Toxicol Lett 2026 Jan 01;:111813. doi: 10.1016/j.toxlet.2025.111813.

Predicting acute developmental toxicity of chemicals in embryos of the African clawed frog (Xenopus laevis): calibration and validation of regression-based quantitative structure activity relationship models for hazard assessment of chemicals in anuran amphibians.

???displayArticle.abstract???
Global decline of amphibian populations has been correlated with a range of endogenous and exogenous variables including their unique physiology and ecology, exposure to chemicals, habitat reduction, climate change, as well as biological hazards such as emerging infectious diseases. The African clawed frog (Xenopus laevis) is an OECD test species used in toxicity testing as a specific proxy for humans and environmentally relevant species, for which acute toxicity data for a range of chemicals have been generated historically by industry, a number of public health agencies and academia. Of particular relevance are mechanistic effects of endocrine-active substances on metamorphosis and the thyroid axis, resulting in developmental toxicity. From such toxicity data, no open-source quantitative structure-activity relationships (QSARs) have been developed as in silico tools to predict such toxicity for data-poor chemicals in X. laevis. Such QSAR models can provide a quantitative starting point for the hazard assessment of chemicals in other anuran amphibians. This manuscript provides a description of the data collection and curation from the largest historical databases including the US EPA ECOTOX knowledgebase and the Ortiz-Santaliestra databases available for Xenopus embryos as acute median lethal concentrations (LC50-12 hours) for a total of 349 unique structures and 1978 individual entries. After data curation, the database contained 359 individual entries for a total of 175 compounds, and were computed using the negative logarithm of molar concentrations expressed as 12hlog 1/LC50 mmol/L. Subsequently, the database was then split into training set, test set and prediction set with 120, 40 and 13 compounds, respectively. These datasets were then used for the development and validation of two different QSAR models: 1. A k-Nearest Neighbours (k-NN) models using istKNN (in silico tools - KNN). 2. A multiple linear regression model (MLR) using the QSARINS (QSAR-INSUBRIA) software version 2.2.4. Overall, the QSAR models performed well for predicting acute toxicity of chemicals in Xenopus embryos and the MLR model performed slightly better than the k-NN model with correlation coefficients of 0.76 and 0.75 and root mean square errors of 0.63 and 0.67, respectively. However, underestimation of predictions for highly toxic compounds were observed and these limitations are discussed for both the k-NN and multiple linear regression model in the light of mechanistic interpretation and expert knowledge. Variability in the experimental datasets as well as under-representation of the most toxic compounds in the database are highlighted as major drivers influencing such underpredictions. Future directions from the present work include the modelling of other endpoints and developmental stages as well as other amphibian species using the available, although limited, data. Overall, it can be foreseen in the near future that such databases and models will be important to develop more performant in silico models, and ultimately to develop NAMs for ecotoxicity assessment of chemicals in anuran amphibians while reducing animal testing.

???displayArticle.pubmedLink??? 41483836
???displayArticle.link??? Toxicol Lett