CoolFace
Datasetpublic

open-biosciences/biosciences-sources

Biosciences RAG Source Documents Dataset Description This dataset contains 140 page-level document chunks extracted from 10 biomedical research papers. The documents form the knowledge base for a Retrieval-Augmented Generation (RAG) system focused on biosciences topics including knowledge graphs, LLM applications in biomedicine, and protein interaction databases. Dataset Summary Total Documents: 140 pages from 10 research papers Domain: Biomedical… See the full description on the dataset page: https://huggingface.co/datasets/open-biosciences/biosciences-sources.

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face
0likes29downloads
Dataset Card

Biosciences RAG Source Documents

Dataset Description

This dataset contains 140 page-level document chunks extracted from 10 biomedical research papers. The documents form the knowledge base for a Retrieval-Augmented Generation (RAG) system focused on biosciences topics including knowledge graphs, LLM applications in biomedicine, and protein interaction databases.

Dataset Summary

  • —Total Documents: 140 pages from 10 research papers
  • —Domain: Biomedical NLP, knowledge graphs, RAG in life sciences
  • —Format: PDF pages with extracted text and metadata
  • —Use Case: RAG knowledge base for biosciences-related queries

Source Papers

PagesTopic
31Biomedical preprint (bioRxiv 2025.08.01.668022)
24Evaluating LLMs for Gene-to-Phenotype Mapping
18Biomedical NLP (arXiv 2405.04819)
16A Hybrid LLM-Knowledge Graph Framework for Biomedical QA
12Knowledge Graph-Enhanced LLM for Pan-Cancer QA
11Biomedical NLP (arXiv 2406.18626)
11Improving LLM Applications in Biomedicine with RAG (systematic review)
8The STRING Database in 2025: Protein Networks with Directionality
5STRING database (Nucleic Acids Research, gkv1277)
4Computational biology (btad570)

Data Fields

  • —page_content (string): Extracted text content from the PDF page
  • —metadata (dict): Document metadata including:
  • —title: Paper title
  • —author: Paper authors
  • —page: Page number
  • —total_pages: Total pages in source document
  • —source: Original file path
  • —format: Document format (PDF)
  • —producer, creator: PDF metadata
  • —keywords, subject: Paper keywords and subject (where available)

Data Splits

This dataset contains a single split with all 140 documents.

Licensing

This dataset is released under the Apache 2.0 license.

Dataset Creation

Created as part of the Open Biosciences RAG evaluation pipeline. Source PDFs are ingested via make ingest which runs scripts/ingest_raw_pdfs.py with LlamaIndex transformations (HeadlineSplitter, SummaryExtractor, EmbeddingExtractor, ThemesExtractor, NERExtractor).

Related Datasets

  • —Golden Testset: open-biosciences/biosciences-golden-testset (12 QA pairs generated from these sources)
  • —Evaluation Inputs: open-biosciences/biosciences-evaluation-inputs (RAG outputs)
  • —Evaluation Metrics: open-biosciences/biosciences-evaluation-metrics (RAGAS scores)