Fraunhofer Institute for Molecular Biology and Applied Ecology IMEAI-supported assessment of chronic mixture risks
In the field of environmental risk assessment, the evaluation of chronic exposure to chemical mixtures is becoming increasingly important, as aquatic organisms are exposed to complex mixtures of substances in reality rather than individual substances. In a study, published in Environment International, the team of the Department Environmental Media Related Ecotoxicology at Fraunhofer IME, headed by Prof. Henner Hollert investigated this issue using the example of a small European river. The aim was to use artificial intelligence to fill data gaps in the chronic toxicity of individual substances and to carry out a realistic risk assessment for mixtures.
The combination of chemical analysis, empirical and AI-modelled toxicity data allows risks to be assessed.
192 organic substances were analytically quantified at six locations. Experimental chronic toxicity data was available for less than half of these. To overcome these gaps, they used an open-source AI-supported prediction model that is validated and trained on approximately 144,000 empirical toxicity measurements for 6,469 substances across 1,842 species. It estimated missing effect concentrations for algae, aquatic invertebrates and fish. On this basis, risk quotients were calculated using the concentration-addition approach, and hotspots with increased chronic risk were identified.
The results show that several river sections pose a relevant ecological risk, with fish being the most sensitive group of organisms. It was particularly striking that human pharmaceuticals – including antibiotics, analgesics and cardiovascular substances – contributed most to the overall risk. The inclusion of AI-based predictions led to higher, presumably more realistic risk estimates compared to approaches based solely on empirical data.
The study illustrates that AI-supported methods can be a robust complement to traditional assessment strategies, provided that the underlying assumptions and functioning of the AI systems are understandable to users. By linking environmental monitoring with advanced modelling methods, a conceptual framework is created for integrating innovative modelling approaches into future ecotoxicological assessments and providing regulatory authorities with a more informed basis for decision-making when dealing with chemical water pollution.