Machine learning supported single-stranded DNA sensor array for multiple foodborne pathogenic and spoilage bacteria identification in milk.

Ensuring food safety through rapid and accurate detection of pathogenic bacteria in food products is a critical challenge in the food supply chain. In this study, a non-specific optical sensor array was proposed for the identification of multiple pathogenic bacteria in contaminated milk samples. Fluorescence-labeled single-stranded DNA was efficiently quenched by two-dimensional nanoparticles and subsequently recovered by foreign biomolecules. The recovered fluorescence generated a unique fingerprint for each bacterial species, enabling the sensor array to identify eight bacteria (pathogenic and spoilage) within a few hours. Four traditional machine learning models and two artificial neural networks were applied for classification. The neural network showed a 93.8 % accuracy with a 30-min incubation. Extending the incubation to 120 min increased the accuracy of the multiplayer perceptron to 98.4 %. This sensor array is a novel, low-cost, and high-accuracy approach for the identification of multiple bacteria, providing an alternative to plate counting and ELISA methods.