From terms to categories: Testing the significance of co-occurrences between ontological categories

The co-occurrence of terms in a text corpus may indicate the presence of a relation between the referents of these terms. We expect co-occurrence-based methods to identify association relations that cannot be found using static patterns. We developed a new method to identify associations between ontological categories in text using the co-occurrence of terms that designate these categories. We use the taxonomic structure of the ontologies to cumulate the number of co-occurrences of terms designating categories. Based on these cumulated values, we designed a novel family of statistical tests to identify associated categories. These tests take both co-occurrence specificity and relevance into consideration. We applied our method to a 2.2 GB text corpus containing fulltext articles and used Gene Ontology's biological process ontology and the Celltype Ontology. The software and results can be found at http://bioonto.de/pmwiki.php/Main/ExtractingBiologicalRelations.