AI-Driven Species Sensitivity Distribution (AI-4-SSD) Framework for Predicting Aquatic Ecological Risks of Chemical Pollutants in Global Near-Coastal Environments

Currently, more than 350,000 chemicals and chemical mixtures have been registered for global production and use, and they are inevitably released into global near-coastal environments during their life cycle. However, the multidimensional adverse effects of these chemicals on marine species populations, assemblages, and biodiversity remain unknown. Herein, we proposed an AI-based framework (AI-4-SSD) for whole-chain predictions of chemical exposure, aquatic toxicity, and risk in the global near-coastal environment. As the core of this framework, a multimodal deep learning model was developed to predict population-level aquatic toxicities of diverse chemicals on eight marine species across three phyla, demonstrating excellent predictive power (<i>R</i>² of 0.85 in the test set). Using the AI-4-SSD framework, we identified six high-risk chemicals threatening marine species assemblages via direct effects on life-history characteristics, including DDT and 6:2/8:2 diPAPs, from approximately 3,000 target chemicals potentially entering the global near-coastal environment. Specifically, in the Black Sea, we found that cumulative risks from coexposure to hundreds of detected chemicals could drive biodiversity loss during 2016-2019, despite individual chemicals posing negligible risks. This work not only provides a user-friendly prediction framework for rapidly identifying high-risk chemicals but also highlights the necessity of mixture risk management for the conservation of marine biodiversity.