Development, validation and application of ERα- and PPARγ-mediated female reproductive toxicity prediction models...

Traditional methods for reproductive toxicity assessment are insufficient for high-throughput chemical screening. Although efforts have been made in recent years to develop predictive models for reproductive toxicity, in-depth modeling of female-specific multi-pathway mechanisms has been lacking. In this study, we developed a female reproductive toxicity prediction model, FeRetoxML, based on adverse outcome pathway (AOP) mediated by estrogen receptor α (ERα) and peroxisome proliferator-activated receptor γ (PPARγ), employing a closed-loop strategy of model development, validation and application, which is the first model established to link PPARγ pathway with rat estrous cycle changes to predict ovarian toxicity. Evaluation on additional datasets showed that the accuracy of FeRetoxML for female reproductive toxicity mediated by the ERα and PPARγ pathways was 84.6 % and 83.3 %, respectively. Also, this study applied the BiLSTM algorithm to construct FeRetoxBiLSTM, which, despite lower predictive performance, exhibited notable application potential. Our study introduced a novel AOP-based modeling approach for high-throughput screening of female reproductive toxicants, and this approach may be applicable to other toxicological endpoints. The models can accurately identify the chemicals with female reproductive toxicity, and provide a promising tool for chemical management, including determining the priority order of in vivo reproductive toxicity testing for chemicals.