Mn,Ce-CDs-Loaded MIL-53(Fe) Nanozymes Multisubstrate Sensor Array Integrated with Multiple Machine Learning Complementary Strategy for Highly Reliable Antioxidant Discrimination and Determination.
Journal:
Analytical chemistry
Published Date:
Jan 4, 2026
Abstract
Accurate antioxidant detection remains crucial for biomedical and food applications, although conventional methods face persistent challenges including structural similarity interference, instrument dependency, and unreliable detection outcomes. To address these limitations, this study developed innovative nanozyme-based multisubstrate sensor array integrated with machine learning. A novel nanocomposite was constructed through immobilizing manganese- and cerium-codoped carbon dots (Mn,Ce-CDs) onto metal-organic frameworks MIL-53(Fe). The Mn,Ce-CDs significantly enhanced electron transfer efficiency and remarkably improved superoxide anion (O2•-)-mediated oxidase-like activity of the composite. This nanozyme effectively catalyzed four chromogenic substrates, 3,3',5,5'-tetramethylbenzidine (TMB), o-phenylenediamine (OPD), 2,2'-azino-bis(3-ethylbenzothiazoline-6-sulfonic acid) (ABTS), and 4-aminoantipyrine (4-AP), to generate distinct multicolor fingerprint patterns, enabling multidimensional signal outputs without requiring exogenous metal ions or multiple probes. The research incorporated seven machine learning algorithms, including linear discriminant analysis (LDA), hierarchical clustering analysis (HCA), artificial neural network (ANN), K-nearest neighbors (KNN), support vector machine (SVM), decision tree (DT), and naive bayes (NB), leveraging their discriminative criterion variations to establish complementary effects that successfully differentiated eight structurally analogous antioxidants (10 nM to 25 μM), with a minimum identification concentration of 10 nM. To rectify misclassification arising from DT and NB algorithms' sensitivity to high-dimensional features, innovative LDA-DT and LDA-NB tandem algorithms were developed. Through employing dimensionality reduction techniques, the classification accuracy of the two algorithms was significantly improved from 86.7% and 93.1% to 100%, respectively. The developed sensor array successfully enabled quantitative analysis of the eight antioxidants and achieved accurate identification of multiple antioxidants in complex matrices including serum, urine, cell lysates, bacterial cultures, and various food samples.
