Fig. 1

KGEV framework stages. Stage 1: A literature mining tool is used to extract relationships from texts such as biomedical literature or clinical notes to produce triples (a head and tail entity linked by a relationship). In our COVID-19 case study, we used SemRep to extract triples from CORD-19 abstracts, which contains literature related to COVID-19. Stage 2: The extracted triples are filtered to keep only high-quality triples. In our case, we filtered by the SemRep confidence score and the number of times a triple appears in the data. We also normalized COVID-19-/SARS-CoV-2-related entities to a small number of topics. The triples from this step are used to construct the initial knowledge graph. Stage 3: Metadata is extracted from the texts that the triples originated from. This metadata, which includes information on the immediate text that supports the extracted triple and document-level information, is stored in a database. Stage 4: Data from other sources are integrated into the KG to fill in the gaps of the triples mined from the texts. In our COVID-19 case study we used eight additional data sources (see Methods for details). Stage 5: The integrated triples are then formatted and stored in the database along with the metadata information. In our case, we used Neo4j as the database because its graphical representation of the data naturally fits the KG data structure