Development and validation of a neural network model for predicting growth of Salmonella Newport on diced roma tomatoes during simulated salad preparation and serving: extrapolation to other serotypes

Authors: Oscar, Thomas

Submitted to: International Journal of Food Science and Technology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 2/26/2018
Publication Date: 3/23/2018
Citation: Oscar, T.P. 2018. Development and validation of a neural network model for predicting growth of Salmonella Newport on diced roma tomatoes during simulated salad preparation and serving: extrapolation to other serotypes. International Journal of Food Science and Technology. 53(7):1789-1801.

Interpretive Summary: A common food handling mistake made in the kitchen of consumers is the failure to separate raw and ready-to-eat food. We developed a model that predicts the growth of Salmonella on diced Roma tomatoes stored for 0 to 8 hours at room temperatures from 61 to 104F. The model was found to provide reliable predictions of the growth of seven other serotypes of Salmonella isolated from chicken parts. Thus, the model can be used with confidence by consumers to assess the safety of salad that might be contaminated with Salmonella and stored at room temperature for an extended period of time before serving.

Technical Abstract: Cross-contamination of diced tomatoes during salad preparation with Salmonella from raw chicken followed by extended holding at room temperature before serving could result in growth of Salmonella and increased risk of salmonellosis. Therefore, a study was undertaken to investigate and model growth of Salmonella on diced tomatoes for the purpose of developing and validating a predictive model for use in risk assessment. Cylindrical portions of Roma tomato pulp with skin (0.14 ± 0.02 g; mean ± SD) in 1.5-ml microcentrifuge tubes were inoculated with a low dose (0.89 ± 0.23 log) of a single strain of a chicken isolate of Salmonella Newport. The inoculated tomato portions (pH = 4.37 ± 0.14) were incubated for 0 to 8 h at 16 to 40C in 4C increments to obtain most probable number (MPN) data for model development. To obtain MPN data to evaluate the model for its ability to interpolate, inoculated tomato portions were incubated for 0 to 8 h at 18 to 38C in 4C increments. An automated, whole sample enrichment, miniature MPN method with a lower limit of detection of one Salmonella cell per tomato portion was used for enumeration. The MPN data were used to develop and validate a multiple-layer feedforward neural network model with two hidden layers of two nodes each. The proportion of residuals (observed - predicted) in an acceptable prediction zone (pAPZ) from -1 log (fail-safe) to 0.5 log (fail-dangerous) was 0.93 (194/209) for dependent data and was 0.96 (86/90) for independent data for interpolation. A pAPZ = 0.7 indicated that the model provided predictions with acceptable accuracy and bias. Thus, the model was successfully validated for interpolation. The model was also successfully validated for extrapolation to seven other serotypes of Salmonella (Montevideo, Hadar, Typhimurium var 5-, 4,5,12:Nonmotile, Kentucky, Thompson, and Heidelberg) but failed validation for extrapolation to three other serotypes of Salmonella (8,20:-:z6, Typhimurium, and Enteritidis).