Assessment of toxicity changes during photocatalytic degradation of antibiotics using T.E.S.T and machine learning

This study selected Doxycycline hydrochloride (DOX) as a representative antibiotic and titanium dioxide nanoparticles (P25) as a typical photocatalyst to investigate the photocatalytic degradation process. The research demonstrates three key innovations: (1) The degradation pathway of DOX by P25 was systematically elucidated through the combination of in situ reactive oxygen species (ROS) verification (confirming the dominant roles of ¹O₂, •O₂^-, and h⁺) and multi-technique intermediate tracking using liquid chromatography-mass spectrometry/two-dimensional correlation spectroscopy (LC-MS/2D-COS). This approach confirmed the direct involvement of ¹O₂, •O₂^-, and h⁺ in the reaction and resolved the sequence of functional group changes. (2) A random forest (RF) model was developed and integrated with T.E.S.T. simulations to construct time-resolved toxicity evolution profiles. The acute toxicity, developmental toxicity, and mutagenicity of intermediates were evaluated using the T.E.S.T. toxicity assessment software. The established RF model demonstrated high prediction accuracy with a coefficient of determination (R²) of 0.932, mean absolute error (MAE) of 0.018, and root mean square error (RMSE) of 0.039. This model enables tracking of photocatalytic intermediate variations and, combined with toxicity simulation, facilitates the prediction of toxicity levels at different degradation stages, thereby deriving toxicity evolution patterns after photocatalytic degradation. (3) This study establishes a novel framework for degradation endpoint optimization through dynamic toxicity prediction, providing a strategic solution for minimizing ecological risks in photocatalytic water treatment.