A cross-domain hybrid learning framework for unified prediction and mechanistic insights into disinfection...

Disinfection by-products (DBPs) generated during drinking water treatment pose significant risks to human health. However, the distinct molecular properties and formation pathways of different DBP categories make it challenging to develop a unified predictive framework. Herein, a cross-domain hybrid learning framework was constructed by integrating the environmental and quantum domains as input for unified DBP prediction. The environmental domain comprised 13 features related to water quality indicators and disinfection parameters, while the quantum domain incorporated density functional theory descriptors for molecular featurization of DBPs. The proposed framework showed accurate and robust performance in predicting 11 DBPs, with an R² of 0.922. Shapley additive explanation analysis was used to interpret the relative contributions of environmental and quantum domains. In the environmental domain, the molecular weight fractions and fluorescent components of natural organic matter exerted distinct regulatory effects on DBP concentrations. The quantum domain linked intrinsic molecular properties with DBP reaction pathways, providing a mechanistic foundation for the prediction of diverse DBPs. Overall, this work presents a unified predictive framework for the 11 investigated DBPs, advancing DBP control strategies and supporting drinking-water safety management.