Modeling by artificial neural networks of silver carp (Hypophthalmichthys molitrixi) with sous vide processing on the effects of storage and processing temperatures on the microbiological status

To evaluate and anticipate the microbial changes of silver carp (Hypophthalmichthys molitrixi) during cold storage (0, 5, 10, 15 & 21 day) at different sous vide processing temperatures (60, 65, 70, and 75 °C), changes in microbial load of Enterobacteriaceae, Lactic Acid bacteria (LAB), Pseudomonas, Psychrotrophs, and total viable count (TVC) were considered. A radial basis function neural network (RBFNN) model was established to predict the changes in the microbial content of silver carp. The critical temperature for inactivation of Enterobacteriaceae and lactic acid bacteria was 65 °C and for Pseudomonas and Psychrotrophs was 70 °C and the highest value (75 °C) was observed for the total viable count. In samples processed at 75 °C, the presence of Enterobacteriaceae, Pseudomonas and Psychrotrophs were not detectable up to 15 days of storage and lactic acid bacteria were not detectable even at the end of the storage period. The optimal ANN topology for modeling Enterobacteriaceae, Pseudomonas, and Psychrotroph contained 9 neurons in the hidden layer, but for TVC and LAB, it was 14 neurons.


49
Introduction 50 Silver carp popularity as a source of food has been notable and this aquatic fish is also used to 51 prevent and clear algal blooms. The silver carp has become popular worldwide because does 52 not require supplementary feed. As a result, silver carp (Hypophthalmichthys molitrix) is one 53 of the most cultured fish species in the world due to its desirable properties such as high 54 nutritional content, fast growth, high feed efficiency, and easy cultivation [1]. Unfortunately, 55 the food safety and spoilage of fishery products is a huge concern because of microbial cross-56 contamination from various sources. As a consequence, because of its neutral pH and high 57 water content, the shelf-life of fish declines rapidly. All of these aspects enhance the need to 58 increase the shelf life and many preservation applications have been taken into consideration 59 including sous-vide. 60 Sous vide is a French word meaning "under vacuum", and sous vide cooking is a processing 61 method in which vacuum-packed foods in heat resistant pouches are cooked under controlled 62 temperature and time [2]. The advantages of sous vide cooked products are that these products 63 are not preserved by low water activity or pH and do not contain preservatives. The food safety 64 of these types of products relies on high heat treatment and cold storage [3]. In this technique, 65 the products are immersed in a hot water/steam oven for a longer time than conventional 66 cooking followed by immediate cooling to ≤ 4 °C [4]. Recently, sous-vide processing was 67 applied and combined with irradiation for mackerel fillets [5]. They concluded that the 68 microbial count of mesophilic and psychrophilic treated by sous vide with irradiation never 69 exceeded the standard limit. The sous-vide processing of ready-to-cook seer fish steaks was 70 optimized by response surface methodology, and they reported the optimized process condition 71 as 3.5% salt concentration, process temperature of 89 °C, and cooking time of 13.5 min.

89
Fifteen sacrisfied silver carp sample (average weight and length of 790±20 g and 283±16 mm, 90 respectively) were obtained from a local warm-water fish farm located in Rasht, in the north of      116 The experimental data were used for developing ANNs where the independent variables were 117 temperature (°C) and storage period (day) and the dependent variable was a microbial count 118 (Log CFU g -1 ). For developing ANNs, 70, 15, and 15%of experimental data were randomly 119 selected for training, cross-validation, and testing, respectively. Multilayer perceptron (MLP) 120 and ANNs were used for modeling the microbial count ( Fig. 1) In the procedure of finding the 121 best ANN structure, a different number of neurons (1 to 15) in the hidden layer was analyzed.

122
For each microorganism, a separate optimal ANN was developed. The training algorithm of 123 ANNs was Levenberg-Marquardt (LV) backpropagation (BP). The transfer function in the 124 hidden layer was Tangent-Sigmoid (tansig) and for the output layer, it was linear. Three 125 replicates were carried out for each experimental run. Two statistical error criteria were taken 126 into account for checking the goodness of fit for each ANN topology. These criteria were the 127 coefficient of determination (R 2 ) and root mean square error (RMSE): where, y pre and y exp indicate the predicted and experimental dependent variable (microbial  Results and discussion 145 The effect of the temperature on the survival of microorganisms is seen in Fig. 2 microbial quality of samples for two weeks. As it is seen in Fig. 3, the presence of 186 microorganisms was not detectable at severe processing temperatures for a specific period 187 (varied depending on the process temperature and type of microorganism) but, after that, they it is recommended to repeat the run and report the average R 2 and RMSE for each network with 207 a different number of neurons in the hidden layer. In this study, the run was repeated five times, 208 and, based on the average R 2 and RMSE, the best ANN was obtained for each group of bacteria 209 (Fig. 4).

210
For Enterobacteriaceae, Pseudomonas, and Psychrotroph the best ANN was obtained when the 211 number of neurons in the hidden layer was 9, but, for LAB and TVC, this was 14 neurons. Fig.   212 5 shows the correlation between predicted (by the optimum network) and experimental data. Simulated results of ANN were used to generate the contour plots of microbial growth during 219 storage (Fig. 6). These graphs provide useful and comprehensive information, especially for 220 optimization purposes. From these graphs, for each group of microorganisms, the microbial 221 load in silver carp fillets processed at a temperature between 60 to 75 °C storage up to 21 days 222 was obtained.  Pseudomonas, Psychrotrophs, and total viable count (TVC)) of control and processed samples. Lactic Acid bacteria (LAB), Pseudomonas, Psychrotrophs, and total viable count (TVC)).

330
Predicted microbial counts were obtained by the best ANN topology.