Building a biomedical ontology recommender web service.

Area of Science:

• Biomedical Informatics
• Data Science
• Bioinformatics

Background:

• Ontologies and terminologies are crucial for annotating biomedical data to enable data integration and translational discoveries.
• Selecting the most suitable ontology for specific datasets is a significant challenge due to the vast number and variety of available biomedical ontologies.
• Researchers often find it difficult and time-consuming to identify appropriate ontologies for their data annotation needs.

Purpose of the Study:

• To develop and present a web service, the Biomedical Ontology Recommender, to assist researchers in selecting appropriate ontologies for annotating biomedical datasets.
• To provide an automated solution for identifying the best-suited ontologies based on dataset metadata or keywords.
• To streamline the process of ontology selection for improved data annotation and analysis.

Main Methods:

• The Biomedical Ontology Recommender service utilizes textual metadata or keywords to suggest relevant ontologies.
• Ontology selection is based on three criteria: coverage (term overlap), connectivity (mapping frequency), and size (number of concepts).
• The service scores ontologies using annotations generated by the National Center for Biomedical Ontology (NCBO) Annotator, drawing from UMLS Metathesaurus and NCBO BioPortal ontologies.

Main Results:

• A functional comparison with existing methods demonstrated the effectiveness of the recommender system.
• Evaluation using real-world use cases showed that recommendation heuristics based on coverage and connectivity were rated 'very relevant' by expert evaluators.
• The best-performing heuristics leverage both the coverage and connectivity of ontologies for accurate recommendations.

Conclusions:

• The Biomedical Ontology Recommender service effectively addresses the challenge of selecting appropriate ontologies for biomedical data annotation.
• The system's reliance on coverage and connectivity criteria provides highly relevant recommendations, as validated by expert evaluation.
• The alpha version of the recommender service is publicly available and integrated into the NCBO BioPortal, facilitating its use by the research community.