Rapid quantitative method for detecting lactose content and crystal form proportion based on infrared spectrum

Abstract

The invention relates to the technical field of rapid quantitative and nondestructive detection of food, and particularly provides a rapid quantitative detection method for lactose content and crystal form proportion based on a near infrared spectrum technology, which comprises the following steps: preparing a lactose-whey protein mixture sample; scanning through an instrument to obtain near infrared spectrums of the samples prepared at different gradients; preprocessing the spectrum to remove noise information, establishing a least square support vector machine method model, and screening out optimal interval data of modeling; establishing a prediction model by applying a principal component analysis method; a correlation coefficient value (R2) and a predicted root-mean-square error (RMSEP) are used as key indexes of model robustness test, so that the lactose content and the crystal form proportion in the lactose-whey protein mixture can be rapidly and accurately predicted within a short time (several seconds) on the premise of not damaging a sample. The method has important reference value for regulating and controlling the structure function and application of the whey protein.

Description

Technical Field

[0001] The present invention relates to the technical field of rapid quantitative and non-destructive detection of foods, and specifically relates to a rapid quantitative method for the lactose content and the α, β-lactose crystal form ratio in a lactose and whey protein mixture based on infrared spectrum detection. Background Art

[0002] The lactose content and crystal state in infant formula milk powder have important influences on the structure, function and application of whey protein. The crystalline or amorphous state of lactose is related to the physicochemical property changes of milk powder. Therefore, the physical state of lactose is considered to be one of the important quality parameters during the processing and storage of milk powder. Lactose has multiple crystal forms, and the main configurations are α-lactose and β-lactose, and their physical properties are very different, such as hygroscopicity, solubility, melting point and crystal shape, which all have influences on the structure, function and application of whey protein and milk powder. If the changes in the content of different lactose crystal forms can be monitored, the influence of lactose crystal forms on the functional quality of whey protein and milk powder can be further controlled. Therefore, it has certain application value to develop a rapid quantitative detection method for the ratio of α, β-lactose crystal forms applicable to laboratories and factories.

[0003] The National Food Safety Standard - Infant Formula Food GB 10765-2021 stipulates that the whey protein content in the protein of milk-based infant formula food should be ≥60%, and the lactose should account for ≥90% of the carbohydrate content. The contents of lactose and whey protein in infant formula milk powder must be within the specified ratio range. Therefore, a rapid method for determining the ratio of lactose and whey protein in infant formula milk powder also needs to be established.

[0004] Currently, the commonly used methods for lactose analysis and detection include Dynamic Vapor Sorption (DVS), Differential Scanning Calorimetry (DSC), Powder X-ray Diffraction (PXRD), etc. DVS can be used to determine the amorphous content of a sample containing a mixture of crystalline and amorphous materials. However, when using DVS samples with a low level of amorphous content, large errors may occur due to the moisture adsorbed on the surface. The main drawback of DSC is that it requires crystalline samples because it measures the melting enthalpy associated with the decomposable melting transitions of the head forms of lactose respectively. However, when non-crystalline or amorphous lactose is heated, the powder undergoes a glass transition and then recrystallizes, and DSC cannot distinguish the glass transition and recrystallization peaks of different lactoses. Powder X-ray Diffraction (PXRD) can be used to identify the isomeric crystal types of lactose. However, if the sample is not crystalline, that is, amorphous, PXRD cannot distinguish the isomeric forms of lactose, and the cost is relatively high, which is not suitable for large-scale production.

[0005] In recent years, spectroscopic techniques have been widely used for quantitative analysis in fields such as food and medicine due to their advantages of not requiring pretreatment of samples, not consuming chemical reagents, being fast, accurate, simple to operate, and low in cost, belonging to green detection technologies. The method of infrared spectroscopic quantitative analysis provides an advanced and efficient approach for the quantitative detection of lactose crystal forms. By monitoring the absorption peaks of different crystal forms in the infrared spectrum, we can accurately predict their content in formula milk powder. Applying infrared spectroscopy to quantitatively analyze the lactose crystal form and content in milk powder is simple to operate and low in cost, suitable for large-scale production and application in factories. Summary of the Invention

[0006] The purpose of the present invention is to establish a rapid quantitative method for the lactose content and the proportion of α- and β-lactose crystal forms in the lactose-whey protein mixture through infrared spectroscopy, so as to solve the problems of complexity, time-consuming, and high cost of traditional methods.

