Finds protein-related entities in English biomedical text with user-supplied labels
What it does
This zero-shot model detects protein entities—including protein complexes, families, groups, and variants—in English biomedical text, with labels supplied at inference time. It reports an F1 of 0.845 on the FSU-PRGE benchmark, uses a GLiNER backbone, and has 220M parameters.
Best for
- Protein-protein interaction mining
- Pathway modeling from literature
- Custom entity extraction in research
What it detects (5 labels)
Trained on the label below; you can pass any other label names at inference time.
- Protein
- Protein complex
- Protein (enumeration)
- Protein family or group
- Protein variant
Quality
| Benchmark | FSU-PRGE · all models on this benchmark |
|---|---|
| F1 | 0.845 |
| Precision | 0.823 |
| Recall | 0.869 |
| Accuracy | 0.967 |
| Source | Reported by the authors (test_results.json in the repository) |
Biomedical Entity Recognition - Various biomedical entities
Use it
pip install gliner
from gliner import GLiNER
model = GLiNER.from_pretrained("OpenMed/OpenMed-ZeroShot-NER-Protein-Base-220M")
print(model.predict_entities("Patient with type 2 diabetes and hypertension.", ["disease"]))All model information shown here is public and belongs to its authors: use and download it freely under each model’s license. The selection, organization, search design and presentation of this catalog are © 2026 Clinical Corpus. Please link to these pages rather than copy them.