Evolutionary combined neural networks for modelling the growth boundaries for a five strain Staphylococcus aureus cocktail against temperature, pH and water activity A. Valero 1* , F. Pérez-Rodríguez 1 , E. Carrasco 1 , C. Hervás 2* , P. Gutierrez 2 , J.C. Fernández 2 , R.M. García-Gimeno 1 , and G. Zurera 1 1 University of Cordoba, Department of Food Science and Technology. Campus de Rabanales s/n. Edif Darwin C1. 14014, Córdoba (Spain) (bt2vadia@uco.es) 2 University of Cordoba, Department of Computing and Numerical Analysis. Campus de Rabanales s/n. Edif Einstein C2. 14014, Córdoba (Spain) (chervas@uco.es) Abstract Staphylococcus aureus contamination of foods is one of the most prevalent causes of gastroenteritis worldwide, which is caused by ingestion of preformed toxins. Ready-to-eat foods without previous heat treatment before consumption are susceptible to be contaminated during processing, storage or handling at both retail points and domestic homes. In this study, a logistic regression based on product units neural networks (LRPU) was carried out to determine the probability of growth of a five strain S. aureus cocktail as a function of storage temperature (8-19ºC), pH (4.5-7.5) and water activity (a w ) (0.856-0.999). A good adjustment of the LRPU model was observed for both training and test datasets (91.10% and 89.36% cases correctly classified respectively). Results confirmed the tolerance to low levels of a w of the working cocktail, since it was capable to grow at 19ºC up to a w levels of 0.867 and at neutral pH (7.0). The storage temperature had a marked effect on S. aureus cocktail growth, since at 8ºC, it was only able to grow at a w levels higher than 0.977. Regarding pH effect, at pH levels below 5.0, probability of growth was lower than 50 % in most cases at temperatures lower than 16ºC. This study provides a solid scientific basis that verifies the criteria established by food industries for the assurance of the microbiological safety of the product until the consumption phase, currently demanded by the European legislation. Keywords: growth/no growth, Staphylococcus aureus, classification, product unit neural networks, food safety Introduction Food poisoning caused by staphylococcal contamination is one of the most prevalent causes of gastroenteritis worldwide, which is caused by the ingestion of food that contains preformed toxins (Jablonski and Bohach 2001). Specifically, S. aureus offen produces the most common types of food intoxication (Jablonski and Bohach 2001). The number of illnesses reported to the Spanish Microbiological Information System (SIM) caused by S. aureus ingestion, were increasing during the last five years, until reaching more than 550 annual cases (http://www.isciii.es/jsps/centros/epidemiologia/informacionMicrobiologica.jsp ). Ready-to-eat foods without a previous heat treatment are susceptible to be contaminated during processing, storage or handling in both retail points and domestic homes (Huang et al., 2001). The key to controlling S. aureus is an understanding of the factors that influence on its growth in foods and the modification of these factors in order to limit potential risks (mainly temperature, pH and water activity). Probability models are characterized by defining the growth/no growth limits of a specific microorganism in a medium as a function of some environmental factors, in a very limited range of conditions. There are several studies reported in the literature regarding the use of logistic equations to describe growth/no growth boundaries (Lanciotti et al., 2001). However, when a strong interaction exists between the variables considered, the use of Product Unit Neural Network models (PUNN) is gaining attention, due to they are more effective in detecting those interactions and they have the ability to implement higher order functions (polynomial) as a particular case (Gurney 1992). The use of PUNN models integrated as