Integrating AI-powered text mining from PubTator into the manual curation workflow at the Comparative Toxicogenomics Database.

Abstract

The Comparative Toxicogenomics Database (CTD) is a manually curated knowledge- and discovery-base that seeks to advance understanding about the relationship between environmental exposures and human health. CTD's manual curation process extracts from the biomedical literature molecular relationships between chemicals/drugs, genes/proteins, phenotypes, diseases, anatomical terms, and species. These relationships are organized in a highly systematic way in order to make them not only informative but also scientifically computational, enabling inferential hypotheses to be formed to address gaps in understanding. Integral to CTD's functionality is the use of structured, hierarchical ontologies and controlled vocabularies to describe these molecular relationships. Normalizing text (i.e. translating raw text from the literature into these controlled vocabularies) can be a time-consuming process for biocurators. To facilitate the normalization process and improve the efficiency with which our scientists curate the literature, CTD evaluated and integrated into the curation process PubTator 3.0, a state-of-the-art, AI-powered resource which extracts and normalizes from the literature many of the key biomedical concepts CTD curates. Here, we describe CTD's long-standing history with Natural Language Processing (NLP), how this history helped form our objectives for NLP integration, the evaluation of PubTator against our objectives, and the integration of PubTator into CTD's curation workflow. Database URL: https://ctdbase.org.