Extraction and analysis of the structure of labels in biomedical ontologies

Abstract

The increasing interest in biomedical ontologies has provoked the development of a significant number of ontologies, and many more are expected to be produced in the near future. A significant proportion of such ontologies have not been created by computer scientists or ontology engineers, but by domain experts. Many such ontologies are rich in implicit knowledge, but are really just plain taxonomies and controlled vocabularies, with little axiomatization. Many of these ontologies have much information within the labels of the classes. There is a great deal of knowledge about the entities described within such labels and text definitions held on classes; these are useful for human users, but not for machine processing. In previous work we proposed a process for enriching ontologies, which included the analysis of such labels, the identification of lexical patterns and the design of corresponding knowledge patterns. However, this process relied on manual intervention. In this paper we present a method to analyze and extract unused information contained in the structure of the labels in biomedical ontologies. The aim of this method is to improve the source ontology. The first step is the identification of lexical patterns based on repetitions of sets of words. Second, such lexical patterns will be examined in existing biomedical ontologies to identify whether those patterns are referencing existing ontological entities. Finally, the results obtained with relevant biomedical ontologies are presented and discussed.