README.md

---
license: apache-2.0
task_categories:
- graph-ml
- tabular-classification
language:
- en
tags:
- biology
- bioinformatics
- knowledge-graph
- graph-neural-networks
- drug-discovery
- medical
- disease-gene-prediction
- protein-chemical-interaction
- medical-ontology
size_categories:
- 100K<n<1M
---

# BioGraphFusion Dataset

[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
[![Paper](https://img.shields.io/badge/Paper-Bioinformatics-green.svg)](https://doi.org/10.1093/bioinformatics/btaf408)
[![arXiv](https://img.shields.io/badge/arXiv-2507.14468-b31b1b.svg)](https://arxiv.org/abs/2507.14468)

## 📊 Dataset Description

This dataset contains the benchmark data used in the paper **"BioGraphFusion: Graph Knowledge Embedding for Biological Completion and Reasoning"** published in *Bioinformatics*.

## 🏗️ Dataset Structure

The dataset includes three biomedical knowledge graph completion tasks with background knowledge integration:

### 1. Disease-Gene Prediction (DisGeNet_cv)

- **Task**: Disease-gene association prediction
- **Background Knowledge**: Drug-Disease relationships from SIDER (14,631 triples) + Protein-Chemical relationships from STITCH (277,745 triples)
- **Main Dataset**: DisGeNet (130,820 triples) focusing on gene targets
- **Description**: Predicts disease-gene associations using multi-source biological knowledge

### 2. Protein-Chemical Interaction (STITCH)

- **Task**: Protein-chemical interaction prediction
- **Background Knowledge**: Drug-Disease relationships from SIDER (14,631 triples) + Disease-Gene relationships from DisGeNet (130,820 triples)
- **Main Dataset**: STITCH (23,074 triples) focusing on chemical targets
- **Description**: Predicts protein-chemical interactions with integrated disease and gene knowledge

### 3. Medical Ontology Reasoning (UMLS)

- **Task**: Medical concept reasoning
- **Background Knowledge**: Various medical relationships from UMLS (4,006 triples)
- **Main Dataset**: UMLS (2,523 triples) with multi-domain entities
- **Description**: Reasons about medical concepts and their hierarchical relationships

## 📈 Dataset Statistics

| Dataset | Task | Background Knowledge Sources | Main Dataset Targets | Total Triples |
|---------|------|------------------------------|---------------------|---------------|
| **Disease-Gene Prediction** | Disease-gene association prediction | Drug-Disease Relationships SIDER (14,631) + Protein-Chemical Relationships STITCH (277,745) | DisGeNet (130,820) Gene | ~423K |
| **Protein-Chemical Interaction** | Protein-chemical interaction prediction | Drug-Disease Relationships SIDER (14,631) + Disease-Gene Relationships DisGeNet (130,820) | STITCH (23,074) Chemical | ~168K |
| **Medical Ontology Reasoning** | Medical concept reasoning | Various Medical Relationships UMLS (4,006) | UMLS (2,523) Multi-domain Entities | ~6.5K |

## 💻 Usage

### Loading the Dataset

```python
from datasets import load_dataset

# Load the complete dataset
dataset = load_dataset("Y-TARL/BioGraphFusion")

# Load specific task
disgenet_data = load_dataset("Y-TARL/BioGraphFusion", "Disease-Gene")
stitch_data = load_dataset("Y-TARL/BioGraphFusion", "Protein-Chemical") 
umls_data = load_dataset("Y-TARL/BioGraphFusion", "umls")
```

## 📝 Citation

If you use this dataset in your research, please cite our paper:

```bibtex
@article{lin2025biographfusion,
  title={BioGraphFusion: Graph Knowledge Embedding for Biological Completion and Reasoning},
  author={Lin, Yitong and He, Jiaying and Chen, Jiahe and Zhu, Xinnan and Zheng, Jianwei and Tao, Bo},
  journal={Bioinformatics},
  pages={btaf408},
  year={2025},
  publisher={Oxford University Press}
}
```

## 🔗 Related Resources

- **Paper**: [Bioinformatics](https://doi.org/10.1093/bioinformatics/btaf408)
- **Preprint**: [arXiv:2507.14468](https://arxiv.org/abs/2507.14468)
- **Code**: [GitHub Repository](https://github.com/Y-TARL/BioGraphFusion)

## 📄 License

This dataset is released under the Apache 2.0 License.

## 🙏 Acknowledgements

We thank the original data providers:

- DisGeNet for disease-gene associations
- STITCH for protein-chemical interactions
- UMLS for medical ontology data

## 📞 Contact

For questions about the dataset, please open an issue in the [GitHub repository](https://github.com/Y-TARL/BioGraphFusion/issues).