You are here:
New features in ToxRefDB to improve modeling applications and data integration
Citation:
Watford, S., L. Pham, J. Wignall, R. Shin, M. Martin, AND K. Paul-Friedman. New features in ToxRefDB to improve modeling applications and data integration. Presented at Society of Toxicology, San Antonio, Texas, March 11 - 15, 2018. https://doi.org/10.23645/epacomptox.6834827
Impact/Purpose:
The objective of this work was to improve ToxRefDB as a predictive modeling resource. Information was added to the database to allow the user to infer "true negatives," i.e., tested but negative, and also to allow the user to connect pathological information in ToxRefDB with CDISC (pathology standards used by the FDA) and MeSH terms. The resultant resource is more available for statistical and predictive modeling, as well as integration with other data resources.
Description:
The Toxicity Reference Database (ToxRefDB) contains publicly available in vivo toxicity studies conducted largely to Environmental Protection Agency (EPA) guidelines, with information for >1000 chemicals and >5000 studies. ToxRefDB has been utilized for training and testing of alternative models yet is a “positives”-only resource, i.e. only effects significantly different from controls were included. The lack of “true negatives” limited available balanced datasets for modeling. Other impediments to modeling have included a lack of quantitative dose-response information and controlled vocabulary for in vivo effects. The objective of this work was to address these challenges to improve ToxRefDB as a predictive modeling resource. First, “guideline profiles” where endpoints are annotated as either required, triggered, recommended, or not required/mentioned were developed based on the EPA 870 series guidelines. Thus, “true negatives” can be inferred and distinguished from “not tested” endpoints. Quantitative data extraction has been completed for subchronic and chronic studies, enabling statistical modeling on nearly 400 endpoints. In this process, units were reviewed and standardized to enable comparison, decreasing the total unit number from >800 to <400. Finally, the lack of controlled vocabulary for in vivo effects has led to challenges in continued data extraction, QA/QC, and integration with other resources. As a solution, ToxRefDB vocabulary was standardized to reflect the language used in corresponding EPA 870 series guidelines and cross-referenced to the National Cancer Institute Thesaurus (NCIt). These cross-references are based on Clinical Data Interchange Standards Consortium (CDISC), an international effort to describe clinical and non-clinical data. NCIt cross-references enable mapping of in vivo pathological effects from ToxRefDB to PubMed (via MeSH terms) and subsequent genes (and other information) that may be relevant for toxicological research. These new additions to ToxRefDB increase the potential for connections between ToxRefDB and other resources and critically support toxicology modeling applications. This abstract does not necessarily reflect U.S. EPA policy.
URLs/Downloads:
DOI: New features in ToxRefDB to improve modeling applications and data integration
SOT2018_POSTER_CLEARED.PDF (PDF, NA pp, 1155.244 KB, about PDF)