Machine learning-assisted multi-channel nanozyme sensor arrays for multiple pesticide tracking, tracing and metabolism analysis

To achieve precise pesticide residue detection and metabolic analysis, we innovatively proposed a machine learning-assisted multi-channel nanozyme sensor array. Five Cu-carboxylate nanozymes with outstanding laccase-like and peroxidase-like activities exhibited significantly distinct responses towards nicosulfuron, 2,4-dichlorophenoxyacetic acid, chlorpyrifos, cypermethrin, and their metabolites. Based on these, a 10-channel sensor array was constructed. Coupled with a Bayesian-optimized random forest (BO-RF) classification model, it enabled simultaneous identification of 4 pesticides and metabolites. Notably, qualitative recognition of 4 pesticides was not affected by variations in concentration or metabolic degree, which exhibited excellent traceback capability. Moreover, the BO-RF model showed outstanding predictive performance in assessing pesticide metabolic stages. The prediction accuracies were all exceed 97 % for 4 pesticides from unmetabolized to fully metabolized states. The practical applicability of the proposed strategy was further validated in spinach, agricultural lake water, and cultivation soil. This work offered a simple, efficient, and intelligent approach for pesticide tracking, traceability, and metabolic analysis.