jashparekh/gaucher-disease-qa

Gaucher Disease QA

A long-form, multi-hop question answering dataset about Gaucher disease, a rare lysosomal storage disorder caused by mutations in the GBA1 gene.

Dataset Description

This dataset contains 25 expert-curated causal reasoning questions grounded in 20 biomedical articles covering the genetics, pathophysiology, clinical manifestations, and treatment of Gaucher disease. Each question requires synthesizing information from multiple source documents to produce a comprehensive answer, making it a challenging benchmark for multi-hop QA and biomedical causal inference.

Key Properties

• Domain: Biomedical / rare disease / genetics
• Question type: Long-form causal reasoning (not extractive)
• Multi-hop: Every question requires evidence from 2-4 source documents
• Total queries: 25
• Total documents: 20
• Total relevance judgments: 86
• Average documents per query: 3.4

Dataset Format (BEIR-style)

The dataset follows the BEIR format with three JSONL files:

queries.jsonl

{"id": "q1", "question": "How does a mutation in the GBA gene disrupt the catalytic activity of glucocerebrosidase?", "answer": "Mutations in the GBA gene can cause the enzyme to fold improperly..."}

corpus.jsonl

{"id": "doc1", "text": "Full article text..."}

qrels.jsonl

{"query_id": "q1", "doc_ids": ["doc1", "doc8", "doc15", "doc19"]}

Topics Covered

The questions span the full causal chain of Gaucher disease:

• Molecular basis: GBA1 mutations, enzyme misfolding, residual activity, genotype-phenotype correlations
• Cellular pathology: Glucocerebroside accumulation, Gaucher cell formation, lysosomal dysfunction
• Organ involvement: Splenomegaly, hepatomegaly, bone marrow infiltration, impaired hematopoiesis
• Neurological complications: Neuronopathic types 2/3, cognitive difficulties, Parkinson's disease link (alpha-synuclein/GCase vicious cycle)
• Metabolic effects: Hypermetabolism (REE 44% elevated), GM3-mediated insulin resistance, AKT pathway disruption
• Clinical management: ERT vs SRT (eliglustat/CYP2D6), splenectomy risks, biomarkers (chitotriosidase), imaging
• Emerging therapies: Gene therapy (AAV vectors, BBB crossing)
• Genetics: Autosomal recessive inheritance, genetic counseling

Usage

from datasets import load_dataset

dataset = load_dataset("jashparekh/gaucher-disease-qa")

Or load the JSONL files directly:

import json

queries = [json.loads(line) for line in open("queries.jsonl")]
corpus = [json.loads(line) for line in open("corpus.jsonl")]
qrels = [json.loads(line) for line in open("qrels.jsonl")]

Intended Use

• Benchmarking multi-hop question answering systems on biomedical text
• Evaluating causal reasoning over scientific/medical document collections
• Testing retrieval-augmented generation (RAG) pipelines on domain-specific content
• Evaluating knowledge graph-based QA systems for rare disease information

Source

All questions and gold answers were expert-curated and verified against the 20 source articles. The articles include clinical reviews, gene therapy research papers, patient-facing medical resources, and comprehensive disease overviews from sources such as GeneReviews, NORD, NIH, Mayo Clinic, and peer-reviewed journals.

Citation

This dataset was created as part of the SARG (Structure-Augmented Reasoning Generation) project. If you use this dataset, please cite:

@article{parekh2025structure,
  title={Structure-Augmented Reasoning Generation},
  author={Parekh, Jash Rajesh and Jiang, Pengcheng and Han, Jiawei},
  journal={arXiv preprint arXiv:2506.08364},
  year={2025}
}

License

MIT