Multi-domain semantic similarity in biomedical research

Area of Science:

• Biomedical Informatics
• Computational Biology
• Data Science

Background:

• Increasingly, biomedical resources are annotated using multiple ontologies across diverse knowledge domains.
• Existing semantic similarity measures often fail to account for this multi-domain nature of annotations.
• Comparing resources like metabolic pathways requires simultaneous consideration of multiple annotation types (e.g., enzymes and metabolites).

Purpose of the Study:

• To develop and evaluate novel methods for multi-domain semantic similarity.
• To enhance the comparison of biomedical resources annotated with concepts from multiple ontologies.
• To improve the prediction of new annotations by leveraging multi-domain information.

Main Methods:

• Proposed two approaches: an aggregative method and an integrative method.
• The aggregative approach calculates domain-specific similarities and averages them.
• The integrative approach merges multiple ontologies into a single, unified ontology for similarity calculation.

Main Results:

• Evaluated approaches on a multidisciplinary epidemiology dataset.
• Demonstrated a significant increase in performance for multi-domain measures over single-ontology measures.
• The proposed methods showed improved capacity to predict new annotations based on existing ones.

Conclusions:

• Multi-domain semantic similarity measures offer superior performance compared to single-domain measures.
• These novel approaches provide a promising foundation for future research in multi-domain similarity.
• The findings suggest the community should consider adopting multi-domain measures for resource comparison.