Prediction and verification of emerging contaminants catalytic degradation based on machine learning and molecular fingerprint.

Abstract

Faced with the increasing presence of recalcitrant emerging contaminants (ECs) in aquatic environments, the development of highly efficient and environmentally friendly treatment technologies has become particularly urgent. Density functional theory (DFT) calculations offer unparalleled value in mechanistic studies by revealing the interaction mechanisms between contaminants and catalytic materials at the electronic structure level. However, the computational intensity of DFT hinders its use for rapid screening of numerous contaminants. To overcome this limitation, we integrated the molecular fingerprint (MF) features of contaminants into the dataset to construct a machine learning (ML) performance prediction model. This enables high-throughput evaluation of the catalytic performance of catalysts while maintaining good predictive accuracy (R2 = 0.81). By introducing SHapley Additive exPlanations (SHAP) and partial dependence plots (PDP), we elucidated the contribution mechanisms of contaminant molecular structural features, catalyst properties, and reaction conditions to performance prediction, identifying the pivotal roles of total pore volume (Vtotal), catalyst concentration (Catalyst-C), and Peroxymonosulfate concentration (PMS-C). Molecular-scale analysis of pollutants revealed that aromatic-related structures enhance reaction activity, while halogen-related substituents inhibit reaction efficiency. Finally, we verified the reliability of the model through experimental validation. This study provides an effective predictive tool and elucidates the structure activity relationship at the atomic scale, providing a theoretical basis and practical guidance for the design of green catalysts and optimization of wastewater treatment processes.