Information from semantic integration of texts and databases
Relations mined from texts and structured information from databases have been mapped to concepts defined in biomedical ontologies and to a predicate dictionary. Concepts and predicates are represented by nodes and edges in this graph and can be queried for relations between concepts. The graph combines relations extracted from Medline abstracts with relations obtained from the UMLS and databases as UniProt, EntrezGene, Comparative Toxicogemics Database, and from the datasets from the Linked Open Drug Data (Drugbank, DailyMed, and Sider). The approach was tested on 61 cerebral spinal fluid and 207 serum compounds of migraine patients. A cloud of all biomedical concepts related to the concept migraine in this graph was used to construct a set of cerebral spinal fluid compound concepts and a set of serum compound concepts. For each of the relations in the cloud provenance is available and provided. These sets were evaluated against two manually created sets of compounds. The evaluation showed that this graph based method retrieves relevant compounds with mean average precision values of 0.32 and 0.59, respectively.
|Graph databases, Medline, Relation mining|
|8th International Conference on Semantic Web Applications and Tools for Life Sciences, SWAT4LS 2015|
|Organisation||Erasmus MC: University Medical Center Rotterdam|
Van Mulligen, E.M, Vlietstra, W.J, Vos, R, & Kors, J.A. (2015). Information from semantic integration of texts and databases. Presented at the 8th International Conference on Semantic Web Applications and Tools for Life Sciences, SWAT4LS 2015. Retrieved from http://hdl.handle.net/1765/82614