Rapid Identification And Quantitative Detection Of Fish Allergen Parvalbumin Based On Mid-infrared Spectroscopy

With the development of the social and economic level,the improvement of people's quality of life,fish and their products are increasingly popular because of their rich nutrition and delicious taste.However,fish is also one of the eight food allergens,and the food allergic reaction caused by eating fish is mainly composed of parvalbumin?Pa?.Some existing detection methods,such as PCR,mass spectrometry,ELISA,etc.,have disadvantages such as high cost,time consuming,and lack of comprehensiveness.Therefore,a method for detecting Pa quickly,accurately and comprehensively is significance.Multi-molecular vibrational spectroscopy enables fast,non-destructive and multi-component simultaneous detection of mixed samples in complex systems,not only to obtain the overall information profile of the sample,but also to obtain the spatial configuration information of the sample.Currently,in the food,materials,and traditional Chinese medicines and other fields this method have been widely used.And mid-infrared spectroscopy used in this research is one of them.Pa were extracted and purified by trichloroacetic acid?TCA?precipitation method,and then the group A,B,and C are made up by Pa of 16 kinds fish,the crude protein,and the crude protein without Pa,their infrared spectrum are collected.The original spectrum and second derivative spectrum are vectorized,the recognition model of Pa is established by a variety of algorithms,such as support vector machine?SVM?,random forest?RF?and neural network?IRN?,to complete the identification of Pa.Using the characteristic information obtained by qualitative detection is used to construct a quantitative model for Pa,and finally integrating all the data into the database to form a Pa infrared spectroscopy database,the detailed information is shown as follows:1.Pa has a Ca²⁺binding site belonging to the EF family,and the structure is resistant to acid.Due to multiple TCA precipitation,the electrophoretic bands of Pa obtained show a distinct band,and Pa are also specifically immunoreactive with the reported EG8antibody.2.Use the purified Pa of fish,its crude protein,and crude protein without Pa as the group A,B and C,each group has 136 samples,each sample has 12 replicates.Every group is divided into training sets and test sets by 7:3.Three models of SVM,RF and IRN are trained.Their parameters are optimized,their final form shown as follows:SVM adopts Gaussian kernel function with penalty factor of 5 and step size of 1.The number of decision trees of RF model is 60,the maximum depth of each tree.is 8.The IRN model has 9 layers,the hidden layer uses the structure of Inception-Resnet-V2,the dropout takes0.75,and the softmax uses logits.The accuracy of the three models on the test set samples is the accuracy of the three models in the group level?A,B,C?is IRN?92.5%,85%,95%?,SVM?75%,85%,75%?,RF?60%,75%,85%?,the accuracy of the three models is the overall level is IRN?97.3%?,SVM?83.3%?,RF?93.3%?.In conclusion,the IRN model is optimal.3.The consensus sequence of Pa got by blast-p,which is mainly composed of Asp,Ser,Glu and Asn.Compared with the infrared characteristic absorption of these four amino acid residues,there are three models of the Pa characteristic spectral region.It can be seen that the IRN model has a higher extraction ratio than the other two models,it was consistent with the internal verification results.At the same time,the obtained characteristic regions were 1644,1543,1265,1240,1155,1051,1010±5 cm^-1,and the peak intensity was different in these regions of the second derivative spectrum belong to different sets.A sample set for external validation of those models was prepared by adding different amounts of Pa to the surimi.The verification results for the three models were:IRN?91%?>RF?70%?>SVM?35%?,and the detection limitation of IRN model for Pa is 0.1⁰.3mg/kg.4.Quantitative prediction model of Pa content in surimi was established by infrared spectroscopy.The Pa content of the calibration group used for model construction was 0,5,10,20,30,40,50,and 60mg per 100 g surimi.This model was based on the linear relationship between the characteristic peak obtained by the qualitative model and Pa content.The results model shows that R²>0.99 and a standard deviation of 2.75.Then,the model was externally verified using samples with Pa contents of 5,15,25 and 35 ug,and the predicted and true values were verified by t test.The results show that when the content of Pa was>0.25mg,the significance P value is greater than 0.05,indicating that the model can accurately predict the content of Pa.5.Integrate the basic information of 16 species of fish,as well as the original and second derivative spectra of the corresponding allergens Pa infrared spectrum,store all the data in the MySQL database,and use Django to build the database's visual framework and the relevant query interface,upload all files into the cloud server,and bind the domain name www.guominyuan.online to the server.Finally,get an open Pa database.