Development of a Bayesian Network Model for Predicting Fish Acute Toxicity from Fish Embryo Toxicity Data
Presenter: Adam Lillicrap, Norwegian Institute for Water Research
Monday, November 30th
10:00AM EST - 11:00AM EST
Reduction of animal testing, wherever possible, is required by legislations such as the EU Directive 2010/63/EU. Fish Embryo Toxicity (FET- OECD TG236) testing has been proposed to be an alternative to the juvenile acute fish toxicity test (AFT- OECD 203). However, FET data are not yet accepted as a replacement to the AFT test for certain regulatory purposes such as REACH. The European Chemicals Agency (ECHA) recommended that a weight-of-evidence (WoE) approach was needed for FET data before it could possibly be used in place of AFT data. Therefore, a Bayesian network (BN) model has been developed to incorporate multiple lines of evidence, in combination with FET data, to predict AFT. Bayesian networks are increasingly being used in ecological risk assessment because they can integrate large amounts of data and other information sources to produce discrete probability distributions, and predict the probability of specified states.
The objectives of this study were:
1) To develop and evaluate a BN model for predicting toxicity of substances to juvenile fish from embryo toxicity data in combination with other relevant information;
2) To apply the BN model in a WoE approach which can support replacing juvenile fish toxicity testing with embryo toxicity testing.
The BN model correctly predicted the AFT toxicity level for 14 substances, and gave lower toxicity for 6 substances. For the 6 substances with an incorrect prediction, 5 substances (2,4-Dichlorophenol, 4-Chlorophenol, Malathion, Naphthalene, Prochloraz) were less toxic to fish than to daphnids or algae, hence the AFT data would not drive the environmental risk assessment for those substances. In the case of 1 of the substances (Juglone), the AFT test was less sensitive than the FET data.