Table 1.

Prominent open-source benchmark datasets for machine learning on biological networks.

Data type	Database	Task type	Prediction tasks
General	Long Range Graph Benchmark (Dwivedi et al. 2022b)	Edge-level	Molecular bond
		Graph-level	Peptide function, peptide structure
General	Open Biomedical Network	Node-level	Protein function
	Benchmark (Liu and Krishnan 2024)	Edge-level	Disease–gene association
General	Open Graph Benchmark (Hu et al. 2020b)	Node-level	Protein function
		Edge-level	Protein–protein association, drug–drug interaction, heterogeneous interaction, vessels in mouse brain
		Graph-level	Molecular property, species-specific protein association
General	SubGNN Benchmarks (Alsentzer et al. 2020)	Subgraph-level	Proteins associated with biological process, rare neurological disorders phenotype-based diagnosis, and rare metabolic disorders phenotype-based diagnosis
General	Temporal Graph Benchmark (Huang et al. 2024)	Node-level	Dynamic node affinity prediction
		Edge-level	Dynamic link prediction
Knowledge graph	PrimeKG (Chandak et al. 2023)	Node-level	Identity of protein/gene, disease, drug, biological process, pathway, phenotype, molecular function, cellular component, exposure, and anatomical region
		Edge-level	Protein–protein interaction, disease–drug indication, disease–drug contraindication, disease-drug off-label use, disease–phenotype association, disease–disease association, disease–protein association, disease–exposure association, phenotype–protein association, pathway–gene association, etc.
Knowledge graph	Phenotype Knowledge Translator (Callahan et al. 2024)	Node-level	Identity of tissue, cell, DNA, RNA, gene, miRNA, variant, protein, disease, biological process, pathway, phenotype, molecular function, cellular component, and chemical
		Edge-level	Tissue-/cell-specific gene expression, gene-variant association, variant-disease association, chemical-disease association, chemical-pathway association, etc.
Molecular design	Protein sEquence undERstanding (Xu et al. 2022)	Edge-level	Protein–protein interaction, contact prediction
		Graph-level	Molecular property (e.g. fold classification, secondary structure prediction)
Molecular design	Tasks Assessing Protein Embeddings (Rao et al. 2019)	Edge-level	Protein–protein interaction, contact prediction
		Graph-level	Molecular property (e.g. fold classification, secondary structure prediction)
Molecular design	Graph Explainability Library (Agarwal et al. 2023)	Graph-level	Molecular mutagenic property, molecular functional group (e.g. benzine rings, fluoride carbonyl)
Neurology	NeuroGraph (Said et al. 2023)	Graph-level	Donor demographics (age and gender), task states (emotion processing, gambling, language, motor, relational processing, social cognition, and working memory), cognitive traits (working memory, fluid intelligence)
Therapeutic discovery	AVIDa-hIL6 (Tsuruta et al. 2024)	Edge-level	Antigen–antibody interaction
Therapeutic discovery	Therapeutic Data Commons (Huang et al. 2021)	Edge-level	Drug–target interaction, drug–drug interaction, protein–protein interaction, disease–gene association, drug–response prediction, drug–synergy prediction, peptide-MHC binding, antibody–antigen affinity, miRNA–target prediction, catalyst prediction, TCR–epitope binding, and clinical trial outcomes
		Graph-level	Molecular property (e.g. synthesizability, drug-likeness)

Databases are categorized by data type. The table is organized alphabetically by data type and database names.