COVID19
Computable knowledge extracted from the literature using machine reading can help researchers best understand and leverage the unprecedented volume of information gathered about the novel coronavirus. We hypothesize that machine interpretation techniques can be used to build graphical models of related concepts, with highly-connected nodes suggesting potentially plausible biological actors. We introduce a new resource, derived from the Semantic MEDLINE database (SemMedDB), reflecting documents also in the COVID-19 corpus. SemMedDB contains concept-relation-concept semantic triples, or predications. After extracting ~106K semantic predications, we imported these into a network and applied network centrality metrics (degree, closeness, betweenness) to identify and substantiate association factors related to COVID-19 for biological plausibility. Filtering the nodes by semantic type to search for drugs, drug targets, biomarkers, or comorbidities associated with complications, we were able to recapitulate agents already in randomized controlled trials for preventing or treating COVID-19 infections, comorbidities associated with lethal complications, many of which made sense upon further inspection. This guilt-by-association analysis demonstrates the value of the information revealed as computable knowledge by machine reading software.