Evaluation of a Bayesian Network for Strengthening the Weight of Evidence to Predict Acute Fish Toxicity from Fish Embryo Toxicity Data

Abstract

The use of fish embryo toxicity (FET) data for hazard assessments of chemicals, in place of acute fish toxicity (AFT) data, has long been the goal for many environmental scientists. The FET test was first proposed as a replacement to the standardized AFT test nearly 15 y ago, but as of now, it has still not been accepted as a standalone replacement by regulatory authorities such as the European Chemicals Agency (ECHA). However, the ECHA has indicated that FET data can be used in a weight of evidence (WoE) approach, if enough information is available to support the conclusions related to the hazard assessment. To determine how such a WoE approach could be applied in practice has been challenging. To provide a conclusive WoE for FET data, we have developed a Bayesian network (BN) to incorporate multiple lines of evidence to predict AFT. There are 4 different lines of evidence in this BN model: 1) physicochemical properties, 2) AFT data from chemicals in a similar class or category, 3) ecotoxicity data from other trophic levels of organisms (e.g., daphnids and algae), and 4) measured FET data. The BN model was constructed from data obtained from a curated database and conditional probabilities assigned for the outcomes of each line of evidence. To evaluate the model, 20 data‐rich chemicals, containing a minimum of 3 AFT and FET test data points, were selected to ensure a suitable comparison could be performed. The results of the AFT predictions indicated that the BN model could accurately predict the toxicity interval for 80% of the chemicals evaluated. For the remaining chemicals (20%), either daphnids or algae were the most sensitive test species, and for those chemicals, the daphnid or algal hazard data would have driven the environmental classification. Integr Environ Assess Manag 2020;16:452–460. © 2020 The Authors. Integrated Environmental Assessment and Management published by Wiley Periodicals, Inc. on behalf of Society of Environmental Toxicology & Chemistry (SETAC)

Development of a hybrid Bayesian network model for predicting acute fish toxicity using multiple lines of evidence

“…Data were then averaged for each chemical, resulting in 152 QSAR values representing modelled toxicity to juvenile fish. More details on QSAR model selection, equation acceptability criteria and results can be found in (Lillicrap et al, 2019).…”

“…The BN model is structured along the four lines of evidence (Figure 1), representing different types of information for a given chemical substance for which a user wants to predict the acute fish toxicity. The parameterisation of all lines is described in Supplementary material (Tables S.1-S.21), while more details on the assumptions are provided by Lillicrap et al (2019).…”

“…Conversely, if the toxicity ratio is between 0.5 and 2, then the toxicity to fish is assumed to be approximately equal to Daphnia and algae, and a toxicity at the same level (but with more uncertainty) is assumed for fish. The full details of these assumptions are described by Lillicrap et al (2019).…”

“…The reasoning behind the CPT of “Toxicity to fish predicted from other taxa” is explained for all 50 combinations of the parent states by Lillicrap et al (2019). In brief, if the ratio of the average toxicity values to Daphnia vs. algae is between 0.5 and 2, it can be assumed that the chemical has a similar mode of action for the two taxa, and that the chemical will affect fish in a similar way.…”

“…The paper will also present an online web interface to the model. More detailed information on this model’s performance for specific chemical substances and endpoints, and the implications for current legislation and guidelines for toxicity testing, are addressed by Lillicrap et al (2019). Feedback from researchers, regulators or other potential users to this first BN version will be used for improving future versions of the model, with the aim of making it a useful tool in a WOE approach for predicting acute fish toxicity.…”