Improving Biomedical Knowledge Graph Quality: A Community Approach

Area of Science:

• Biomedical Informatics
• Data Science
• Knowledge Representation

Background:

• Biomedical knowledge graphs (KGs) are crucial tools in research and translation.
• Existing KGs often lack consistent design, documentation, and dissemination practices, unlike ontologies.
• This opacity limits their external reuse and comparability.

Purpose of the Study:

• To establish and apply a set of evaluation criteria for biomedical KGs.
• To assess the transparency and reusability of existing biomedical KGs.
• To identify areas for standardization and improvement in KG development.

Main Methods:

• Developed evaluation criteria based on data standards and principles from related fields.
• Applied these criteria to a diverse set of 16 biomedical KGs.
• Analyzed variations in KG models, source integration, and terminology.

Main Results:

• Many biomedical KGs, even those appearing to follow best practices, obscure essential information for reuse.
• Significant variations exist in KG models, data integration, and node type terminology across different KGs.
• Lack of transparency is a common issue, impeding external use.

Conclusions:

• Community-wide adoption of shared criteria and standards (e.g., BioLink, KGX) is essential.
• Improved transparency and standardization will enhance KG reusability and comparability.
• Maturing KG standards is critical for maximizing their utility in biomedical research.