Automated Mechanistic Literature Classification and Prioritization for Polychlorinated Biphenyls (PCBs)
On this page:
Mechanistic information is increasingly prominent in risk assessment activities and in reducing the use of vertebrate animals in research. We developed an automated keyword search approach to expedite identification, analysis, and prioritization of mechanistic literature, while minimizing bias and errors inherent to manual categorization. 7,890 keywords were curated to characterize 20 mechanistic categories. The approach was validated using a published database of 120 studies also subjected to manual categorization. Here, we customized the approach to also identify health outcome-specific mechanistic literature applied to a database of 69,587 studies containing search terms related to PCBs. Two additional keyword categories comprised of 233 and 67 keyword terms were developed to capture studies relevant for developmental exposure effects and placental effects, respectively. File preparation and run time for the algorithm required 20 hours, providing several orders of magnitude in time savings compared to manual review. The approach identified 1,371 studies with developmental terms and 857 studies with placental terms. The output enables analyses within and across keyword categories, providing instant insights into potential modes of action. For example, 138 and 16 studies containing developmental terms also included gene expression or epigenetic terms, respectively. Also, while placental and developmental studies shared top signal transduction pathway terms (MAP kinase, TGFß, VEGF, and insulin signaling), NRF2 pathway terms mapped only to developmental studies. These findings confirm the utility of this approach to identify and analyze mechanistic literature for very large databases and demonstrate the value of this approach for characterization of health outcome-specific mechanistic information.