Towards automatic classification within the ChEBI ontology

Abstract

Abstract *Background* Appearing in a wide variety of contexts, biochemical 'small molecules' are a core element of biomedical data. Chemical ontologies, which provide stable identifiers and a shared vocabulary for use in referring to such biochemical small molecules, are crucial to enable the interoperation of such data. One such chemical ontology is ChEBI (Chemical Entities of Biological Interest), a candidate member ontology of the OBO Foundry. ChEBI is a publicly available, manually annotated database of chemical entities and contains around 18000 annotated entities as of the last release (May 2009). ChEBI provides stable unique identifiers for chemical entities; a controlled vocabulary in the form of recommended names (which are unique and unambiguous), common synonyms, and systematic chemical names; cross-references to other databases; and a structural and role-based classification within the ontology. ChEBI is widely used for annotation of chemicals within biological databases, text-mining, and data integration. ChEBI can be accessed online at "http://www.ebi.ac.uk/chebi/":http://www.ebi.ac.uk/chebi/ and the full dataset is available for download in various formats including SDF and OBO.*Automated Classification*The selection of chemical entities for inclusion in the ChEBI database is user-driven. As the use of ChEBI has grown, so too has the backlog of user-requested entries. Inevitably, the annotation backlog creates a bottleneck, and to speed up the annotation process, ChEBI has recently released a submission tool which allows community submissions of chemical entities, groups, and classes. However, classification of chemical entities within the ontology is a difficult and niche activity, and it is unlikely that the community as a whole will be able or willing to correctly and consistently classify each submitted entity, creating required classes where they are missing. As a result, it is likely that while the size of the database grows, the ontological classification will become less sophisticated, unless the classification of new entities is assisted computationally. In addition, the ChEBI database is expecting substantial size growth in the next year, so automatic classification, which has up till now not been possible, is urgently required. Automatic classification would also enable the ChEBI ontology classes to be applied to other compound databases such as PubChem. *Description Logic Reasoning*Description logic based reasoning technology is a prime candidate for development of such an automatic classification system as it allows the rules of the classification system to be encoded within the knowledgebase. Already at 18000 entities, ChEBI is a fair size for a real-world application of description logic reasoning technology, and as the ontology is enhanced with a richer density of asserted relationships, the classification will become more complex and challenging. We have successfully tested a description logic-based classification of chemical entities based on specified structural properties using the hypertableaux-based HermiT reasoner, and found it to be sufficiently efficient to be feasible for use in a production environment on a database of the size that ChEBI is now. However, much work still remains to enrich the ChEBI knowledgebase itself with the properties needed to provide the formal class definitions for use in the automated classification, and to assess the efficiency of the available description logic reasoning technology on a database the size of ChEBI's forecast future growth.*Acknowledgements*ChEBI is funded by the European Commission under SLING, grant agreement number 226073 (Integrating Activity) within Research Infrastructures of the FP7 Capacities Specific Programme, and by the BBSRC, grant agreement number BB/G022747/1 within the “Bioinformatics and biological resources” fund.