A Collection of Benchmark Data Sets for Knowledge Graph-based Similarity in the Biomedical Domain.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Knowledge graph (KG) entity comparison is vital for data integration and machine learning.
• Biomedical KGs are crucial for predicting protein interactions, gene-disease associations, and protein localization.
• Evaluating KG semantic similarity measures is challenging due to the difficulty in creating gold standard datasets.

Purpose of the Study:

• To develop a comprehensive set of benchmark datasets for evaluating semantic similarity measures in biomedical knowledge graphs.
• To address the non-trivial challenge of building gold standard datasets for large-scale biomedical KGs.

Main Methods:

• Created 21 benchmark datasets by leveraging proxies for biomedical entity similarity.
• Utilized data from Gene Ontology and Human Phenotype Ontology.
• Explored proxy similarities based on protein sequence, protein family, protein-protein interactions, and phenotype-based gene similarity across four species.

Main Results:

• The developed datasets offer varying sizes and species coverage, with different annotation completion levels.
• Provided semantic similarity computations using state-of-the-art measures for each dataset.
• Facilitated a standardized evaluation framework for biomedical KG similarity methods.

Conclusions:

• The new benchmark datasets simplify the evaluation of knowledge graph-based semantic similarity measures in the biomedical domain.
• These resources enable more robust comparison and development of computational methods for biological data analysis.
• The study provides a valuable resource for researchers in bioinformatics and computational biology.