Machine Learning-Enabled Time-Resolved Nanozyme-Encoded Recognition of Endogenous Mercaptans for Disease Diagnosis

With their important role in regulating intracellular redox balance and maintaining cell homeostasis, endogenous mercaptans are recognized as biomarkers of many diseases in clinical practice, and thus establishing efficient yet simple methods to distinguish and quantify endogenous mercaptans is of great significance for health management. Here, we propose a machine learning-enabled time-resolved nanozyme-encoded strategy to identify endogenous mercaptans in the presence of potential interferents for disease diagnosis. Diethylenetriaminepenta(methylenephosphonic) acid was first employed to coordinate with Mn³⁺ to prepare a new amorphous nanozyme, which exhibited excellent oxidase-like activity in catalyzing the oxidation of colorless 3,3',5,5'-tetramethylbenzidine to its blue oxide. The addition of endogenous mercaptans (cysteine, homocysteine, and glutathione) could competitively suppress the chromogenic process to different extents due to their discrepant antioxidant abilities, providing specific fingerprints over time for each species. With this mechanism, a time-resolved sensor array with the nanozyme as a sole sensing unit was constructed to accurately identify different types and levels of mercaptans and their various mixtures with the help of pattern recognition. Furthermore, machine learning was combined with the sensor array to construct a stepwise prediction model consisting of concentration-independent classification and concentration-associated regression, which could not only differentiate cancer cells from normal ones based on intracellular glutathione but also evaluate the severity of cardiovascular diseases according to serum homocysteine, showing great application potential in disease diagnosis.