Maillard reaction freshening of Spanish mackerel umami peptides based on GA-BP neural network

【Objective】 The study aimed to establish a predictive model using genetic algorithm（GA） and multi-layer feed-forward neural network algorithm（BP neural network） to optimize the key parameters in the Maillard reaction process of umami peptides derived from Spanish mackerel by-products, providing a reference for the development of Spanish mackerel flavourings and the promotion of green processing applications of Spanish mackerel resources. 【Method】 Using Spanish mackerel by-products as the raw material, an appropriate amount of D-xylose was added to enhance fresshness through the Maillard reaction. Single-factor experiments were conducted to separately investigate the effects of D-xylose mass concentration, initial pH, reaction time and reaction temperature on the browning value（A_{420 nm}）, final pH and sensory scores of the reaction products. On this basis, a BP neural network was established with D-xylose mass concentration, initial pH, reaction temperature and reaction time as the input layer, and the sensory scores of the products as the output layer, followed by optimization using GA. Amino acid analysis was performed to compare the changes in amino acids before and after the Maillard reaction, analyzing the variation in freshness. 【Result】 The results of the single-factor experiments showed that when the D-xylose mass concentration was 40 g/L, the initial pH was 6.0, the reaction time was 90 min, and the reaction temperature was 120 ℃, the A_{420 nm} value, final pH and sensory scores of the umami peptides derived from Spanish mackerel by-products reached their optimal levels. After 7 iterations using 69 sample groups in the GABP neural network model, the mean square error（MSE） reached a minimum value of 0.005287, and the sample correlation coefficient（R） reached a maximum value of 0.98317, resulting in the most accurate fitting model. The model was analyzed using 18 sample groups and it was found that the R=0.98787 for these samples, indicating that the established GA-BP neural network model could well predict the results of the Maillard reaction under different process parameters.Using this model, the optimal process conditions were obtained: D-xylose mass concentration of 36 g/L, initial pH of 5.4, reaction time of 70 min, and reaction temperature of 119 ℃. Under these conditions, the sensory score of the umami peptides was 9.58, which was close to the predicted value（9.62）. After the Maillard reaction on hydrolyzed amino acids of Spanish mackerel by-products umami peptides, content of umami amino acids increased, especially the glutamic acid content, which rose from 56.21 mg/g to 70.39 mg/g, an increase of 25.23%. The sweet amino acids increased from 103.98 mg/g to 155.64 mg/g, an increase of 49.68%. Conversely, most free amino acids decreased after the Maillard reaction, with a loss rate of 27.76%. 【Conclusion】 The Maillard reaction freshness enhancement process optimized based on the GA-BP neural network model can greatly improve the freshness characteristics of umami peptides derived from Spanish mackerel by-products.