Linking micropollutant mixtures and macroinvertebrate ecological health using AI-based toxicity predictions

This study investigates how micropollutant mixtures affect the ecological health of benthic macroinvertebrate communities by combining ecotoxicological predictions with macroinvertebrates and water chemistry data. Using the AI-based model TRIDENT, we predicted the toxicity (EC₁₀) of 559 micropollutants. Substances were grouped according to their EC₁₀ through a cluster analysis. The micropollutants' effects on the ecological health, represented by the multimetric index, was evaluated in 207 sampling sites with beta regressions using two approaches to assess the toxic pressure. The first model considered the maximum toxic pressure of the entire water sample as the single explanatory variable. The second model incorporated the maximum toxic pressure of every cluster as separate explanatory variables, and translated better the effects of pollution on the multimetric index (pseudo R-squared = 0.28) compared to the other model (pseudo R-squared = 0.15). Additionally, we identified substances that drove the toxic pressure of our samples. Another beta regression showed that a large amount of the communities' health (pseudo R-squared = 0.24) could be explained by four indicator substances alone. Our findings reveal that micropollutant contamination plays a key role in the degradation of aquatic ecosystems, and that summarizing a mixture of micropollutants to a single-substance metric underestimates this contribution.