[0007] The advantages and preparation methods of this experiment are as follows:

[0008] A rapid quantitative method for different crystal form proportions of lactose and its content in a lactose-whey protein mixture based on infrared spectroscopy detection, comprising the following steps: (1) Design a WPI-lactose mixed solution system containing different whey protein isolate (WPI) - lactose ratios and a WPI-lactose mixed solution system containing different α-lactose / β-lactose ratios; (2) Use a Spotlight400 / 400N Fourier transform infrared / near-infrared imaging system (PerkinElmer, Norwalk, CT, USA) to obtain spectral data, with a scanning range of 5000~900 cm -1 , the number of scans is 20 times, and the resolution is 8 cm -1 ; Take the spectrum of the crystal in air as the background spectrum, and collect the background spectrum once every 2 hours; Keep the experimental temperature at room temperature throughout the experiment; Collect the spectrum of each sample 3 times, calculate the average spectrum and use it for data processing; Analyze the sample data with chemometric software and select the spectral pretreatment method; Use chemometric software to establish a single mid-infrared prediction model respectively; (3) Pretreat the full-band data of the original infrared spectrum by S-G, Der1st, Der2nd, MC, Standard, and MSC methods; (4) There may be correlations between different wavelength variables. Adopt the method of extracting characteristic variables to simplify the established prediction model, and use the PCA method to extract characteristic bands to achieve a more in-depth study of the prediction of lactose crystal form proportion and the ratio of lactose to whey protein; (5) Use R language software to establish a partial least squares (PLS) prediction model. The prediction performance parameters of the model include the correlation coefficient value (R 2 ) and the root mean square error of prediction (RMSEP).

[0009] Compared with the prior art, the beneficial effects of the present invention are as follows: A rapid quantitative detection method for predicting the lactose crystal form ratio and the ratio of lactose to whey protein provided by the present invention is a method capable of on-line monitoring and quickly and accurately measuring the lactose content and crystal form ratio; by artificially changing the ratio to control the content of lactose and different crystal forms in the sample, spectral information is obtained under the condition of ensuring the integrity of the sample, and after screening the characteristic spectral intervals by near-infrared spectroscopy and combined interval partial least squares method, principal component analysis is carried out, and the method of extracting characteristic variables is used to simplify the established prediction model. Finally, a robust prediction model for the lactose content in the lactose-whey protein mixture and a prediction model for the different crystal form ratios of lactose in the α, β-lactose and whey protein mixture are established, so that the lactose content and the ratio of α-lactose to β-lactose content can be quickly and accurately predicted within a short time (a few seconds) without destroying the sample. It can be substituted into the model to predict the different crystal form ratios of lactose in milk powder during the future production application process, and by controlling the lactose crystal form, the physical and chemical properties and structural functions of whey powder can be adjusted, thereby providing a reference value for producing infant formula milk powder with better quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flow chart of a rapid quantitative detection method for the proportion of lactose content and the proportion of two crystal forms of lactose in the lactose-whey protein mixed solution system provided by the present invention;

[0011] Figure 2 It is the mid-infrared (a) original spectrum diagram of the mixed solution system containing different proportions of lactose and whey protein and the mid-infrared (b) original spectrum diagram of the mixed solution containing different proportions of α, β-lactose mixture and whey protein;

[0012] Figure 3 It is the prediction model effect diagram of the single mid-infrared (a) of the mixed solution system containing different proportions of lactose and whey protein and the single mid-infrared (b) of the mixed solution containing different proportions of α, β-lactose mixture and whey protein.

