A machine-built self-updating model of COVID-19 mechanisms

The Ecosystem of Machine-maintained Models with Automated Analysis (EMMAA, emmaa.indra.bio) is a framework for automatically building a set of disease-related models and keeping them up to date using the latest results from the scientific literature. EMMAA uses the INDRA system (indra.bio) to run multiple machine reading systems on literature relevant for a model, and extract molecular and higher-level causal mechanisms from these publications. During daily update cycles, extractions from newly published literature are aligned with the existing model and are used to corroborate existing relations or extend the model with new relations. Using INDRA assembly, the EMMAA COVID-19 model combines mechanisms from structured sources (Reactome, VirHostNet, DrugBank, CTD, etc.) with ones obtained from existing literature on coronaviruses, as well as the emerging body of new literature on COVID-19/SARS-CoV-2. The EMMAA dashboard allows exploring and curating the content of the COVID-19 model, with each mechanism shown in the context of the specific sentences from which it was derived, and linked back to the underlying literature.

A key feature of EMMAA is its support for user queries of each maintained model. Through a structured query interface, users can run queries representing questions such as “how does losartan affect ACE2?” or “what are the upstream small-molecule inhibitors of TMPRSS2?” against the COVID-19 model. EMMAA generates mechanistic paths to answer these queries, with each relation in the path annotated with literature or database evidence. Further, users can register queries to get daily email updates on any new results pertaining to their queries.

We aligned the EMMAA COVID-19 model with the COVID-19 Disease Map model (https://covid.pages.uni.lu/) by finding overlaps between entities (human or viral proteins, small molecules, biological processes, etc.) and their relations across the two models. This alignment can be used in multiple ways to accelerate and enrich the manual curation of the COVID-19 Disease Map model. First, the EMMAA model provides literature references, with specific evidence sentences to annotate existing links in the Disease Map model. Second, the EMMAA model can be used to find entities and relations not yet included in the Disease Map model in a way that can be prioritized based on the underlying evidence. Finally, EMMAA quantifies the contribution of specific new publications to the COVID-19 model, and this information facilitates the triaging of new literature (about 300 new COVID-19 publications each day) to focus expert curation.

These results are accessible publicly at https://covid19.indra.bio.