Hugging Face's logo

Datasets:
FuhaiLiAiLab
/
BioMedGraphica

Tasks:
Graph Machine Learning
Token Classification
Feature Extraction
Languages:
English
Tags:
biomedical
knowledge-graph
multi-omics
data-integration
graph-ml
drug-discovery
License:
Dataset card Data Studio Files
xet
Community
BioMedGraphica / README.md
HemingZhang's picture
HemingZhang
Upload README.md with huggingface_hub
0c25469 verified
preview code
|
raw
history blame contribute delete
5.78 kB
metadata
annotations_creators:
  - expert-generated
language:
  - en
license: mit
pretty_name: BioMedGraphica
tags:
  - biomedical
  - knowledge-graph
  - multi-omics
  - data-integration
  - graph-ml
  - drug-discovery
  - text-mining
  - bioinformatics
size_categories:
  - 1M<n<5G
task_categories:
  - graph-ml
  - token-classification
  - feature-extraction
  - other

BioMedGraphica

BioMedGraphica is an all-in-one platform for biomedical data integration and knowledge graph generation. It harmonizes fragmented biomedical datasets into a unified, graph AI-ready resource that facilitates precision medicine, therapeutic target discovery, and integrative biomedical AI research.

Developed using data from 43 biomedical databases, BioMedGraphica integrates:

  • 11 entity types
  • 30 relation types
  • Over 2.3 million entities and 27 million relations

✨ Highlights

  • Multi-omics integration: Genomic, transcriptomic, proteomic, metabolomic, microbiomic, exposomic
  • Graph AI-ready: Outputs subgraphs ready for GNNs and ML models
  • Soft matching: Uses BioBERT for fuzzy entity resolution (disease, phenotype, drug, exposure)
  • GUI software: Provides Windows-based interface for end-to-end pipeline
  • Connected graph variant: Isolated nodes removed for efficient downstream training

📊 Dataset Statistics

Metric Count
Total Entities 2,306,921
Total Relations 27,232,091
Connected Entities 834,809
Connected Relations 27,087,971
Entity Types 11
Relation Types 30

🧬 Entity Types

Entity Type Count Percentage (%) Connected Count Connected (%)
Promoter 230,358 9.99 86,238 10.33
Gene 230,358 9.99 86,238 10.33
Transcript 412,326 17.87 412,039 49.36
Protein 173,978 7.54 121,419 14.54
Pathway 6,793 0.29 1,930 0.23
Metabolite 218,335 9.46 62,364 7.47
Microbiota 621,882 26.96 1,119 0.13
Exposure 1,159 0.05 1,037 0.12
Phenotype 19,532 0.85 19,078 2.29
Disease 118,814 5.15 22,429 2.69
Drug 273,386 11.85 20,918 2.51
Total 2,306,921 100 834,809 100

🔗 Relation Types

Relation Type Count Percentage (%)
Promoter-Gene 230,358 0.85
Gene-Transcript 427,810 1.57
Transcript-Protein 152,585 0.56
Protein-Protein 16,484,820 60.53
Protein-Pathway 152,912 0.56
Protein-Phenotype 478,279 1.76
Protein-Disease 143,394 0.53
Pathway-Protein 176,133 0.65
Pathway-Drug 1,795 0.01
Pathway-Exposure 301,448 1.11
Metabolite-Protein 2,804,430 10.30
Metabolite-Pathway 12,198 0.04
Metabolite-Metabolite 931 0.003
Metabolite-Disease 24,970 0.09
Microbiota-Disease 22,371 0.08
Microbiota-Drug 866 0.003
Exposure-Gene 28,982 0.11
Exposure-Pathway 301,448 1.11
Exposure-Disease 979,780 3.60
Phenotype-Phenotype 23,427 0.09
Phenotype-Disease 181,192 0.67
Disease-Phenotype 181,192 0.67
Disease-Disease 12,006 0.04
Drug-Protein 84,859 0.31
Drug-Pathway 3,065 0.01
Drug-Metabolite 3,589 0.01
Drug-Microbiota 866 0.003
Drug-Phenotype 93,826 0.34
Drug-Disease 39,977 0.15
Drug-Drug 3,882,582 14.26
Total 27,232,091 100

📦 Access and Downloads

🧪 Validation

  • Hard matching for structured identifiers (e.g. Ensembl, HGNC)
  • BioBERT-based soft matching for flexible terms (e.g., diseases, phenotypes, drugs)
  • Case study and benchmarking with Synapse dataset

📚 Citation 

@article{zhang2024biomedgraphica,
title={BioMedGraphica: An All-in-One Platform for Biomedical Prior Knowledge and Omic Signaling Graph Generation},
author={Zhang, Heming and Liang, Shunning and Xu, Tim and Li, Wenyu and Huang, Di and Dong, Yuhan and Li, Guangfu and Miller, J Philip and Goedegebuure, S Peter and Sardiello, Marco and others},
journal={bioRxiv},
year={2024}
}