Improving the Data Quality of Drug Databases using Conditional Dependencies and Ontologies

International audienceMany health care systems and services exploit drug related information stored in databases. The poor data quality of these databases, e.g. inaccuracy of drug contraindications, can lead to catastrophic consequences for the health condition of patients. Hence it is important to ensure their quality in terms of data completeness and soundness. In the database domain, standard Functional Dependencies (FDs) and INclusion Dependencies (INDs), have been proposed to prevent the insertion of incorrect data. But they are generally not expressive enough to represent a domain-specific set of constraints. To this end, conditional dependencies, i.e. standard dependencies extended with tableau patterns containing constant values, have been introduced and several methods have been proposed for their discovery and representation. The quality of drug databases can be considerably improved by their usage. Moreover, pharmacology information is inherently hierarchical and many standards propose graph structures to represent them, e.g. the Anatomical Therapeutic Chemical classification (ATC) or OpenGalen's terminology. In this article, we emphasize that the technologies of the Semantic Web are adapted to represent these hierarchical structures, i.e. in RDFS and OWL. We also present a solution for representing conditional dependencies using a query language defined for these graph oriented structures, namely SPARQL. The benefits of this approach are interoperability with applications and ontologies of the Semantic Web as well as a reasoning-based query execution solution to clean underlying databases