[0013] Figure 4 It is the mid-infrared (a) PCA loading diagram of the mixed solution system containing different proportions of lactose and whey protein and the mid-infrared (b) PCA loading diagram of the mixed solution containing different proportions of α, β-lactose mixture and whey protein;

[0014] Figure 5 It is the prediction model effect diagram of the single mid-infrared (a) of the mixed solution system containing different proportions of lactose and whey protein and the single mid-infrared (b) of the mixed solution containing different proportions of α, β-lactose mixture and whey protein after PCA processing. DETAILED DESCRIPTION OF THE INVENTION

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] A rapid quantitative method for detecting lactose content and crystal form ratio based on infrared spectroscopy, characterized by comprising the following steps:

[0017] (1) Disperse whey protein isolate (WPI9410) in deionized water at about 60 °C for 15 min using a high-shear disperser, then stir (1200 r / min) and recombine in a 200 I stainless steel mixing tank for 2 hours, and then gently stir (500 r / min) overnight at 4 °C. Prepare the lactose solution in deionized water at 70 °C for 2 hours. Prepare whey protein isolate (WPI)-lactose with a total solid content of 20% (w / w) by microtitration method, and use 1 mol / L HCl or 1 mol / L NaOH solution for pH adjustment, with the end point controlled at 6.80 ± 0.05. The pH meter (Mettler Toledo FE28) is used to monitor the mixed solution system in real time during the adjustment process. The mass ratio of WPI to lactose is in the range of 1:6 to 1:9, and 50 equally spaced gradient points are set by linear interpolation method; (2) In the binary system of α-lactose monohydrate and β-lactose anhydrate, set 50 equal ratio gradients according to the crystal molar ratio from 100:0 to 0:100, and compound the above crystal form ratio gradient samples with WPI at a fixed mass ratio of 7:1 (w / w). Lactose is derivatized before mixing with whey protein to prevent the change from the α form to the β form in the solution. Derivatization step: Dissolve dry lactose (1 mg) in the derivatization mixture [containing 22% trimethylsilylimidazole (Fluka), 19.5% dimethyl sulfoxide (Acros Organics) and 58.5% pyridine (Acros Organics), 2.25 mL], and vortex for 2 min; (3) Comprehensively consider the interpretability and prediction accuracy of the model by comparing the correlation coefficient value (R 2 ) and the root mean square error of prediction (RMSEP) of the model; On this basis, construct an optimal quantitative prediction model for lactose and crystal form.

[0018] Example 1

[0019] (1) Spectral data were acquired using a Spotlight 400 / 400N Fourier transform infrared / near-infrared imaging system (PerkinElmer, Norwalk, CT, USA). The environmental temperature was 20 °C and the relative humidity was 45%. The mid-infrared spectral scanning parameters were as follows: the scanning range was 5000 - 900 cm -1 , the number of scans was 20 times, and the resolution was 8 cm -1 ; the spectra of each sample were collected 3 times, and the average spectra were calculated and used for data processing; the final spectral diagrams of α / β-lactose and ordinary lactose-whey protein mixtures are as Figure 2 shown.

[0020] (2) To eliminate the interference of noise and spectral line translation and improve the robustness and prediction accuracy of the model, 6 spectral preprocessing methods were used to optimize the original data. Savitzky-Golay (S-G) can reduce the interference of noise signals on data points and improve the signal-to-noise ratio of spectral data. The number of smoothing points used in this study was 8. First derivative (Der1st) and second derivative (Der2nd) can perform derivative processing on the original spectra, improve the resolution of spectral data, and to a certain extent eliminate the signal overlap caused by combination frequencies and overtones in spectral data. Mean centralization (MC) can enhance the differences between sample spectra, thereby improving the robustness and prediction ability of the model. Standardized spectral preprocessing (Standard) is often used to solve the problems of amplitude drift and dimensional differences. Multiplicative scatter correction (MSC) can eliminate the linear scattering interference of spectral data.

