Neural Network Prediction Of Beef Flavor Compounds Produced By Hydrolysis-Maillard Reaction Of Beef Suet Residue

The residue of suet after boiling and extrusion is skimmed tallow,which is rich in protein,and one of its potential high-value utilization pathways is the preparation of beef flavor compounds by hydrolysis and Maillard reaction.However,the reaction is a nonlinear system with a complex process,and there are problems of low accuracy and overfitting if conventional orthogonal tests and response surface methods are used for optimization.For this reason,this paper uses multi-model mechanism neural network(MMM-NN)and curve prediction model(CPM)to study the hydrolysis and Maillard reaction process,as follows:Firstly,based on different hydrolysis conditions and Maillard reaction conditions,the Maillard reaction products(MRPs)were prepared from skimmed tallow as raw material,and the composition of components,molecular weight distribution and flavor characteristics were studied and determined by high performance liquid chromatography,gas chromatography and sensory evaluation.The results showed that the hydrolysate of skimmed tallow contained 17 free amino acids,including a large amount of Glu and Tyr,and the molecular weight distribution of peptides was mainly concentrated in the range of 180 Da to 2000 Da.There were mainly 48 flavor active components in MRPs,among which the most diverse were aldehydes,and the highest contents were ethyl palmitate and nonanal.Accordingly,a process-MRPs volatile component-flavor database was constructed.Secondly,based on the database,a flavor prediction model was constructed using MMMNN.The flavor precursor amino acids in hydrolysate and the flavor active components in MRPs were used as input,and the sensory evaluation scores were used as output to train the model.The results showed that after training,the prediction errors of the model for the sensory evaluation scores of fresh,bitter,meaty,burnt,aromatic and sour flavors of skimmed tallow MRPs were 0.94,0.50,0.96,0.24,0.78 and 0.31,respectively,indicating that the model could predict the sensory evaluation scores accurately by flavor components and reduce the errors caused by subjectivity and sensory fatigue of sensory evaluation.Again,based on the database,molecular sensory science was used to identify the key flavor components,and the relative odor activity value(ROAV)-hydrolysis time course curves of nine key flavor components were fitted and validated using three curve models.The results showed that the ROAV of 2-ethyl-3,5-dimethylpyrazine was the largest and always 100,and the ROAV of ten flavor components,such as nonanal and octanal,were greater than 1.Among the three curve models,no negative values of CPM appeared,and the curve had the most accurate fitting effect through each data point,according to which the changes of ROAV of key flavor components in MRPs with hydrolysis time could be accurately predicted,so that different hydrolysis times can be selected according to flavor requirements in actual production.Finally,17 amino acids were added as auxiliary ingredients for the Maillard reaction,and based on the sensory evaluation scores,the MMM-NN was trained by Latin hypercube sampling with Adam algorithm based on the above two models.The best amino acid combinations were selected from them to regulate the flavor of beef sauce,and MRPs with various flavor requirements were prepared and optimized under the premise of cost consideration.The results showed that no decrease in accuracy occurred after 40,000 iterations,indicating that the model was not overfitted and the prediction was close to the true value(accuracy>99%),and the fitting results showed that the model could accurately predict the sensory evaluation scores according to the actual situation.The flavor orientation of the two beef sauces changed after adding MRPs.In conclusion,the hydrolysis-Maillard reaction process of skimmed tallow was optimized using MMM-NN and CPM,the non-linear relationship between flavor compounds and sensory evaluation during the reaction was examined,and the ROAV-hydrolysis time relationship of key flavor compounds was clarified.which realized the controlled Maillard reaction preparation of beef flavor compounds and provided technical support to improve the utilization and added value of skimmed tallow.