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 (R2 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.