Ontology portal

Abstract

Biological curation is the cornerstone of modern Computational Biology. Specialised Biocurators can be assimilated to the museum catalogers in the world of Computational Biology. They turn unidentifiable objects from unwieldly mediums into a powerful model from which researchers can benefit from (Bourne and McEntyre, 2006). Whist this process alone would already be a significant scientific contribution, the task of a biocurator is even more comprehensive. Researchers would want to add their findings to the exhibt; or pick aspects of this exhibit to modify for their own personal research with minimal effort. Because of this Biocurators need to be specialised in a multitude of different domains to do this indispensable, non-trivial task. In this project we explore what it takes to build an automated system that can facilitate this process. We hypothesise that using text mining algorithms and ontologies we can describe complex biological interactions found in literature. By using ontologies stored in a Neo4j graph we built a Knowledge Graph of relations mined from PubMed abstracts. Despite our model being capable of multiple interaction types, we evaluated our experiment using the CDR gold standard created for BioCreative V. We also asked the help of a domain expert to evaluate the relations obtained when running the system on full text papers. Our results obtained were encouraging and show that the system has potential if more effort is invested. We managed to obtain an F-score of 25% on the CDR dataset which is comparable to other systems designed specifically for chemical disease relationships.