Exploiting Semantic Predications in a Graph Database

Knowledge extraction using semantic relations is crucial for accurate and valid knowledge management in biomedicine. The Semantic MEDLINE Database (SemMedDB) contains semantic predications extracted with the SemRep semantic interpreter. Predications are structured as subject-predicate-object triples and can be represented as a directed network. Arguments correspond to UMLS Metathesaurus concepts, while predicates correspond to relations in the UMLS Semantic Network (e.g., SemRep extracts the predication Ethionine-CAUSES-Lesion from the sentence Ethionine can trigger lesions.). SemMedDB has predications from all of MEDLINE and is available as a MySQL database. MySQL is generally efficient, but modeling networks using a relational database causes a large number of many-to-many relations. Complex join queries are then needed to retrieve such data. In this poster we present the Neo4j graph database as an alternative for storing, retrieving, and mining SemMedDB.