Abstract: By combining chemical and biological databases, we developed a knowledge-based database to refine toxicity data and provide training datasets for artificial intelligence models. As a case study, ToxCast-based models were developed using various machine and deep learning algorithms to predict toxicities. Artificial intelligence models offer a new opportunity to assess the potential toxicity of a vast number of environmental chemicals. However, many toxicity prediction models are inherently black boxes, making interpretation challenging for toxicologists and hindering regulatory acceptance. The mechanisms leading to the onset of apical toxicity are complex, and in the absence of process evidence, it is difficult to trust the results. The adverse outcome pathway framework holds promise in addressing this issue. This study aimed to develop explainable artificial intelligence models for predicting toxicity using ToxCast data within an adverse outcome pathway framework. Initially, we identified adverse outcome pathways leading to reproductive toxicity from the AOP Wiki. We then collected in vitro bioactivity data from ToxCast and in vivo toxicity data from ECOTOX DB. These data were integrated into adverse outcome pathways and each assay- adverse outcome pathway pair underwent assessment to determine relevance to taxa within ecosystems. Finally, machine learning models were developed to predict each mechanistic and apical endpoint-based toxicity. Models for each key event and apical endpoint achieved high performance.