作者：陆雨顺 陈璐 钱永忠 许彦阳

刊物名称：Critical Reviews in Environmental Science and Technology

发表年月：June 2026

摘要内容：Chemical contaminants in the environment typically occur as complex mixtures. Non-additive effects among constituents of mixtures (synergism or antagonism) pose substantial challenges to conventional risk assessment paradigms that are based on single-chemical evidence. This review analyzes how deep integration of artificial intelligence (AI) with New Approach Methodologies (NAMs) and computational toxicology can transform mixture risk assessment. We first examine the limitations of traditional joint-toxicity models—concentration addition (CA) and independent action (IA)—in capturing nonlinear interactions and toxicokinetic processes. We then highlight emerging applications of machine learning for toxicity screening of previously uncharacterized chemicals, prediction of non-additive effects, and automated evaluation of toxicity endpoints. Next, we discuss how AI enables a multiscale risk assessment framework from mechanistic initiation to system-level prediction by modeling molecular initiating events (MIEs), integrating adverse outcome pathway (AOP), and strengthening organoid-based assays and physiologically based pharmacokinetic (PBPK) modeling. Finally, we outline the basic data and regulatory validation strategies required to build next-generation intelligent risk-assessment systems, with the goal of providing a new technical roadmap and scientific decision support for health risks under complex real-world exposures.

原文链接： https://doi.org/10.1080/10643389.2026.2693546