We reconstructed hierarchical structures of Interlinking Ontology for Biological Concepts by life-science experts converting the skos:broader and dcterms:subject relations to rdfs:subClassOf relations. As a result, we could efficiently create more than 125,000 rdfs:subClassOf in the ontology, and we expect that the relations would be leveraged for inferring biological molecules' functions by the inheritance approach.
Interlinking Ontology for Biological Concepts (IOBC) is developed from Japan Scientific and Technological Thesaurus (JST thesaurus). It contains approximately 120,000 terms related to life science and chemical science. So far we sub-classified the skos:related relations within the IOBC to 35 relations such as SIO:000225 (has function) [1]. Consequently, we could infer biological and chemical functions to biological molecules using the sub-classified and sub-class-of relations of IOBC [2]. However the hierarchical structure has been described using skos:broader and dcterms:subject without distinguishing some ontological relationships such as sub-class-of, and is-part-of because it was developed from the thesaurus. In this study, we attempted to reconstruct the hierarchical relations for the purpose of more rigorously describing the relationships among biological concepts and inferring biological molecules’ functions at scale.
In order to improve the hierarchical structures of IOBC, four lifesciences experts attempted the followings. (1) The experts considered separating each “skos:broader” relation into each of “rdfs:subClassOf”, and “SIO:000068(is part of)” because they were represented by skos:broader in IOBC. In addition, (2) the experts considered converting each “dcterms:subject” relation to “rdfs:subClassOf,” because some upper concepts were represented by dcterms:subject instead of rdfs:subClassOf in IOBC. Furthermore, they examined whether there was more appropriate term upper concept for each rdfs:subClassOf in IOBC. As the result, the experts defined 35 newly categorized terms as upper concepts. The newly defined terms include “biological phenomenon, process, and state,” “substance”, etc.
In the above reconstruction processes, each relation was decided by 3 experts’ majority rule, and discussion, and the ontological consistency was finally confirmed by an expert.
We separated 147,460 skos:broader into 110,582 rdfs:subClsssOf, and 135 SIO:000068 (is part of) (1-1 and 1-2 in Table.1). Meanwhile, we converted 118,355 dcterms:subject to 15,092 rdfs:subClassOf (2 in Table.1) . This ontology is available in BioPortal (http://purl.bioontology.org/ontology/IOBC). Table.1 Reconstruction examples of the hierarchical structures # Before reconstruction After reconstruction 1-1 erythrocyte aggregation (skos: erythrocyte aggregation (rdfs:subClassO broader) hemagglutination f) hemagglutination 1-2 antigen recognition (skos:broa antigen recognition (SIO:000068 (is part der) antigen-antibody reaction of)) antigen-antibody reaction 2 cell aggregation (dcterms:subje cell aggregation (rdfs:subClassOf) biolog ct) Immunology ical phenomenon, process, and state
The expert curation greatly and exactly increasing the number of rdfs:subClassOf, and SIO:000068 (is part of) relations in the IOBC, we can more largely and accurately discover new functions, and roles of biological concepts such as chemical compounds and gene products using the is-a hierarchy and the part of structure, and combinating the IOBC, with other ontologies, such as ChEBI.