Aligning relationships in the UMLS: Methods and Preliminary Results

Dr. Olivier Bodenreider
Lister Hill Nagional Center, National Library of Medicine


Abstract

Unlike the categorization link defined between Metathesaurus concepts and Semantic Network types, no such correspondence exists between the relationships at these two levels of the Unified Medical Language System (UMLS). We present and discuss two methods for aligning Metathesaurus relationships with their counterparts in the UMLS Semantic Network. The first approach attempts to elicit the semantics of Metathesaurus relationships through an examination of their relata at different levels: concept, high-level ancestors and semantic types. The second approach examines the frequency of association between a given Semantic Network relationship and the actual relationships observed in the Metathesaurus between the concepts categorized by these semantic types. Of the 139 relationships present in the Metathesaurus, 80 (58%) could be aligned with Semantic Network relationships. The remaining relationships are vocabulary internal, used, for example, for vocabulary management or to indicate strictly lexical relationships. This approach is a preliminary step towards the development of a comprehensive ontology of relationships for the biomedical domain.