Lessons learned from cross-validating alignments between large anatomical ontologies.

Objectives: To compare the alignments of two large anatomical ontologies (the Foundational Model of Anatomy and GALEN) produced by three ontology alignment systems (AOAS, FALCON and PRIOR) in the framework of the Ontology Alignment Evaluation Initiative during its 2006 campaign.

Materials: Number of mappings identified by AOAS: 3,132, FALCON: 2,518 and PRIOR: 2,589.

Methods: Three approaches to analyzing and comparing the results were utilized: computing the overlap among result files, manual review of some 2,000 mappings and structural validation.

Conclusions: The generic systems FALCON and PRIOR identify many false positives and false negatives. With a stricter and domain-specific lexical similarity model, AOAS has a better precision, but is more sensitive to missing synonyms and misspellings.