README.md

metadata

tags:
  - biology
  - chemistry
size_categories:
  - 10K<n<100K
dataset_info:
  features:
    - name: pmid
      dtype: string
    - name: section
      dtype: string
    - name: text
      dtype: string
    - name: annotations
      sequence:
        sequence: int64
  splits:
    - name: train
      num_bytes: 120073099
      num_examples: 17619
  download_size: 61504820
  dataset_size: 120073099
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Alzheimer's Disease Tagged Articles Dataset

This dataset contains a collection of scientific articles related to Alzheimer's disease, annotated with biomedical entities (such as diseases, genes, and species) using NCBI’s PubTator tools.

Data Sources and Annotation Tools

• Entity annotations were generated using NCBI's standard models and API calls to the PubTator3 API:
  • TaggerOne and GNORM for gene_species_tagged_articles.json
  • TaggerOne alone for disease_tagged_articles.json
  • PubTator3 API for 'converted_from_bioc.json'

These tools are widely used in biomedical NLP for tagging mentions of diseases, genes, and species within PubMed articles.

Dataset Structure

The dataset is a flattened version of the original JSON files, where each record represents a tagged article section. The structure includes:

• pmid: PubMed ID of the article
• section: Type of section (e.g., title, abstract)
• text: Lowercased content of the section
• annotations: List of [start, end] character offsets identifying entity mentions

Example:

{
  "pmid": "34379990",
  "section": "abstract",
  "text": "clinical progression of tauopathies may result...",
  "annotations": [[0, 26], [45, 53], ...]
}

Preprocessing

The dataset has been simplified:

• Original nested structures were flattened
• Tuples were converted to JSON-compliant lists
• Only title and abstract sections are currently included

Future Work

• Extend tagging beyond titles and abstracts to include full-text sections (e.g., introduction, methods, results)
• Add entity labels (e.g., Disease, Gene, Species) in a future version

Usage

from datasets import load_dataset

dataset = load_dataset("AbrehamT/tagged_articles")

License

This dataset is released under the MIT License.

Acknowledgments

This dataset relies on NCBI’s PubTator, GNORM, and TaggerOne models. Credit goes to the developers of these tools and the researchers whose articles form the dataset foundation.