CROssBARv2-KG

This repository provides the dataset for the CROssBARv2 Knowledge Graph (KG), a heterogeneous and general-purpose biomedical KG-based system.

The CROssBARv2 KG comprises approximately 2.7 million nodes spanning 14 distinct node types and around 12.6 million edges representing 51 different edge types, all integrated from 34 biological data sources. We also incorporated several ontologies (e.g., Gene Ontology, Mondo Disease Ontology) along with rich metadata captured as node and edge properties.

Building upon this foundation, we further enhanced the semantic depth of CROssBARv2. This was achieved by generating and storing embeddings for key biological entities, such as proteins, drugs, and Gene Ontology terms. These embeddings are managed using the native vector index feature in Neo4j, enabling powerful semantic similarity searches.

Data

The dataset is organized into two primary directories:

• nodes/: Contains CSV files defining the biological entities.
• edges/: Contains CSV files defining the relationships between these entities.

nodes

Each file corresponds to a specific biological entity type and contains a primary identifier column that provides a unique ID along with additional metadata fields. Identifiers follow the Compact URI (CURIE) standard (e.g., uniprot:Q9H161) as defined in the Bioregistry.

The table below lists the primary identifier column name for each node file:

File	ID Column Name
Cellular_component.csv	id
Biological_process.csv	id
Molecular_function.csv	id
Protein.csv	id
Phenotype.csv	hpo_id
Pathway.csv	pathway_id
Ec.csv	ec_number
Compound.csv	compound_id
Gene.csv	id
Side_effect.csv	meddra_id
Drug.csv	drugbank_id
Drug.csv	drugbank_id
Organism.csv	id
Domain.csv	id
Disease.csv	disease_id

The following table demonstrates the CURIE format used for each node type within the KG:

Node Type	CURIE
Protein	uniprot:Q9H161
Gene	ncbigene:60529
OrganismTaxon	ncbitaxon:9606
ProteinDomain	interpro:IPR000001
Drug	drugbank:DB00821
Compound	chembl:CHEMBL6228
GOTerm (BiologicalProcess, MolecularFunction, CellularComponent)	go:0016072
Disease	mondo:0054666
Phenotype	hp:0000012
SideEffect	meddra:10073487
EcNumber	eccode:1.1.1.-

edges

Each file represents a relationship between two biological entity types in the KG.
Every edge file contains source and target columns, which specify the identifiers of the nodes being linked.
Other columns serve as metadata for the relationship.

The table below lists the filename along with the source and target identifier column names and a description of the edge type:

File	Source Column	Target Column	Description
Pathway_orthology.csv	pathway1_id	pathway2_id	Evolutionarily conserved pathway relationship
Go_to_go.csv	source	target	Hierarchical or regulatory connection between functional terms
Gene_to_disease_edge.csv	gene_id	disease_id	Gene-disease link via expression or function Genetic variant linked to disease risk/pathology
Ec_hierarchy.csv	child_id	parent_id	Hierarchical classification of enzyme functions
Disease_to_drug_edge.csv	disease_id	drug_id	Therapeutic intervention with a drug
DTI.csv	drugbank_id	uniprot_id	Direct target binding or modulation
Phenotype_hierarchical_edges.csv	child_id	parent_id	Hierarchical classification of phenotypes
Protein_to_phenotype.csv	protein_id	hpo_id	Protein involvement in a phenotypic condition
Drug_to_side_effect.csv	drugbank_id	meddra_id	Associated adverse effect of a drug
Disease_to_disease_comorbidity_edge.csv	disease1	disease2	Comorbid occurrence in patients
DGI.csv	entrez_id	drugbank_id	Positive/negative regulation of gene expression
Drug_to_pathway.csv	drug_id	pathway_id	Target involvement in a biological pathway
PPI.csv	uniprot_a	uniprot_b	Physical/functional protein-protein association
Disease_to_disease_association_edge.csv	disease_id1	disease_id2	Statistical or mechanistic disease link
Phenotype_to_disease.csv	hpo_id	disease_id	Disease-related clinical trait
Orthology.csv	entrez_a	entrez_b	Evolutionary relationship between genes
Side_effect_hierarchy.csv	child_id	parent_id	Hierarchical classification of side effects
Organism_to_disease_edge.csv	organism_id	disease_id	Pathogen-induced disease etiology
DDI.csv	drug1	drug2	Pharmacological or biochemical interaction
Protein_to_ec.csv	protein_id	ec_id	Enzymatic reaction catalysis
Reactome_hierarchical_edges.csv	child_id	parent_id	Hierarchical or functional pathway connection
Tf_gene_edges.csv	tf	target	Regulatory influence on gene expression
Disease_hiererchical_edges.csv	child_id	parent_id	Hierarchical classification of diseases
Protein_to_pathway.csv	uniprot_id	pathway_id	Functional participation in a biological pathway
Protein_has_domain.csv	source_id	target_id	Protein contains structural/functional domain
Protein_belongs_to_organism.csv	source_id	target_id	Protein origin specific to an organism
Protein_to_go.csv	source	target	Molecular activity performed by protein / Cellular component localization / Biological role or process involvement
Pathway_to_pathway.csv	pathway_id1	pathway_id2	Hierarchical or functional pathway connection
Gene_encodes_protein.csv	source_id	target_id	Genetic encoding of a protein product
Disease_to_pathway.csv	disease_id	pathway_id	Regulation of biological pathway activity
Domain_to_go.csv	source	target	Functional role enabled by domain / Structural or localization role of domain / Biological role or process involvement of domain
CTI.csv	chembl	uniprot_id	Direct target binding or modulation

How to Use

You can easily load this dataset using the Hugging Face datasets library:

from datasets import load_dataset

# Example 1: Load the Protein nodes
proteins = load_dataset("HUBioDataLab/CROssBARv2-KG", data_files="nodes/Protein.csv")

# Example 2: Load Drug-Target Interactions (Edges)
dti_edges = load_dataset("HUBioDataLab/CROssBARv2-KG", data_files="edges/DTI.csv")