Detection and differentiation of Salmonella Typhimurium and Salmonella Enteritidis in poultry meat using a machine learning-enabled paper chromogenic array sensor approach

Authors: JIA, ZHEN - Auburn University; White, Stephen; ZHANG, BOCE - University Of Florida; KU, WEI-SHINN - Auburn University; GAO, XIAN - Auburn University; HARPER, LUCAS - Auburn University; KAFLE, RANJU - Auburn University; GARNER, LAURA - Auburn University; MOREY, AMIT - Auburn University; OWEN, KATHRYN - Auburn University

Submitted to: Food Research International
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/24/2026
Publication Date: 8/25/2026
Citation: Jia, Z., White, S.N., Zhang, B., Ku, W., Gao, X., Harper, L., Kafle, R., Garner, L., Morey, A., Owen, K. 2026. Detection and differentiation of Salmonella Typhimurium and Salmonella Enteritidis in poultry meat using a machine learning-enabled paper chromogenic array sensor approach. Food Research International. 192:112583. https://doi.org/10.1016/j.foodcont.2026.112583.
DOI: https://doi.org/10.1016/j.foodcont.2026.112583

Interpretive Summary: A critical food safety need is for rapid, accurate detection of Salmonella serotypes associated with foodborne illness. This study developed a paper colorized reaction sensor for specific metabolites with machine-learning for sensitive detection optimization. This approach has many advantages as it can be non-destructive, continuous, and provides simultaneous detection and differentiation of Salmonella Typhimurium (ST) and Salmonella Enteritidis (SE) in chicken breast. The present data distinguished Salmonella from background microflora. Furthermore, this approach achieved differentiation of Salmonella serotypes. Three machine learning models were evaluated for their performance in serotype identification based on color patterns. Across all storage conditions, the SVM and KNN models consistently outperformed XGBoost in serotype identification accuracy. At room temperature, the KNN and SVM models achieved near-perfect training accuracy (~100%), with testing accuracies of 80–88% in a broader range of samples for identifying and distinguishing Salmonella Typhimurium and Salmonella Enteritidis in mono-serotype, dual-serotype, and serotype cocktail as early as 6 hours. At 4°C, serotype detection of the ML-PCA approach using the KNN and SVM models was achieved as early as 1 day, with testing accuracy exceeding 80%. These results demonstrate that the ML-PCA approach enables non-destructive, continuous, and simultaneous serotype-level detection and differentiation of Salmonella on poultry meat. This methodology offers a potential tool for continuous Salmonella monitoring, improving food safety, and reducing the risk of Salmonella outbreaks.

Technical Abstract: Rapid and accurate detection of Salmonella serotypes associated with foodborne illness is critical for maintaining food safety. This study developed and evaluated the capability of a machine learning-enabled paper chromogenic array sensor (ML-PCA) approach for non-destructive, continuous, and simultaneous detection and differentiation of Salmonella Typhimurium (ST) and Salmonella Enteritidis (SE) in chicken breast in the presence of background microflora during storage at room temperature and refrigerated temperature (4°C). At room temperature, the ML-PCA approach achieved accuracies of 80–88% for detecting ST and SE in mono-serotype, dual-serotype, and serotype cocktail as early as 6'h. At 4°C, ST and SE could be detected as early as 1'd, with testing accuracy exceeding 80%. These results demonstrated that the ML-PCA approach enabled non-destructive, continuous, and simultaneous detection and differentiation of ST and SE on poultry meat. This methodology offers a potential tool for continuous Salmonella monitoring across the food system.