Dual-Recognition AIEgens-Lysozyme Assemblies for Interpretable Machine Learning-Enabled Sensor Arrays: Toward Rapid...

Array-based sensing technology holds immense potential for the rapid identification of pathogenic bacteria. Nevertheless, developing a universal strategy that simultaneously achieves high-throughput discrimination of diverse foodborne pathogens while maintaining mechanistic interpretability poses substantial challenges. Herein, we introduce an intelligible sensing platform that synergizes dual-recognition nanoassemblies with interpretable machine learning for the instantaneous discrimination of 10 clinically relevant foodborne pathogens. Three aggregation-induced emission luminogens (AIEgens), functionalized with ethynyl, aldehyde, and phenylboronic acid moieties, are assembled with lysozyme to generate six cross-reactive sensing elements. Molecular docking and dynamics simulations reveal that the binding modes, ranging from purely hydrophobic to synergistically hydrophobic-polar interactions, govern the differential fluorescence responses toward bacterial surfaces. Upon exposure to bacteria, the sensor array produces unique fingerprints, enabling 100% classification accuracy across 7 machine learning algorithms. Crucially, the integration of XGBoost with SHapley additive explanations transcends conventional black-box analytics, unraveling the contribution of individual sensors and unveiling feature importance, which determines the dominant sensing elements and infers the pivotal role of aldehyde and phenylboronic acid groups in bacterial recognition. Their application potential is validated in the complex milk matrices, maintaining 100% classification accuracy and revealing the shifts of detection environment-dependent feature importance. This work establishes a new paradigm for intelligent pathogen surveillance, where the analytical performance and molecular insight converge to guide the rational design of next-generation sensing systems.