Effective measures to protect the environment from toxic chemicals require data on the ecological impacts of chemicals—yet sufficient data is available for only 3.5% of chemicals on the market in the EU to assess species sensitivity. A recently published study proposes an approach using AI to make a significant contribution toward meeting this data need. Instead of predicting the potential effects of each chemical in isolation, the researchers utilized the full range of data from all compounds and all species. 

The goal was to take an unknown species-chemical pair and predict the concentration of the chemical that is lethal to 50% of the species population (LC50). The pairwise AI learning approach enabled the prediction of more than 16 million LC50 values, allowing the researchers to create so-called hazard heatmaps for all species-chemical pairs—maps using different colors to represent the sensitivity of the species in question to the chemicals. Furthermore, they were able to create maps showing the distribution patterns of individual hazards for individual species and for species groups. 

Publication:
Posthuma, L., Price, T., and Viljanen, M. (2025) Environmental Science & Technology, 59, 16250–16260. Improving the ecotoxicological risk assessment of chemicals through pairwise learning. https://pubs.acs.org/doi/10.1021/acs.est.5c01289

Source and further information:
https://environment.ec.europa.eu/news/new-ai-approach-bridges-data-gaps-improve-toxicity-assessment-chemicals-2026-04-22_en?pk_source=ec_newsroom&pk_medium=email&pk_campaign=sfep_news&pk_content=issue633_na1587