Giles, O. | bioRxiv Science

bioRxiv2020

Reference ontology and database annotation of the COVID-19 Open Research Dataset (CORD-19)

The COVID-19 Open Research Dataset (CORD-19) was released in March 2020 to allow the machine learning and wider research community to develop techniques to answer scientific questions on COVID-19. The dataset consists of a large collection of scientific literature, including over 100,000 full text papers. Annotating training data to normalise variability in biological entities can improve the performance of downstream analysis and interpretation. To facilitate and enhance the use of the CORD-19 data in these applications, in late March 2020 we performed a comprehensive annotation process using named entity recognition tool, TERMite, along with a number of large reference ontologies and vocabularies including domains of genes, proteins, drugs and virus strains. The additional annotation has identified and tagged over 45 million entities within the corpus made up of 62,746 unique biomedical entities. The latest updated version of the annotated data, as well as older versions, is made openly available under GPL-2.0 License for the community to use at: https://github.com/SciBiteLabs/CORD19

immunology↗

bioRxiv2020

Optimising biomedical relationship extraction with BioBERT: Best practices for data creation

Text mining is widely used within the life sciences as an evidence stream for inferring relationships between biological entities. In most cases, conventional string matching is used to identify cooccurrences of given entities within sentences. This limits the utility of text mining results, as they tend to contain significant noise due to weak inclusion criteria. We show that, in the indicative case of protein-protein interactions (PPIs), the majority of sentences containing cooccurrences ({bsim}75%) do not describe any causal relationship. We further demonstrate the feasibility of fine tuning a strong domain-specific language model, BioBERT, to analyse sentences containing cooccurrences and accurately (F1 score: 88.95%) identify functional links between proteins. These strong results come in spite of the deep complexity of the language involved, which limits the accuracy even of expert curators. We establish guidelines for best practices in data creation to this end, including an examination of inter-annotator agreement, of semisupervision, and of rules based alternatives to manual curation, and explore the potential for downstream use of the model to accelerate curation of interactions in the SIGNOR database of causal protein interactions and the IntAct database of experimental evidence for physical protein interactions.

bioinformatics↗