To establish a method for the detection and identification of lysozyme in complex food matrices by gold nanoparticle (AuNP)-aptasensor array.

The 3 lysozyme-specific aptamers were selected as recognition elements, and a sensor array was constructed using AuNP color changes as the response signal. Based on the differences in binding affinity between lysozyme and the 3 aptamers, combined with the aggregation state and color change rules of AuNPs under salt-induced conditions, the characteristic color response differences of the samples were employed as lysozyme “fingerprints”. Supervised machine learning via linear discriminant analysis was applied to enhance model orthogonality, enabling recognition and prediction of unknown samples.

The 3 aptamers exhibited distinct binding affinities for lysozyme, with dissociation constants (K_D) of 141.528 nmol/L(Seq-1), 67.671 nmol/L (Seq-2) and 38.404 nmol/L (Seq-3). The limits of detection of lysozyme for the 3 aptamers Seq-1, Seq-2 and Seq-3 alone were 14.45, 15.739 and 1.95 nmol/L, respectively. The difference in binding affinity leaded to different degrees of protection for AuNPs by aptamers, causing AuNPs to present different aggregation states under salt-induced conditions and exhibit gradient color changes from red to blue. Characteristic “fingerprint spectra” were formed based on the color differences between food matrices with and without lysozyme added. A linear discriminant analysis model was constructed based on the data matrix of “3 nucleic acid aptamers ×18 kinds of food matrices ×5 parallel experiments”, achieving 100% accurate identification and classification of 18 kinds of food matrices with 100 nmol/L lysozyme added. The variance contribution rates of the 3 discriminant factors were 88.0%,7.6%, and 4.4% respectively. In addition, the recognition accuracy of this model for 36 blind samples reached 100%,indicating that this method could effectively distinguish lysozyme in different food matrices.

This study presents an AuNP-aptasensor array integrating pattern recognition and machine learning strategies, enabling reliable lysozyme detection in complex matrices. The approach provides a novel paradigm for the developing aptamer-based detection methods that mitigate matrix effects.