Cascade machine learning reveals the regulatory mechanism of dissolved organic matter on nanoparticles aggregation...

The aggregation of nanoparticles (NPs) in aquatic environments is influenced by multiple environmental conditions, with dissolved organic matter (DOM) exerting complex regulatory effects on NPs aggregation. However, traditional models often struggle to capture and predict this intricate process. To address this, here we developed a novel cascade machine learning framework to predict the extent and rate of NPs aggregation for the first time. A comprehensive dataset, encompassing a wide range of NPs types, DOM sources, and water chemistry conditions from the literature, was established. A two-stage cascade model demonstrated high predictive potential, achieving precision predictions for aggregation extent (train R²=0.94, test R²=0.92) and aggregation rate (train R²=0.76, test R²=0.51). Mechanistic analysis revealed that the model can successfully distinguish the distinct dual roles of DOM. For aggregation extent, DOM directly mediates particle interactions, acting as either a polymer bridge (quantified by high attachment efficiency, α) that overrides electrostatic stability, or as a steric stabilizer; whereas for aggregation rate, DOM indirectly controls the process by modulating the overall systemic stability (represented by the critical coagulation concentration, CCC), which acts as the primary rate-limiting factor. This study provides a mechanistically interpretable framework that quantitatively distinguishes the direct (α-related) and indirect (CCC-related) effects of DOM, offering a more accurate scientific basis for the environmental risk assessment of NPs.