README.md

---
pretty_name: PrionNER
license: other
license_name: prionner-mixed-data-license
language:
- en
task_categories:
- token-classification
task_ids:
- named-entity-recognition
tags:
- biomedical
- pubmed
- prion-disease
- named-entity-recognition
- brat
- json
configs:
- config_name: fine
  default: true
  data_files:
  - split: train
    path: "data/fine/json/train/*.json"
  - split: test
    path: "data/fine/json/test/*.json"
- config_name: coarse
  data_files:
  - split: train
    path: "data/coarse/json/train/*.json"
  - split: test
    path: "data/coarse/json/test/*.json"
---

# PrionNER

This repository accompanies the paper "PrionNER: A Named Entity Recognition Dataset for Prion Disease Biomedical Literature" and provides a data-only Hugging Face release of PrionNER.

PrionNER is a named entity recognition dataset for prion disease biomedical literature. This release contains canonical train/test splits, fine-grained and coarse-grained annotations, and multiple synchronized data views derived from the same document set.

## Dataset Summary

- Total documents: 317
- Train documents: 247
- Test documents: 70
- Fine-grained schema: 31 defined labels, 30 observed in the released data
- Coarse-grained schema: 15 labels
- Fine/coarse entity count: 4,655 train and 1,650 test
- Discontinuous entities: 97 train and 34 test

The fine-grained schema defines `VPSPr`, but that label does not appear in the current released splits.

## Configurations

This dataset card defines two main configurations for loading with `datasets`:

- `fine`: fine-grained JSON documents with `train` and `test` splits
- `coarse`: coarse-grained JSON documents with `train` and `test` splits

Example:

```python
from datasets import load_dataset

fine = load_dataset("daotuanan/PrionNER", "fine")
coarse = load_dataset("daotuanan/PrionNER", "coarse")
```

The repository also includes auxiliary release files:

- `data/raw/`: original BRAT source pairs
- `data/raw_text/`: text-only files
- `data/fine/brat/` and `data/coarse/brat/`: BRAT exports
- `data/fine/conll/` and `data/coarse/conll/`: CoNLL exports
- `metadata/`: schema files, BRAT config, and summary metadata

## Data Format

Each JSON document includes:

- `doc_id`: document identifier
- `text`: full document text
- `entities`: list of entity annotations

Each entity includes:

- `id`: original BRAT entity ID
- `label`: entity type
- `start`, `end`: document-level character offsets
- `text`: entity surface form
- `spans`: one or more span objects
- `is_discontinuous`: present when the entity spans multiple non-contiguous segments

## Licensing and Rights

This Hugging Face repository uses `license: other` because it contains mixed-rights data.

- The PrionNER annotation layer and project-authored metadata are intended to be released under `CC BY 4.0`, to the extent the repository maintainers hold the necessary rights in those materials.
- The underlying article titles, abstracts, and other source text are included publicly in this repository, but they are **not** blanket-relicensed by the project.
- Please review `LICENSE`, `DATA_LICENSE.md`, and `THIRD_PARTY_RIGHTS.md` before redistributing the source text.

## Provenance

Some released text may be derived from PubMed or MEDLINE records. Where PubMed- or MEDLINE-derived material is used, please acknowledge the U.S. National Library of Medicine as the source of the bibliographic data and do not imply endorsement by NLM or the U.S. Government.

## Citation

If you use PrionNER, please cite the repository and the associated paper:

- "PrionNER: A Named Entity Recognition Dataset for Prion Disease Biomedical Literature"

```bibtex
@misc{dao2026prionnernamedentityrecognition,
      title={PrionNER: A Named Entity Recognition Dataset for Prion Disease Biomedical Literature},
      author={An Dao and Nhan Ly and Thao Tran and Yuji Matsumoto and Akiko Aizawa},
      year={2026},
      eprint={2605.28375},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2605.28375},
}
```