[0021] (3) Partial Least Squares (PLS) regression models are one of the most widely used linear algorithms for solving regression problems. During the modeling process, the spectral data were randomly divided into a training set and a test set at a ratio of 7:3, and the pls (version 2.8-2) in the R package was used for modeling. After the model was established, the leave-one-out cross-validation method and the Jackknife method were used to evaluate the prediction performance of the model to minimize the risk of overfitting. Using Correlation coefficients values (R 2 ) and Root Mean Square Error for predictions (RMSEP) as the model prediction performance evaluation indicators. R 2The closer it is to 1, the closer the predicted value is to the actual value; the RMSEP value is used to measure the deviation between the predicted value and the true value, and the smaller the value, the better the model prediction effect. R 2 The calculation formulas for R and RMSEP are as follows:

[0022]

[0023] In the formula: n is the number of samples in the test set; y i is the true value of the i-th sample; is the average value of the true values, that is f(xi) is the prediction result of the model, representing the predicted value of the i-th sample.

[0024] Table 1 Performance of the full-band PLS model for lactose-whey protein infrared spectra

[0025]

[0026] Table 2 Performance of the full-band PLS model for α,β-lactose-whey protein infrared spectra

[0027]

[0028] From the results analysis in Table 1 and Table 2, it can be seen that different spectral preprocessing methods have different effects on the original spectra of the samples, and there are significant differences in the prediction results of the established PLS models. When no spectral preprocessing is performed on the lactose-whey protein, the prediction result is the worst (R 2 = 0.8067, RMSEP = 0.0032), and when mean centering preprocessing is performed on α,β-lactose-whey protein, the prediction result is the worst (R 2 = 0.8027, RMSEP = 0.1224); as different spectral preprocessing methods are used to denoise the original spectra of lactose-whey protein samples, the results of the PLS models are improved to varying degrees. The best prediction effects are obtained by using the smoothing filter function for preprocessing in the models established for lactose-whey protein and α,β-lactose-whey protein. For the former, R 2 = 0.9899 and RMSEP is 0.0013, and for the latter, R 2 = 0.9606 and RMSEP is 0.0615; compared with the prediction results of the model without spectral preprocessing method, R 2 of the former is increased by 22.71%, and R 2 of the latter is increased by 19.67%, and the results are as Figure 3 shown.

[0029] Example 2

[0030] (1) Selection of spectral characteristic bands Spectral data analysis involves a large number of samples, and the spectral data of each sample is huge, which will cause the spectral matrix to contain a large amount of redundant data, resulting in low spectral analysis efficiency and poor model prediction accuracy. Therefore, Principal Component Analysis (PCA) is used to select characteristic wavelengths. PCA is a common linear unsupervised recognition method that can transform a large number of related variables into a few principal components that retain most of the original data information as much as possible. It has the advantages of not requiring calibration, being applicable to any type of data, especially complex multivariate data sets, etc. PCA is run using FactoMineR (version 2.8) in the R package.

[0031] (2) Based on the PLS prediction model established in Example 1, PCA is used to extract the characteristic spectra of each observation wavelength. The Factor loadings obtained after processing are as Figure 4 shown, where PC1 and PC2 can reflect the Factor loadings scores of PCA for each observation wavelength. When the corresponding score is greater than zero, it indicates that a strong spectral absorption band appears in this region, and this region is selected as the characteristic band for PLS modeling. The results are as Figure 5 shown. From Figure 5 it can be seen that compared with the full-spectrum PLS model, the performance of the model after PCA processing has been improved to varying degrees. For the former, R 2 = 0.9943 and RMSEP is 0.0001; for the latter, R 2 = 0.9668 and RMSEP is 0.0577. Compared with the model prediction results without PCA processing, for the former, R 2 has increased by 0.44%, and for the latter, R 2 has increased by 0.65%. This shows that the application of these two characteristic spectral band selection methods enables the model to better capture the key information in the spectral data and further improves the prediction performance.

