Integrative QSAR and q-RASAR models for Danio rerio acute toxicity from short-term exposures: Mechanistic insights and data gap filling

The Toxic Substances Control Act (TSCA) mandates the U.S. EPA to monitor all chemicals used in the country, over 86,000 to date, posing a major challenge for comprehensive toxicity testing. Due to the impracticality of experimentally assessing every compound, computational approaches such as quantitative structure-activity relationships (QSAR) and quantitative read-across structure-activity relationships (q-RASAR) provide efficient alternatives for estimating aquatic toxicity, which is critical for safeguarding both aquatic ecosystems and human health. In this study, we curated acute LC₅₀ toxicity data for Danio rerio (zebrafish), a widely used model in ecotoxicology, from the EPA's ToxValDB. Datasets were categorized by experimental duration (2, 3, and 4 h), study type (mortality), exposure conditions, and chemical class. This resulted in curated sets of 97 (2 h), 45 (3 h), and 356 (4 h) compounds. We developed six predictive models, three QSAR and three q-RASAR to forecast zebrafish toxicity. The q-RASAR model shows higher predictive performance than the QSAR model across all durations, with statistically significant improvements for the 3 h dataset (both parametric and non-parametric tests) and the 4 h dataset under non-parametric analysis. To assess practical applicability, we predicted toxicity for over 1100 external compounds lacking zebrafish data, effectively addressing key ecotoxicological data gaps. The integration of QSAR and q-RASAR across exposure durations not only enhances predictive accuracy but also provides mechanistic insights into toxicity pathways. These models offer a valuable tool for regulatory risk assessment, supporting more informed and sustainable environmental decision-making.