An Intelligent Sensing Paradigm Enabled by Phosphate-Regulated Nanozyme Activation and Machine Learning for Antioxidant Discrimination at Neutral pH

Nanozymes with oxidase-like activity hold great promise for biosensing but are severely limited by their low catalytic efficiency at neutral pH. Herein, we propose a strategy to address this limitation by using endogenous phosphorylated metabolites (e.g., ATP, CTP, GTP) as modulators to reactivate Ce-MOF nanozymes under neutral conditions. Mechanistic studies reveal that phosphate coordination tunes the Ce 3+ /Ce 4+ ratio, thus facilitating superoxide radical generation and thereby restoring oxidase-like activity. Capitalizing on the distinct regulatory profiles of ATP, CTP, and GTP, a triple-channel colorimetric sensor array was constructed. This array enabled the fingerprint-based discrimination and quantification of seven structurally similar antioxidants, achieving detection limits in the range of 0.140 to 1.91 μM. For reliable concentration-independent discrimination of antioxidants, a two-step machine learning (ML) framework was developed. An optimized K-nearest neighbors (KNN) model achieves 98.1% accuracy for concentration-independent qualitative recognition, and linear discriminant analysis (LDA) is subsequently applied for accurate quantification. The integrated platform successfully analyzed commercial health supplements, achieving a 92.0% blind identification accuracy and quantitative results with ∼3.40–6.10% error relative to labeled values. Furthermore, the platform demonstrated satisfactory performance in spiked fetal bovine serum samples, with recoveries ranging from 80.4% to 118% for three model antioxidants. This work not only provides a fundamental strategy for engineering adaptive nanozymes but also establishes an intelligent sensing paradigm for complex sample analysis under neutral conditions.