Machine Learning-Assisted Tissue-Residue-Based Risk Assessment for Protecting Threatened and Endangered Fishes in the Yangtze River Basin

Assessing pollutant risks to threatened and endangered (T&E) species is crucial for their conservation. However, traditional risk assessment methods for bioaccumulative pollutants to T&E fishes is challenging due to uncertainties in exposure-based toxicity relationships and data gaps. Tissue-residue concentration-response relationships provide a more reliable approach. This study employed machine learning (ML) algorithms to predict tissue-residue toxicity of bioaccumulative pollutants to T&E fishes, and found the extreme gradient boosting (XGBoost) model performed best, with an external validation R2 of 0.85 and a root-mean-squared error of 0.81. It was then used to predict the developmental toxicity of 22 bioaccumulative flame retardants to 98 T&E fishes from the Yangtze River basin, across four life stages. Results showed embryonic and juvenile stages were most sensitive, with organophosphate flame retardants (OPFRs), particularly (4-methylphenyl) diphenyl phosphate (CDPP) and isodecyl diphenyl phosphate (IDPP), exhibiting higher toxicity than novel brominated flame retardants (NBFRs) and polybrominated diphenyl ethers (PBDEs). Ecological risk assessment for T&E fishes revealed that aryl-OPFRs posed the highest risks, with CDPP exhibiting a risk quotient (RQ = 4.07) four times higher than the safety threshold, significantly exceeding the risks associated with NBFRs and PBDEs. This study established a novel ML-assisted tissue-residue-based risk assessment method for bioaccumulative pollutants to T&E fishes, which is significant for global T&E species conservation.