ClinicalTrials.gov
Operated By	U.S. National Library of Medicine
Use	Clinical trial evidence
API	ClinicalTrials.gov API v2

Fields Collected:

• NCT ID (trial identifier)
• Trial phase
• Study status
• Enrollment count
• Start/completion dates
• Sponsor information

PubMed

Attribute	Value
Source	PubMed
Operated By	NCBI
Use	Literature evidence
API	E-utilities API

Fields Collected:

• PMID
• Article title
• Publication year
• Journal name
• Abstract

WHO ICTRP

Attribute	Value
Source	WHO ICTRP
Operated By	World Health Organization
Use	International clinical trials
Coverage	Southeast Asia registries (TCTR, etc.)

---

Data Processing Pipeline

NPRA Data → DrugBank Mapping → TxGNN Prediction → Evidence Collection → Report Generation

Step 1: NPRA Data Processing

1. Download pharmaceutical products dataset from data.gov.my
2. Filter for active registrations (PRODUCT APPROVED)
3. Extract and normalize active ingredients
4. Parse ingredient names and dosage information

Step 2: DrugBank Mapping

1. Normalize drug names (remove salts, standardize spelling)
2. Match to DrugBank IDs using fuzzy matching
3. Validate mappings against DrugBank database
4. Mapping Rate: ~73% of unique ingredients

Step 3: TxGNN Prediction

1. Load TxGNN model and knowledge graph
2. Run knowledge graph inference for drug-disease pairs
3. Run deep learning model for confidence scores
4. Filter and rank predictions

Step 4: Evidence Collection

1. Query ClinicalTrials.gov for each drug-disease pair
2. Search PubMed for relevant literature
3. Check ICTRP for regional trials
4. Aggregate and format evidence

---

Data Quality Measures

Validation Checks

• Drug name normalization accuracy
• DrugBank ID verification
• Clinical trial ID format validation
• PubMed ID existence check

Coverage Statistics

Metric	Value
NPRA products processed	27,938
Unique ingredients	~2,500
DrugBank mapped	508 (73.87% of matchable)
KG predictions	41,560
DL predictions (score ≥0.7)	176,021

---

Data Update Schedule

Data Source	Update Frequency
NPRA Registry	Weekly
DrugBank	Monthly
ClinicalTrials.gov	On-demand
PubMed	On-demand

---

Terms of Use

NPRA Data

• Licensed under CC BY 4.0 by Malaysia Government
• Attribution required

DrugBank Data

• Used under academic license
• Attribution required

Clinical Trial Data

• Public domain (ClinicalTrials.gov)
• No restrictions

PubMed Data

• NCBI Disclaimer applies
• No restrictions on article metadata

---

Citation Requirements

When using MyTxGNN data, please cite:

1. TxGNN Model: Huang et al., Nature Medicine (2023)
2. NPRA Data: data.gov.my with CC BY 4.0 attribution
3. DrugBank: DrugBank Online with appropriate license citation

---

Data Disclaimer
Data is provided for research purposes only. While we strive for accuracy, users should verify critical information from primary sources. Data may be subject to change without notice.