MLSE machine learning algorithm-assisted priority screening of aquatic toxicity risk of DBPs and the guidance of environmental regulations

Disinfection by-products (DBPs) are ubiquitously formed during the water treatment processes. Due to the widespread occurrence, an urgent need exists to conduct systematic research on their aquatic toxicity risk and their toxicity mechanisms. In this study, an assessment system of a total of 833 DBPs aquatic toxicity risk was developed by integrating three organisms: Scenedesmus sp., Daphnia magna, and Danio rerio. Molecular docking was performed for all DBPs to quantify their potential aquatic toxicity. Four molecular structure–guided multi-learner stacking ensemble (MLSE) models were constructed to predict the lacking data of the toxicity risk of 51 DBPs, facilitating the development of a priority screening list of DBPs aquatic toxicity risk. Feature visualization revealed that high-risk DBPs tend to exhibit more complex molecular architectures, greater conformational flexibility, and increased non-hydrogen atom counts. Molecular dynamics simulations and amino acid analysis further indicated that these compounds possess higher hydrophobicity, form more hydrogen bonds, and interact with toxicity target proteins primarily through van der Waals and electrostatic forces. Density functional theory (DFT) calculations emphasized the importance of source control and priority monitoring of key DBP precursors. The purpose of this study is to assess DBPs aquatic toxicity risk, providing support for priority screening and ecological risk assessment.