Esta web utiliza cookies. Si continúas navegando consideramos que aceptas su uso.
Más información
Aceptar
I Jornadas Científicas del IMIB-Arrixaca
Acceso Personal
Contacto
Aviso Legal
Inicio
Bienvenida
Comités
Ponencias
Comunicaciones
Programa
Fechas clave
Inscripciones
Selección comunicaciones
Enviar comunicación
Patrocinadores
Sede
Imprimir
SUGGESTING MISSING RELATIONS IN BIOMEDICAL ONTOLOGIES BASED ON LEXICAL REGULARITIES
Autores:
MANUEL QUESADA MARTÍNEZ
,
JESUALDO TOMÁS FERNÁNDEZ BREIS
, DANIEL KARLSSON,
Grupos de investigación:
[GI/IMIB/E170/2011] TECNOLOGÍAS DE MODELADO, PROCESAMIENTO Y GESTIÓN DEL CONOCIMIENTO
Comunicación:
Antecedentes:
The number of biomedical ontologies has increased significantly in recent years. Ontologies provide a set of concepts, terms, and relations that describe the semantics of a particular biomedical domain. Biomedical ontologies have hidden semantics, which means that they are rich in content in natural language, that is, for humans, but the content that can be understood by machines is smaller. Biomedical ontologies are more and more used for the codification of clinical data. SNOMED CT is an example of such resources, and it is considered a common reference for health terms used to annotate Electronic Health Records (EHRs). The fact that the machine-processable content is small also limits the exploitation of the clinical data by means of machines.
Métodos:
Our objective is to take advantage of the hidden semantics in biomedical ontologies for analysing and improving the machine-processable content. We have developed a method that applies the “lexically suggest, logically define” principle. This principle means that what is expressed in natural language for humans in the ontologies should be also expressed in the form of logical axioms for machines. The method uses natural language processing techniques to extract and analyse the textual content associated with the terms of biomedical ontologies. The method identifies and analyses lexical regularities (LRs), which are groups of consecutive tokens appearing in several ontology terms. The output of the method contains those LRs associated with terms that do not satisfy the “lexically suggest, logically define” principle, what is measured by the logical relation metric. Small values of this metric suggest that logical relations might be missing.
Resultados:
The method is implemented as part of the OntoEnrich framework (http://sele.inf.um.es/ontoenrich). This tool aims to automate the application of the method making the analysis of the results easier for biomedical experts. We have applied the method to a SNOMED-CT module (July 2015 version) of 18440 terms and 18443 formal relations. 19774 LRs were obtained, and the logical relation of 585 was different from 1. The median value of the logical relation in absolute terms is 1, which is a good indicator; more than 50% of the LRs capture just 1 missing logical relation. In particular, the highlighted LR “congenital stenosis” is further analysed as it was declared as a potential issue by SNOMED-CT developers (“artf229197”). This LR exhibits 42 terms, but roughly 50% of them are not related with the term “Congenital stenosis (morphologic abnormality)”. This means that a query to retrieve the patients whose disease has associated morphology Congenital stenosis would exclude many patients of interest, so potentially producing many false negatives.
Conclusiones:
Our method detects LRs that capture lexical relations not logically defined. Its application to a SNOMED-CT module identified 585 LRs with missing logical relations. The “congenital stenosis” LR automatically captured a problem previously reported by the SNOMED-CT team. This makes us think that the analysis of the other 584 cases might contribute to SNOMED-CT Quality Assurance. A further validation of the results with domain experts is currently in process.
Instituto de Investigación Sanitaria Acreditado
Inicio
Grupo de Investigación
Miembros
Proyectos
Colaboraciones
Servicios
Recursos formativos
Producción Científica
Publicaciones
Tesis
Novedades
Noticias
Eventos
Convocatorias
Agenda