Enriching SARS-CoV-2 sequence data in public repositories with information extracted from full text articles

Project Summary In response to the COVID-19 pandemic, scientists have published over one hundred thousand research articles and made available over eight hundred thousand virus genome sequences. These sequences, along with their metadata, can be used to understand virus evolution and spread and their implications for public health, a field of study called genomic epidemiology. However, these sequence records do not typically contain patient metadata such as demographics, clinical severity, or comorbidities, preventing researchers from uncovering trends in population health. To understand the severity of the problem, we analyzed nearly 748 thousand SARS-CoV-2 records from GISAID and 60 thousand from GenBank for the presence of patient metadata finding age and gender were represented in < 1% of GenBank records and in GISAID, 26% included sex, and 24% had age. For other fields, the amount of missing data is even more pronounced, with neither resource providing information on a patient's race and only GISAID specifying severity (i.e. ICU) in less than 5% of records. To address missing virus metadata, researchers could utilize the publication associated with the new sequences, however, the virus sequence record is often never updated with a link to the publication. From the set of records that we analyzed, 3.4% (of 748K) in GISAID and < 1% (of 117K) in GenBank had a link to a publication. This greatly hinders secondary data analysis of these sequences and limits the ability to use them at scale to uncover associations between the viral genome, transmission risk, and health outcomes. The goal of this proposal is to enhance genomic epidemiology and population health of COVID-19 with a framework to continuously and automatically enrich SARS-CoV-2 nucleic acid sequence metadata in public databases such as GenBank and GISAID with metadata in associated published articles. We will incorporate input from clinicians at the front-line of patient care during the pandemic and build on our NIH funded work (R01AI117011), which used Natural Language Processing (NLP) to enrich the geographic metadata of a sequence record using its corresponding published article. We have used these data in virus phylogeographic models and shown the benefit of using enriched metadata for modeling virus evolution and spread. Theavailability of SARS-CoV-2 sequences, paired withfull- text COVID-19 articles and preprints, presents an opportunity for metadata enrichment and scientific discovery beyond our prior work. Our specific aims are to: (1) enrich SARS-CoV-2 sequence metadata using text extracted from publications and (2) derive key epidemiologic insights for different patient demographics using our enriched SARS-CoV-2 sequence dataset. We will leverage our prior joint work funded by the NIH to enable the secondary use of enriched metadata for genomic epidemiology to improve our understanding of SARS-CoV-2 evolution and spread among different population groups. We will disseminate the enriched data through our GeoBoost2 data dashboard, GenBank LinkOut and the i2b2 platform. The latter will more immediately allow integration with COVID-specific clinical data shared by the 4CE Consortium.