A Decision Support System to Predict Acute Fish Toxicity
Madsen, Anders L.; Moe, S. Jannicke; Braunbeck, Thomas; Connors, Kristin A.; Embry, Michelle; Schirmer, Kristin; Scholz, Stefan; Wolf, Raoul; Lillicrap, Adam David
Peer reviewed, Journal article
Published version
View/ Open
Date
2022Metadata
Show full item recordCollections
- NGI articles [1085]
Original version
Proceedings of Machine Learning Research (PMLR). 2022, 186 253-264.Abstract
We present a decision support system using a Bayesian network to predict acute fish toxicity from multiple lines of evidence. Fish embryo toxicity testing has been proposed as an alternative to using juvenile or adult fish in acute toxicity testing for hazard assessments of chemicals. The European Chemicals Agency has recommended the development of a so-called weight-of-evidence approach for strengthening the evidence from fish embryo toxicity testing. While weight-of-evidence approaches in the ecotoxicology and ecological risk assessment community in the past have been largely qualitative, we have developed a Bayesian network for using fish embryo toxicity data in a quantitative approach. The system enables users to efficiently predict the potential toxicity of a chemical substance based on multiple types of evidence including physical and chemical properties, quantitative structure-activity relationships, toxicity to algae and daphnids, and fish gill cytotoxicity. The system is demonstrated on three chemical substances of different levels of toxicity. It is considered as a promising step towards a probabilistic weight-of-evidence approach to predict acute fish toxicity from fish embryo toxicity.