Claims

1. A rapid quantitative method for detecting lactose content and crystal form ratio based on infrared spectroscopy, characterized in that: The following steps are involved: (1) Designing WPI-lactose mixed solution systems containing different ratios of whey protein isolate (WPI) to lactose and WPI-lactose mixed solution systems containing different ratios of α-lactose to β-lactose; (2) Preprocessing and extracting characteristic spectra of the collected mid-infrared spectral data to obtain a prediction model for the lactose content and the proportion of different crystal forms in the WPI-lactose mixed solution system; (3) By comparing the correlation coefficient values ​​of the models (R 2 ) and the root mean square error of prediction (RMSEP) were used to comprehensively consider the explanatory power and prediction accuracy of the model; on this basis, the optimal quantitative prediction model for lactose and crystal form was constructed.

2. The method for quantitative nondestructive detection of lactose and its crystal forms according to claim 1, characterized in that: The training model for determining the lactose content and different proportions of α and β crystalline lactose in a WPI-lactose mixture includes the following steps: (1) preparing a WPI-lactose mixed solution system with a total solid content of 20% (w / w), wherein the mass ratio of WPI to lactose is in the range of 1:6 to 1:9, setting 50 equally spaced gradient points by linear interpolation, and adjusting the pH to 6.8 with 1 mol / L HCl or 1 mol / L NaOH solution; (2) In a two-component system of α-lactose monohydrate and β-lactose anhydrate, 50 isocratic gradients were set according to the crystal molar ratio from 100:0 to 0:100, and the above crystal ratio gradient samples were compounded with WPI at a fixed mass ratio of 7:1 (w / w).

3. The method for quantitative nondestructive detection of lactose and different crystal forms according to claim 1, characterized in that: The infrared spectroscopy data of the samples were collected using a Fourier transform infrared spectrophotometer (PerkinElmer, Norwalk, CT, USA) with a resolution of 8 cm -1 Each sample was scanned 20 times, with a scanning range of 5000 to 900 cm -1 Before measurement, the atmospheric spectrum is scanned as background for correction to avoid atmospheric interference and reduce instrument noise, and the spectral data is stored in the form of absorbance.

4. The method for quantitative nondestructive detection of lactose and different crystal forms according to claim 1, characterized in that: In order to eliminate the interference of noise and spectral line shift and improve the robustness and prediction accuracy of the model, the sample spectral data were preprocessed (smoothing filter function, first-order derivative, second-order derivative, mean centering, standardization, multivariate scattering correction) and characteristic spectrum extraction (principal component analysis) using R language software. The smoothing filter function calculated the moving average filter according to the Savitzky-Golay() function, and the filter window size was set to 11; the diff() function was used to calculate the first-order derivative and the second-order derivative, and the lag period of the derivative was set to 10; the scale() function was used to calculate the mean centering and standardization; the msc() function was used to calculate the multivariate scattering correction; the PCA() function was used to calculate the principal component analysis; the "prospectr", "pls", "reshape2", "ggplot2", "randomForest", "caret", "glmnet", "Metrics", and "xxIRT" packages in the R language software were used to calculate the prediction performance.

5. The method for quantitative nondestructive detection of lactose and different crystal forms according to claim 1, characterized in that: Spectral data analysis, modeling and model evaluation, and the use of chemometric PLS to establish a single mid-infrared prediction model.

6. The method for quantitative nondestructive detection of lactose and different crystal forms according to claim 1, characterized in that: R of quantitative nondestructive detection method for lactose and its different crystal forms 2 The calculation formula is as follows: Where n is the number of samples in the test set; yi is the true value of the i-th sample The value is the average of the true values.

7. The method for quantitative nondestructive detection of lactose and different crystal forms according to claim 1, characterized in that: The RMSEP calculation formula for the quantitative nondestructive testing method of lactose and different crystal forms is as follows: Where: n is the number of samples in the test set; yi is the true value of the i-th sample; is the average value of the true value, that is f(xi) is the model prediction result, which represents the predicted value of the i-th sample.

Images