synapticstudio/rxguard-ddi-arabic

---
license: cc-by-nc-sa-4.0
language:
- ar
tags:
- drug-interactions
- pharmacy
- arabic
- egyptian-arabic
- medical
- ddi
- rxguard
size_categories:
- 10K<n<100K
---

# RxGuard DDI Arabic Dataset v3

Egyptian Arabic fine-tuning dataset for drug interaction checking — built from RxGuard DB.

## Stats
- **Total samples:** 17,334
- **Format:** OpenAI chat JSONL (system + user + assistant)
- **Language:** Egyptian Arabic (عربي مصري)

## Data Types (16 types)

| Type | Count | % |
|------|-------|---|
| ddi | 11,903 | 68.7% |
| herb | 1,629 | 9.4% |
| food | 845 | 4.9% |
| drug_disease | 626 | 3.6% |
| safe_pair | 600 | 3.5% |
| multi_drug | 400 | 2.3% |
| cyp450 | 270 | 1.6% |
| drug_info | 267 | 1.5% |
| side_effects | 242 | 1.4% |
| refusal | 195 | 1.1% |
| faers_ae | 155 | 0.9% |
| renal_hepatic | 138 | 0.8% |
| blackbox | 37 | 0.2% |
| nti | 11 | 0.1% |
| allergy | 8 | 0.0% |
| same_ingredient | 8 | 0.0% |

## Changes from v1 → v3
- **v1:** 13,497 samples, 12 types
- **v3:** 17,334 samples, 16 types
- **New:** Drug-Disease Contraindications (626), FAERS real-world adverse effects (155), Same-ingredient warnings (8), Out-of-scope refusal (195)
- **Expanded:** Herb interactions (164 → 1,629 via CYP450), Food interactions (141 → 845 via CYP450)

## Data Sources
- OpenFDA (drug labels, FAERS adverse event reports)
- RxNorm, DDInter, DrugBank references
- CYP450 cross-referencing (herbs/foods as enzyme actors)
- Clinical rule-based drug-disease contraindications
- Manually curated Egyptian pharmacy context

## Usage
```python
from datasets import load_dataset
ds = load_dataset("synapticstudio/rxguard-ddi-arabic")
```