TCM-MKG

Dataset Description

TCM-MKG (Traditional Chinese Medicine Multidimensional Knowledge Graph) is a large-scale biomedical knowledge resource for Traditional Chinese Medicine (TCM)-related research.

This repository is a Hugging Face distribution of the publicly released TCM-MKG V1.0 dataset. It preserves the original Zenodo source tables and additionally provides a Hub-friendly tabular export under original_kg/ for convenient machine access.

This repository should be understood as a format conversion and redistribution of the released TCM-MKG data. It does not claim newly collected data, newly curated biomedical evidence, or a newly constructed knowledge graph. The underlying knowledge content comes from the original TCM-MKG authors and their integration of publicly available resources.

The dataset can support research in:

• Traditional Chinese Medicine knowledge representation
• Knowledge graph reasoning and completion
• Knowledge graph embedding
• Drug discovery and drug repurposing
• Network pharmacology
• AI-assisted biomedical research

Data Source

The original dataset is:

Traditional Chinese Medicine Multidimensional Knowledge Graph (TCM-MKG V1.0)

Zenodo DOI:

https://doi.org/10.5281/zenodo.13763953

The original Zenodo release provides the source tables D1-D24, the supplementary table SD1, and the accompanying documentation PDF. This Hugging Face release keeps those raw files and also exposes a tabular nodes.tsv / edges.tsv export for downstream knowledge graph and machine learning workflows.

Users should cite the original dataset when using this resource.

Files

• original_kg/nodes.tsv: node table with columns node_id, node_type, name, aliases, source_table, and extra_json
• original_kg/edges.tsv: edge table with columns source_id, target_id, relation, source_table, evidence, and attributes_json
• D1_*.tsv to D24_*.tsv: original source-specific tables from the Zenodo release
• SD1_predicted_InChIKey_EntrezID.tsv: supplementary source table from the Zenodo release
• TCM-MKG_Open_Source_Documentation.pdf: original documentation file

In this release, original_kg/nodes.tsv contains 369,912 rows and original_kg/edges.tsv contains 48,849,793 rows.

Dataset Structure

The dataset contains multidimensional biomedical knowledge related to:

• TCM terminology
• Chinese patent medicines
• Chinese herbal pieces
• Natural products and chemical identifiers
• Biological targets
• Diseases and ontology mappings
• Protein-protein interactions

The Hub-oriented graph export can be used as entity-relation-entity style data:

• nodes.tsv stores normalized entity records
• edges.tsv stores graph relations together with provenance through source_table and JSON attributes

The source_table field links exported graph rows back to the original source-table identifiers such as D3, D6, or D23.

Usage

This dataset can be used for:

• Knowledge graph completion
• Entity and relation representation learning
• Link prediction
• Drug discovery and drug repurposing
• Traditional Chinese Medicine mechanism exploration
• Network pharmacology analysis
• Biomedical knowledge graph research

Example loading with Hugging Face datasets:

from datasets import load_dataset

nodes = load_dataset("JX-Lab/TCM-MKG", "nodes", split="train")
edges = load_dataset("JX-Lab/TCM-MKG", "edges", split="train", streaming=True)

For very large graph operations, streaming is recommended for edges.tsv.

Citation

If you use this dataset, please cite the original TCM-MKG dataset:

@dataset{zeng_jingqi_2024_13763953,
  author       = {Zeng, Jingqi and Jia, Xianpei},
  title        = {Traditional Chinese Medicine Multidimensional Knowledge Graph (TCM-MKG)},
  year         = {2024},
  publisher    = {Zenodo},
  version      = {V1.0},
  doi          = {10.5281/zenodo.13763953},
  url          = {https://doi.org/10.5281/zenodo.13763953}
}

Disclaimer

This repository provides a Hugging Face-compatible distribution of the publicly available TCM-MKG dataset.

The original data were collected, integrated, and released by the original authors from publicly available resources. This repository does not claim ownership of the original data and should not be interpreted as introducing a newly built knowledge graph beyond format conversion for repository access.

Users are responsible for complying with the original dataset license and for appropriately citing the original authors and original data sources when using this resource.