We propose a silver standard based on the UMLS Metathesaurus to align NCI, FMA and SNOMED CT. This silver standard aims at being exploited within the OAEI and SEALS Campaigns. The UMLS Metathesaurus (UMLS-Meta) [1] is currently the most comprehensive e ort for integrating independently-developed medical thesauri and ontologies. UMLS-Meta is being used in many applications, including PubMed and ClinicalTrials.gov. The integration of new UMLS-Meta sources combines automatic techniques, expert assessment, and auditing protocols (see [2] for a review of current methods). In its 2009AA version, UMLS-Meta integrates more than one hundred thesauri and ontologies, including SNOMED CT, FMA and NCI, and contains more than 6 million entities. UMLS-Meta provides a list with more than two million unique identi ers (CUIs). Each CUI can be associated to entities belonging to di erent sources. Pairs of entities from di erent sources with the same CUI are synonyms and hence can be represented as an equivalence mapping. Thus, UMLS-Meta mappings could be considered as a silver standard to align ontologies such as SNOMED CT, FMA or NCI. The Ontology Alignment Evaluation Initiative (OAEI) could be bene ted from UMLS-Meta so that it could be used as the input dataset for a new challenging track within the evaluation campaign. However, in our previous work [3, 4], we showed that UMLSMeta contains a signi cant number of logic errors when the rich semantics of the ontology sources is taken into account together with the UMLS-Meta mappings.
Our experiments were based on the UMLS-Meta version 2009AA and the
corresponding versions of FMA (version 2.0), NCI (version 08.05d) and SNOMED CT
(version 20090131), which contain 66,724, 78,989 and 304,802 entities,
respectively. After extracting the relevant parts of UMLS, we obtained 3,024 mapping
axioms between FMA and NCI, 9,072 between FMA and SNOMED CT and
19,622 between SNOMED CT and NCI. Note that mappings are considered as
OWL 2 axioms (see [
When reasoning over each of the source ontologies independently, all their entities were found satis able. However, after the respective integrations via UMLS-Meta mappings, we obtained a huge number of unsatis able entities, namely 5,015 when integrating FMA and NCI, 16,764 with FMA and SNOMED CT, and 76,025 with SNOMED CT and NCI.
We designed three logic-based principles [
UMLS-Meta represents a reference of correspondences among ontologies, however, the direct integration of these ontologies with UMLS-Meta mappings leads to a huge number of unintended logical consequences. We propose instead a revised set of UMLS-Meta mappings which could be considered as a silver standard to evaluate ontology matching techniques over FMA, NCI and SNOMED CT. The proposed silver standard and related sources are available for download in http://krono.act.uji.es/people/Ernesto/umlsassessment.
The silver standard could be improved by ontology matching tools if they are able to nd new valid correspondence which do not lead to logic errors. We also intend to design less aggressive techniques in order to preserve the maximum number of UMLS-Meta mappings. Additionally, ontology sources could be revised before the automatic assessment in order to smooth the incompatibilites.