{"data":{"kind":"file","path":"README.md","version_id":"k6wbpmdwbkxqnfvgu8411ktn","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2093,"modified_at":"2026-01-13T23:06:43.798000","content_hash":"2acfd0e382a77db4552d59079a3528b0d971409c5e29207f59789c47a783bc58"},"entries":[],"content":"# OpenMed ADE Environment\n\nAdverse Drug Event (ADE) detection environment for RL fine-tuning using the ADE Corpus V2.\n\n## Task Description\n\nGiven a sentence from medical literature, classify whether it describes an adverse drug event:\n\n| Class | Label | Description |\n|-------|-------|-------------|\n| A | ADE_PRESENT | Clear adverse effect caused by a drug is described |\n| B | ADE_POSSIBLE | Potential or uncertain adverse effect mentioned |\n| C | NO_ADE | No adverse drug event described in the sentence |\n\n## Dataset\n\n- **Source**: [ade-benchmark-corpus/ade_corpus_v2](https://huggingface.co/datasets/ade-benchmark-corpus/ade_corpus_v2)\n- **Original**: Gurulingappa et al., 2012 (Journal of Biomedical Informatics)\n- **Combined Examples**: ~23,500 (from classification + relation extraction configs)\n- **Train/Eval Split**: 80/20\n\n## Reward Structure\n\n| Component | Weight | Description |\n|-----------|--------|-------------|\n| Accuracy | 80% | Exact match on ADE classification (A-C) |\n| Thinking | 15% | Quality of reasoning in `<think>` tags |\n| Format | 5% | Proper `\\boxed{}` answer format |\n\n## Example\n\n**Input:**\n```\nSentence: Intravenous azithromycin-induced ototoxicity.\nDrug mentioned: azithromycin\nPotential adverse effect: ototoxicity\n```\n\n**Expected Output:**\n```\n<think>\nThe sentence clearly states that azithromycin administered intravenously induced\nototoxicity (hearing damage). The word \"induced\" establishes a direct causal\nrelationship between the drug and the adverse effect.\n</think>\n\\boxed{A}\n```\n\n## Usage\n\n```python\nfrom OpenMed_ADE import load_environment\n\nenv = load_environment()\n```\n\n## Citation\n\n```bibtex\n@article{gurulingappa2012benchmark,\n  title={Benchmark corpus to support information extraction for adverse drug effects},\n  author={Gurulingappa, Harsha and Rajput, Abdul Mateen and Roberts, Angus and Fluck, Juliane and Hofmann-Apitius, Martin and Toldo, Luca},\n  journal={Journal of Biomedical Informatics},\n  volume={45},\n  number={5},\n  pages={885--892},\n  year={2012},\n  publisher={Elsevier}\n}\n```\n\n## License\n\nPublic domain (original ADE Corpus)\n","encoding":"utf-8","truncated":false,"total_bytes":2093},"status":null}