Assessing the growth dynamics of Listeria monocytogenes and microbiota in RTE poultry: A combined BP-ANN and traditional approach

Abstract

Ready-to-eat (RTE) poultry meats are increasingly popular as convenient snacks, but their refrigeration creates conditions that may facilitate the growth of foodborne pathogens, such as Listeria monocytogenes, posing significant food safety risks. To address this issue, a Backpropagation Artificial Neural Network (BP-ANN) model incorporating environmental factors was established to predict the growth of L. monocytogenes and background microbiota (BM) in RTE poultry meats—duck wings (DW), duck tongues (DT), and duck necks (DN)—at 4 °C, 10 °C, 15 °C, and 25 °C. Besides, the traditional predictive model was also developed to simulate the growth dynamics and maximum growth rates (μ_max). The Baranyi model, identified as the best fit based on RMSE (0.28 ± 0.05 log CFU/g) and AIC (5.68 ± 2.47), was employed as the foundation for a competition model incorporating the Jameson effect, demonstrating that BM reached the stationary phase faster, significantly inhibiting L. monocytogenes growth. With increasing temperature, the μ_max of L. monocytogenes rose from 0.05 ± 0.02 to 0.68 ± 0.09 h^-1, while that of BM increased from 0.06 ± 0.02 to 0.77 ± 0.08 h^-1. Notably, DW provided more favorable conditions for microbial growth compared to DN and DT. In addition, the BP-ANN model effectively captured complex nonlinear interactions among temperature, pH, A_w, and meat types, achieving high predictive accuracy (R² = 0.9882). It thus offered a complementary explanation to traditional modeling. Model validation using an independent dataset at 8 °C, 12 °C, and 20 °C confirmed high predictive reliability of developed models, with error margins ranging from 0.2 to 0.5 log CFU/g. These findings provide valuable tools for predicting microbial growth in RTE poultry products, aiding in risk assessment, and informing temperature-dependent storage strategies to improve food safety.