Mechanistic Prediction and High-Throughput Risk Screening of Aquatic Contaminants Using Daphnia magna AOP Network

Traditional toxicological testing cannot meet regulatory demands for numerous chemicals in aquatic environments. New Approach Methodologies (NAMs) such as High-Throughput Screening (HTS) and Adverse Outcome Pathway (AOP) framework offer sustainable alternatives. Here, we built a comprehensive Quantitative Structure-Activity Relationship (QSAR) prediction and risk assessment framework that integrates a multidomain deep learning model, applicability domain assessment, and toxicity risk prioritization. Specifically, by applying in vitro HTS bioactivity to multiple Molecular Initiating Events (MIEs) and Key Events (KEs) in Daphnia magna AOP network, we compiled a data set (6971 chemicals × 10,024 multidimensional structure descriptors) exhibiting high chemical structural diversity and strong representativeness of environmentally relevant chemicals. QSAR prediction models covering application domains of multiple MIEs/KEs were further developed. The 6971 chemicals were risk-prioritized based on the AOP network topology, with the ranking results consistent with existing toxicological evidence. In addition, chemicals with potential combined toxic risks were identified. An out-of-the-box aquatic toxicity prediction and risk assessment tool is provided, enabling automated multitarget bioactivity prediction, reliability and decidability evaluation, and toxic risk profiling for new chemicals. This study represents a concrete application of "AI for Science" in ecotoxicology, aiming to reduce the dependence on traditional animal experiments. It simplifies the identification of potential toxic effects and bypasses in vivo validation processes to advance the sustainable management of environmental chemicals.