General Regression Neural Network Model for Growth of

General Regression Neural Network Model for Growth of Salmonella Serotypes on Chicken Skin for Use in Risk Assessment Thomas P. Oscar, Ph. D. USDA, ARS Princess Anne, MD

Risk Assessment Data Gaps • Strain variation • Microbial competition • Initial dose • Food matrix

Must have been that chicken! Strain Variation • Salmonella enterica serotypes (> 2, 300) – Top three in chickens are: Enteritidis Typhimurium Kentucky

Strain Variation Autoclaved chicken meat at 25 C J. Food Safety (2000) 20: 225 -236.

Microbial Competition J. Food Prot. (2003) 66(2): 200 -207; (2006) 69(2): 276 -281.

Microbial Competition • Natural Antibiotic Resistance – Bad for public health – Good for predictive microbiology Salmonella Typhimurium DT 104 J. Food Prot. (2006) 69(9): 2048 -2057. J. Food Prot. (2008) 71(6): 1135 -1144. J. Food Prot. (2009) 72(2): 304 -314.

Initial Dose Food Microbiol. (2007) 24: 640 -651.

Regression Modeling Observed W Observed N(t) Observed PI Primary Model Observed m W Tertiary Model Predicted W Model PI Model Predicted PI Secondary Models m Model Primary Model Predicted m Predicted N(t) Observed Nmax J. Food Prot. (2005) 68(12): 2606 -2613. Nmax Model Predicted Nmax

Neural Network Modeling • General Regression Neural Network (GRNN) – Better performance than regression models – User-friendly commercial software • Compatible with Monte Carlo simulation software Jeyamkondan et. al. , 2001

Objective • To develop a GRNN and simulation model for growth of Salmonella on chicken skin as a function of serotype for use in risk assessment. – Short-term temperature abuse (0 to 8 h)

Materials and Methods • Experimental Design (3 x 10 x 5 x 2) – Serotypes (Typhimurium, Kentucky, Hadar) • 30 C in BHIB for 23 h at 150 opm } prehistory – Temperature (5, 10, 15, 20, 25, 30, 35, 40, 45, 50 C) – Time (0, 2, 4, 6, 8 h) – Trial (1, 2) – Sample (a, b) 5 l 7 cfu

Materials and Methods CFU MPN

Materials and Methods Poultry isolates • Plating Media – XLH-CATS for Typhimurium – XLH-NATS for Kentucky – XLH-TUGS for Hadar Ingredients XL = xylose lysine H = HEPES C = chloramphenicol A = ampicillin T = tetracycline S = streptomycin N = novobiocin U = sulfisoxazole G = gentamicin MPN drop plate

General Regression Neural Network Output ŷ Predicted Value -0. 01 N(x) D(x) … … Summation Layer 7. 37 Pattern Layer Distance Function Specht, 1991 S T t Serotype Temp. time Input Layer

Step 1 Enter data

Step 2 Define the data set

Step 3 Set the training parameters

Step 4 Train the GRNN

Step 5 Review results

J. Food Prot. (2006) 69(9): 2048 -2057.

J. Food Prot. (2006) 69(9): 2048 -2057.

Step 6 Predict

Step 7 Integrate with risk assessment

Serotype (%) Temperature ( C) Time (h) Log change Scenario T_K_H Min_ML_Max Correlation Min_50%_Max A 31_58_11 5_20_50 0_2_8 0 -0. 21_0. 09_4. 8 B 31_58_11 5_20_50 0_2_8 -1 -0. 16_0. 04_0. 5

Conclusion #1 • Easy to develop • Low cost • Flexible predictions • Superior performance Neural network modeling outperforms regression modeling in predictive microbiology applications

Conclusion #2 • Cocktail of Typhimurium_Kentucky_Hadar – Overly ‘fail-safe’ predictions for Kentucky.

Conclusion #3 GRNN model was successfully validated for risk assessment model? Data Gaps Strain variation Microbial competition Initial dose Food matrix

Acknowledgements • Thank you for your attention! • Thanks to Jaci Ludwig of ARS and Celia Whyte and Olabimpe Olojo of UMES for their outstanding technical assistance on this project. I hope it was Kentucky!