New Study Uses Machine Learning and Raman Spectroscopy for Highly Accurate Foodborne Pathogen Detection

Food safety testing is an important field in the global economy to ensure the food products being sent to market are of high quality. A recent study from researchers at Tianjin University examines this issue closely by proposing a new method that can rapidly identify food-borne pathogens in food products. This study, published in Food Bioscience, presents a method that combines Raman spectroscopy with machine learning to rapidly and accurately identify foodborne pathogens (1). Led by Di He of Tianjin University, the study demonstrates how deep learning algorithms can overcome challenges posed by the biochemical similarity of different pathogenic serotypes, achieving a prediction accuracy of 98.4% (1).

Foodborne illnesses are a common occurrence in modern society. Currently, it is estimated that 1 in 6 Americans get sick from contaminated food (2). People who acquire foodborne illnesses consume food that contains harmful bacteria, parasites, or other agents (2). Although there are testing methods out there, these methods often require time-intensive culturing and molecular identification testing techniques, limiting their efficiency in large-scale food safety monitoring (1). Raman spectroscopy has emerged as a promising alternative for pathogen detection, but its effectiveness has been hindered by the high degree of biochemical similarity among serotypes of the same genus (1).

Assortment of healthy food dishes. Top view. | Image Credit: © Yaruniv-Studio - stock.adobe.com

In this study, He and colleagues tried implementing machine learning (ML) methods to improve discrimination and prediction accuracy among multiple foodborne pathogen serotypes. Their study evaluated the effectiveness of Raman micro-spectroscopy in combination with various ML models for identifying and predicting distinct serotypes of pathogenic bacteria. The research team optimized multiple ML techniques, including four unsupervised algorithms: K-means, agglomerative nesting (AGNES) clustering, spectral clustering, and density-based spatial clustering of applications with noise (DBSCAN) (1). Using these algorithms allowed the researchers to learn more about foodborne pathogens and their phylogenetic linkages (1).

The researchers also tested eight supervised learning models. These models were K-nearest neighbor (KNN), decision tree (DT), random forest (RF), support vector machine (SVM), adaptive boosting (AdaBoost), Gaussian naive Bayes (GNB), convolutional neural network (CNN), and vision transformer (1). By comparing the ML approaches to the deep learning models, the researchers found that deep learning models outperformed traditional machine learning approaches, particularly in scenarios with interference from multiple bacterial genera (1).

Among the tested models, the study introduced and evaluated a dual-scale CNN architecture, which significantly outperformed existing single-scale CNN models. This novel model achieved a remarkable 98.4% prediction accuracy in distinguishing foodborne pathogen serotypes (1). The dual-scale CNN structure enhances identification capabilities by simultaneously capturing both local feature peaks and global spectral patterns, resulting in a 4.9% increase in accuracy compared to single-scale CNN models (1). Furthermore, in mixed data sets containing multiple genera, the dual-scale CNN model demonstrated an even higher accuracy of 99.2% when looking at complex samples (1).

The findings from this study strongly support the integration of ML-driven Raman spectroscopy into food safety testing protocols. Unlike traditional pathogen detection methods, which can take hours or even days, this advanced approach offers rapid and precise results, enabling real-time monitoring of microbial contamination in food production and distribution chains (1).

In the conclusion of their article, He and colleagues proposed future research directions. They suggest that future studies could explore its application in detecting a wider range of foodborne pathogens, optimizing spectral data preprocessing techniques, and integrating this system into automated food safety monitoring platforms (1). Because food safety regulations are becoming more stringent, methods such as the one proposed by He and colleagues are needed.

On average in the United States, foodborne illnesses result in 128,000 hospitalizations and 3,000 deaths each year (2). To reduce these numbers, it is paramount that detecting foodborne pathogens becomes more effective. As this study shows, Raman spectroscopy and ML can potentially help in this space, which could positively impact and solve global food security challenges.

References

1. Sun, J.; He, D.; You, Y. Raman Spectroscopy Powered by Machine Learning Methods for Rapid Identification of Foodborne Pathogens. Food Biosci. 2025, 66, 106281. DOI: 10.1016/j.fbio.2025.106281
2. U.S. Food & Drug Administration, Foodborne Pathogens. FDA.gov. Available at: https://www.fda.gov/food/outbreaks-foodborne-illness/foodborne-pathogens (accessed 2025-03-24).