Toxicity prediction and ecological risk assessment of new contaminants to rare and endangered species using machine learning-QSAR: a case study of conserving Gobiocypris rarus in the Yangtze River Basin

Rare and endangered species play a crucial role in ecosystems, but pollutants, especially new contaminants such as per- and polyfluoroalkyl substances (PFASs), increasingly threaten their survival. It is vital to conduct ecological risk assessments for these species. However, conventional toxicity testing methods are ethically and practically infeasible for rare and endangered species. Using modeling approaches to predict toxicity is a viable solution to assess ecological risks. This study proposed a machine learning, quantitative structure-activity relationship (ML-QSAR) method for predicting toxicity, using a case study of Gobiocypris rarus, an endangered fish species endemic to China's Yangtze River Basin. Six algorithms were applied to construct models based on molecular descriptors and life stage encoding. The random forest-based model showed optimal performance for predicting both acute and chronic toxicity (R2Acute = 0.99, R2Chronic = 0.93). Feature importance analysis revealed a key mechanistic divergence: life stage was the dominant factor in acute toxicity; in contrast, chronic toxicity was primarily driven by molecular interaction descriptors. The optimal models were applied to predict the toxicity of 73 pollutants detected in the Gobiocypris rarus' habitat. The ecological risk assessment indicated that current environmental concentrations of 12 PFASs remain low, but it is important to note their persistence and potential long-term ecological risks. This study presents a novel, non-testing approach for quantifying chemical threats to endangered species for which data are limited. Integrating life-stage-specific mechanisms into an ML-QSAR method provides a powerful tool for supporting evidence-based conservation and pollutant management in vulnerable aquatic ecosystems.