Linking AhR and Nrf2 activation to neurotoxicity through adverse outcome pathway-based machine learning.

Chemical-induced neurotoxicity poses a challenge for safety assessment due to the complexity of central nervous system responses and the limitations of conventional assays. We developed an adverse outcome pathway (AOP)-guided machine-learning framework that links molecular initiating events, using the activation of the aryl hydrocarbon receptor (AhR) and Nrf2-mediated oxidative stress responses as examples, to neurotoxicity. A curated dataset was assembled from public sources and the literature, key molecular features such as nitrogen-containing groups were identified as structural alerts associated with neurotoxicity. Within this framework, AhR and Nrf2 activity information (measured or predicted) was integrated with structural fingerprints to enhance biological interpretability while maintaining robust predictive performance (e.g., AUC value > 0.80) in both cross-validation and external validation. The AOP-based model was further applied to virtual screening of 7576 compounds, with blood-brain barrier (BBB) permeability and applicability domain (AD) constraints ensuring reliable predictions. Molecular docking of 12 prioritized persistent organic pollutants (POPs) further supported these predictions, revealing strong binding affinities with AhR (up to -10.2 kcal/mol) and Nrf2 (up to -7.9 kcal/mol) through interactions such as hydrogen bonding and π-π stacking. Overall, this strategy highlights AOP-based machine learning as a powerful approach to connect structure, mechanism, and outcomes, thereby improving chemical risk assessment and prioritization.