Relation extraction between biomedical entities from scientific texts

The current growth in biomedical digital publications has made necessary the development of tools that are capable of synthetize the information. To this end, a specific task has become fundamental, the capability of extracting biomedical relations between named entities, such as genes, chemicals or diseases. BioRE (Biomedical Relation Extraction) has allowed investigators to build taxonomies based on biomedical relations, and facilitate the creation of ontologies and graph-based databases. Methods for extracting relations have evolved in recent years in a variety of approaches, from co-concurrence statistical methods and Rule-based methods, to Machine Learning and Deep Learning methods. Recent approaches are mainly based on pre training language models created from biomedical corpus. This work proposes BioREx a BioRE system based on the state-of-the-art, BioBert, which was fine-tuned with a dataset specifically adapted for this task. Our model is composed of two classifiers created from representations on biomedical data in to perform intra sentence relation classification between Genes-Diseases, Chemical-Disease and ChemicalsGenes. A web platform capable of extracting relations from biomedical text has been also created based on these classifiers. The BioRE solution was evaluated showing a F1 score of 82.37, 86.74 and 69.1 to classify Genes-Diseases, Chemical-Disease and Chemicals-Genes, respectively