Clustering of bisphenols based on toxicity predictions for key aquatic species: Daphnia magna, Pimephales promelas, and Oryzias latipes

Abstract

The in silico assessment of chemical toxicity is crucial for regulatory frameworks like REACH, which support the use of QSAR models and read-across techniques to predict the properties of compounds. This study addresses the challenge of evaluating bisphenol A (BPA) alternatives, for which specific predictive models are often lacking. Utilizing VEGA software, we examined three ecotoxicological endpoints: toxicity in Daphnia magna (Daphnia magna Acute (EC50) Toxicity model (IRFMN)), Pimephales promelas (Fathead Minnow LC50 96 h toxicity (EPA)), and Oryzias latipes (Fish Acute (LC50) toxicity model (IRFMN)). We employed Self-Organizing Maps (SOM) to cluster bisphenol compounds based on similarities to experimental data from model training sets. Principal Component Analysis (PCA) was used to reduce dimensionality and visualize data, with color-coding to indicate predicted properties. Our results reveal that while BPA is often a cluster indicator due to its extensive inclusion in training sets, BPA alternatives frequently exhibit similar toxicological concerns. The clustering approach provides a nuanced understanding of the potential risks associated with BPA alternatives, suggesting that many may not offer significant safety improvements over BPA itself.