Objectives

Although the global vaccination programme for COVID-19 is progressing, it will not eradicate the virus. It is vital that the medical community furthers its understanding of the pathobiology of COVID-19 and develops treatments. Since the onset of the pandemic over 40 000 articles have been published. To help researchers extract the information needed for drug development from this research corpus and to understand it in the context of other sources of biomedical data, we developed a unified knowledge representation.

Methods

Using the TypeDB database, we developed a knowledge graph of the biological domain, that through logical inference, enables new relationships between entities to be identified. Entities text-mined from the CORD-19 corpus by machine learning were semantically enriched using SemMedDB and ingested along with datasets from Uniprot, DGldb, Human Protein Atlas, Reactome, DisGeNET .

Results

We ingested 13 587 publications and identified 28 513 paragraphs that mentioned COVID-19-related entities of interest such as genes, proteins, drugs and coronaviruses. Across all data sources, 253 540 entities were included in our knowledge graph, resulting in 2 282 950 relations.

Conclusions

The Bio Covid knowledge graph enables users to query and analyse large amounts of data from structured and unstructured sources. Using classical artificial intelligence, connections between entities can be inferred, and in contrast to machine learning techniques (connectionist AI), the reasons and sources of this inference identified. Bio Covid will support study of the mechanisms of coronaviral infection, the immune response, and help find targets for the development of treatments more efficiently, thus providing clinical and economic benefits.