AI Revolutionizes Environmental Risk Assessment of Chemicals
The bioconcentration factor (BCF) is a critical measure of chemical accumulation in fish compared to their surrounding water. Until recently, it was believed to be a constant for each chemical, but a groundbreaking study led by Professor Heinz Köhler from the University of Tübingen's Institute of Evolution and Ecology challenges this notion. The research reveals that BCF varies depending on the test concentration, casting doubt on the bioaccumulation data used in the EU's licensing procedure for over half of the chemicals that potentially accumulate in fish. To address this, the team has developed an AI tool, BCFpro, which enables researchers to assess the bioaccumulating properties of substances with high accuracy. This tool is freely available, and the team's findings are published in the Journal of Hazardous Materials.
The concentration of chemicals in the food chain is a significant concern, especially since it affects humans as well. Professor Köhler highlights the potential for massive chemical buildup in the human body, with harmful effects often only becoming apparent after a long time. The BCF in fish is a widely accepted benchmark for assessing chemical risks and standardizing bioaccumulation data in animals.
However, the study reveals a surprising twist: the BCF does not provide a consistent criterion for each chemical. The test concentration for the surrounding water body significantly influences the BCF, with higher concentrations generally resulting in lower BCFs. This finding was mathematically proven and physiologically explained by the research team, which included collaboration with the German Federal Environment Agency and the Universities of Yale and Athens. The team evaluated thousands of studies to reach this conclusion, noting that this effect had not been previously identified in chemical hazard classification regulations worldwide.
To address this complexity, the team employed deep learning, an AI machine learning method, to develop BCFpro. This program can predict experimental data on the BCF with 90% certainty. By using artificial networks similar to brain neurons, deep learning processes complex datasets and extracts valuable patterns and features. BCFpro can also identify critical values for chemicals under worst-case scenarios, where bioaccumulation is particularly severe.
The study's findings are alarming. When BCFpro was used to review chemicals categorized as non-accumulating dangerously in animals, it revealed that over 60% of substances that should have been identified as bioaccumulating were not. This highlights the importance of conducting chemical tests under environmentally relevant conditions to obtain realistic risk assessment values. To ensure standardized and reliable chemical categorization, the research team is making BCFpro freely available.
BCFpro's ability to predict the bioaccumulation of new chemical developments also holds significant potential for reducing animal testing. By focusing on practical applications and challenging existing methods, the University of Tübingen researchers are contributing to the improvement of ecotoxicological methods, enhancing environmental safety and animal welfare.
