Work

Integrating ToxCast/ToxRefDB Data with Quantitative Adverse Outcome Pathway for Risk Assessment of Endocrine Disrupting Chemicals using Machine Learning

Year: 2026
Type: preprint
Abstract: Endocrine disrupting chemicals (EDCs) pose complex, multi-mechanistic health risks that are difficult to evaluate using conventional in vitro or in vivo approaches alone. Here, we developed a quantitative adverse outcome pathway (qAOP) model integrating ToxCast and ToxRefDB data to enable mechanism-based, high-throughput risk assessment of EDCs. Molecular initiating events (MIEs) for 40 representative EDCs were prioritized from ToxCast bioassay data and mapped onto the AOP wiki database, yielding six integrated AOP networks comprising 22 MIEs, 167 key events, and 40 adverse outcomes (AOs) spanning organ- and gender-specific endpoints. Chemical–MIE potencies were quantified using Toxicological Priority Index (ToxPi) scores derived from ToxCast AC50 values, while MIE–AO probabilistic relationships were weighted using the PageRank algorithm applied to AOP network topology. Matrix multiplication of these two probabilistic layers generated integrated toxicity scores predicting EDC–AO associations. Model performance was validated against in vivo ToxRefDB data using machine learning-based receiver operating characteristic (ROC) analysis, achieving AUC values of 0.86 and 0.73 for hepatotoxicity and reproductive toxicity, respectively. This data-driven qAOP framework provides a mechanistically transparent alternative to black-box QSAR models, bridges in vitro and in vivo toxicological evidence, and offers a scalable, animal-free strategy for prioritizing and screening emerging EDCs and structural analogues within Integrated Approaches to Testing and Assessment (IATA) frameworks.
Source: ChemRxiv
Institution: Kyoto University
Language: English
Open Access status: green
OpenAlex ID: W7169870959
Cites:
Cited by:
Related to: 9