Predicting the acute toxicity of organophosphate esters (OPEs) to aquatic organisms by modelling the structure-toxicity relationships using partial least square regression

Abstract

Organophosphate esters (OPEs) have been used worldwide as organophosphate flame retardants (OPFRs) since brominated flame retardants (BFRs) were banned. Due to the toxicity of these OPEs, environmental concerns and ecological risks arose. However, there are still large gaps in the understanding of their toxicity to organisms and the mechanisms of toxicity. After collecting the existing toxicity information and obtaining molecular descriptors of OPEs, a partial least square (PLS) regression model was used in this study to quantify the structure-toxicity relationships of OPEs. Based on the regression results, the acute toxicity of the remaining OPEs lacking acute toxicity data was predicted, and the risk level of total common OPEs was classified. The acute toxicity of 15 chemicals was collected, and >1660 molecular structure descriptors were obtained. The cross-validation results of the partial least square regression indicated that two principal components met the regression requirements with the selected features, and the regression equations of these chemicals were generated with selected molecular descriptors. The influence of physicochemical properties, such as hydrophobicity/molecular weight, in traditional perception of OPE toxicity was not that obvious, and acute toxicity was mainly influenced by the autocorrelation coefficients. However, the regression results indicated that the correlation between autocorrelation coefficients calculated based on different physicochemical properties and toxicity was different. According to the prediction result based on PLS regression, CDP may pose a high risk and halogenated alkyl-substituted OPEs such as TCEP may be less toxic. The results of the present study may help inform the environmental management and risk assessment of emerging chemicals such as OPEs.