Fingerprinting Technique Coupled With Pattern Recognition Applied In The Determination Of Milk Quality

The computer science provides a reliable guarantee on biological science.The development of fingerprinting technique provides a technical guidance for the research of biological sample components including protein and lipid and technical mean to detecting biomarker.Therefore,computer science algorithm can be used to analysis the fingerprinting and dig the information of the measured data.The integrity of data acquisition was obtained by fingerprinting technique.Furthermore,the difference and internal rules of fingerprinting were performed by statistical analysis method.There have been many targeted approaches to detect milk adulteration in the dairy industry.At present,the existing methods were used to detect a certain kind of adulterated substances.The milk compound has been tested on each if the adulterants were unknown.The detection method was very time-consuming and labor-intensive for substantial milk samples to be determined.Moreover,the reconstructed milk samples could not be detected by the targeted method.Therefore,based on the above problems,the fingerprinting technique combined with pattern recognition was evaluated for the rapid detection of economically motivated adulteration?EMA?of milk.In the study,the research objects were milk adulterated with other plant protein and reconstructed milk.The milk adulteration was detected by flow injection mass spectrometric,gel electrophoresis,attenuated total reflectance Fourier transform infrared?ATR-FTIR?spectral,ultra-high liquid chromatography-mass spectrometry?UPLC-Q-TOF-MS?,three-dimensional endogenous fluorescence spectroscopy,and two-dimensional exogenous fluorescence spectroscopy from intact protein and peptide fingerprinting of all samples combined with pattern recognition method.The objective of this research will provide a new method and idea for milk adulteration.The main results of this study were as follows.?1?In terms of milk adulterated with plant and whey protein,flow injection mass spectrometry?FIMS?and gel electrophoresis combined with patternrecognition were evaluated for the rapid detection of exogenous protein of milk based on the intact protein.Two possible proteins around the average masses of18,277.5 and 18,362.5 Da were found.The peaks were tentatively identified as?-lactoglobulin B?UniProt Entry ID:P02754,theoretical average mass:18,281.2 Da?and?-lactoglobulin A?theoretical average mass:18,367.3 Da?,respectively by the molecular weight determination?MoWeD?algorithm.Moreover,PLS-DA and SVM classification models were built to evaluate their ability for automated detection of adulterated milks.The PLS-DA and SVM modeling resulted a prediction accuracy of 100%.FIMS intact proteinscombined with chemometrics was demonstrated to detect possible adulteration by soybean,pea,and whey protein isolates in milk within 2 min per sample at levels as low as 0.5%?w/v?.Identifying the type of adulterant is helpful intracing the possible sources of economically motivated adulterants?EMA?.Therefore,multi-class SVM and PLS-DA were evaluated.In comparison to PLS-DA and BP-ANN,SVM showed a better classification performance in differentiating authentic and adulterated samples.SVR model was to quantify the adulterated levels of each fraudulent protein in milk.Moreover,the models of soybean and whey protein adulterants outperformed other models with high coefficients of determination(R²_{test set}>0.8000)and low prediction error.Gel electrophoresis combined with PCA analysis can successfully discriminateadulterated and authentic milk samples.The limit of detection was 5%.?2?With regard to reconstructed milk detection,attenuated total reflectance Fourier transform infrared?ATR-FTIR?spectral fingerprint combined with chemometrics was established to identify pure milk from their counterparts adulterated with powdered milk.Five algorithms including random forest?RF?,naive Bayes?NB?,soft independent modeling of class analogy?SIMCA?,and partial least-squares-discriminant analysis?PLS-DA?with statistically unbiased performance estimation under the model population analysis framework?MPA?were compared to discriminate reconstructed milk from fresh counterparts.To evaluate the performance of a classifier,ten metrics were used to assess overall effectiveness of the classifiers.The support vector machine?SVM?respectively yielded 95,92,98,91,and 95%for average accuracy,sensitivity,specificity and Matthews correlation coefficient in the test set without complex pretreatment,a best performance among all classifiers.These pretreatment methods were multiplicative scatter correction?MSC?,standard normal variate?SNV?,Savitzky-Golay smoothing,Savitzky-Golay processing first derivatives and second derivatives.The results indicated the classification accuracy of PLS-DA was 100%and high coefficients of determination of PLSR was 0.8846 and low prediction error using ATR-FTIR by SNV preprocessing.?3?The reconstructed milk was identified using proteomics method on the basis of intact protein and peptide fingerprints of milk samples.The classification accuracy of reconstructed and fresh milk was 100%using PLS-DA and SVM models combined with intact protein and peptide fingerprints of milk samples,respectively.A low-level and mid-level data fusion approaches were adopted for the joint analysis of intact protein and peptide fingerprints of milk samples.The results indicated that the fresh and reconstructed milk could be discriminated by mid-level data fusion approach using PLS-DA and PLSR combined with PCA analysis,the limit of detection was 0.5%.?4?The reconstructed milk detection methods were established by three-dimensional endogenous fluorescence spectroscopy and two-dimensional exogenous fluorescence spectroscopy combined with chemometrics.After obtaining three-dimensional endogenous fluorescence of milk samples,the corresponding three-dimensional spectroscopy and contours were plotted by MATLAB.From perspective of visualization,there was an obvious difference between fresh milk and milk adulterated with 5%milk powder.Moreover,three-dimensional endogenous fluorescence combined with parallel factor analysis.It was found that fresh milk and reconstructed milk was different from projection score.There was partial overlap of fresh milk and reconstructed milk.Therefore,combined with classification models,SVM accuracy reach 95.83%.Next,for two-dimensional endogenous fluorescence spectroscopy of milk samples,the fluorescence signals of fresh and reconstructed milk increased with different extent after fluorescent dye added into milk samples.Two-dimensional exogenous fluorescence spectroscopy combined with PCA could discriminate fresh and reconstructed milk.Moreover,the maximum emission wavelength of loading from PC1 and PC2 was 450⁵00 nm.Moreover,two-class SVM and PLS-DA model coupled with two-dimensional exogeneous fluorescence spectra yielded100%accuracy.In summary,in our study,we built milk adulteration method using mass spectrometry and optical spectrum fingerprints combined with pattern recognition.This research can provide data base for milk adulteration and a new idea for constructing big data platform of milk quality control.