MIR Rapid Batch Detection Method for Lactoferrin Content in Milk

Abstract

The present invention belongs to the field of dairy cow performance measurement and milk quality detection, and particularly relates to a rapid batch detection method for the content of lactoferrin in milk by MIR. In terms of the selection of characteristic bands, the applicant breaks the common practice of using algorithms to screen features, but instead uses the method of manual selection + multiple traversals. Finally, more characteristic bands and wave points are selected for modeling, and the range is wider. 437 characteristic wave points are screened out. At the same time, the spectral MIR of the second measurement of the same milk sample is selected for modeling, which improves the model accuracy of the first spectral measurement data modeling. Finally, the optimal combination of data preprocessing methods and model algorithms is screened out, and the optimal parameters are determined, improving the accuracy of the model. The method of the present invention can rapidly, accurately and low-costly detect the content of lactoferrin in dairy products, realizing rapid batch detection, and will be widely applicable to dairy cow performance measurement and milk quality detection.

Description

Technical Field

[0001] The present invention belongs to the field of dairy cow performance determination and milk quality detection, and particularly relates to a rapid batch detection method for the content of lactoferrin in milk. Background Art

[0002] Lactoferrin (LF) is an iron-sugar binding protein transformed from transferrin, and mainly exists in the milk of humans, cows, etc. [1-3] . Lactoferrin is an active ingredient in cow's milk. The content of lactoferrin in colostrum is about 0.8 g / L, and the average concentration in normal milk is about 0.3 g / L. [4-5] . Lactoferrin is one of the four main proteins in human milk, accounting for about 20% of the total protein in human milk. The amino acid sequence homology between bovine lactoferrin and human lactoferrin reaches 69%, and they have similar biological activities. Cow's milk is one of the main sources of lactoferrin that is most easily obtained. The "National High-quality Milk Project" clearly puts forward the index requirement for the bioactive substances in high-quality pasteurized milk as lactoferrin ≥ 25 mg / L. Lactoferrin can be used as a protein source of amino acids and is beneficial to the biological utilization of iron. It has the functions of resisting bacteria and killing bacteria [6-7] , and has a certain preventive and adjuvant therapeutic effect on children's diseases such as children's diarrhea and children's Helicobacter pylori infection, and has an improving effect on infantile anemia. However, due to the short shelf life of raw milk and pasteurized fresh milk, it is necessary to perform rapid detection on them, and there is no rapid detection method in China at present [8] , so it is urgent to establish a rapid batch determination method for lactoferrin in milk.

[0003] At present, the determination methods of lactoferrin mainly include spectrophotometry, high performance liquid chromatography, enzyme-linked immunosorbent assay, high performance capillary electrophoresis, surface plasmon resonance technology, etc. Among them, the most commonly used are high performance liquid chromatography and enzyme-linked immunosorbent assay. High performance liquid chromatography has high purity requirements, takes a long time to purify and concentrate lactoferrin, has cumbersome operations, expensive instruments, and high technical requirements for experimental personnel. Enzyme-linked immunosorbent assay (ELISA) has good applicability for the detection of lactoferrin and is well recognized by the dairy-related industries. However, it needs to be diluted many times before measuring samples, and the probability of error increases due to the large dilution factor, and the used kits are expensive. Therefore, it is not suitable for widespread use in large batches in production practice.

[0004] Mid-infrared spectroscopy (MIR) is a rapid and cost-effective tool for recording phenotypes at the population level [9]. Generally, the infrared band of 2.5 - 25 μm is defined as the mid-infrared region. The mid-infrared spectrum is an absorption band caused by the vibration of specific functional groups, which is suitable for the identification of organic compound structures. The band density is proportional to the number of functional groups and can be used for quantitative analysis. As an analytical tool, infrared spectroscopy is increasingly used in different fields of animal production. During the period from 1990 to 2000, research was carried out on using MIR to detect the contents of proteins, triglycerides, metabolites, etc. in blood. [10-11] , and foreign scholars explored using MIR to analyze fats, proteins, lactose, etc. in milk. Although foreign research on detecting milk components based on MIR has started, there are problems such as low accuracy and inaccurate characteristic bands; while in China, the research on new technologies for detecting milk components using MIR is still in its infancy, and there is no report on lactoferrin.

[0005] Compared with other countries, due to the influence of domestic climate, geographical environment, feeding conditions, etc., Chinese dairy cows are quite different from foreign dairy cows, and the milk quality also has its own characteristics. Foreign detection models may not necessarily be suitable for Chinese dairy cows.

[0006] To address the above problems, the present invention uses a variety of pretreatment methods to analyze the measured milk MIR data, constructs a prediction model for lactoferrin in milk, and establishes a method for rapidly determining the content of lactoferrin using MIR, providing a reference basis for the establishment of a rapid, batch, and non-destructive detection technology for lactoferrin components in milk with domestic intellectual property rights based on MIR and the genetic research of milk components. Summary of the Invention

[0007] The purpose of the present invention is to provide a rapid batch detection method for the content of lactoferrin in milk using MIR, which is simple, fast, and has a high accuracy compared with the true value.

[0008] To achieve the above purpose, the present invention takes the following technical measures:

[0009] A rapid batch detection method for the content of lactoferrin in milk using MIR includes the following steps:

[0010] 1. The characteristic bands in the infrared spectrum of the milk sample collected are: 945.21 cm -1 -1373.45 cm -1 , 1755.39 cm -1 -1797.83 cm -1 , 2010.02 cm -1 -2318.66 cm -1 , 2384.24 cm -1 -2600.29 cm -1 , 2731.46 cm -1 -2924.36 cm-1 with the MIR data in 3622.66 cm -1 - 4112.63 cm -1 ;

[0011] 2. Substitute the obtained MIR data into the SNV + PLS - DA (n_component = 16) model to output the predicted results of the lactoferrin content.

[0012] In the above - mentioned method, preferably, there is a difference of two wave points allowed before and after each segment in step 1.

[0013] The protection scope of the present invention also includes: the above - mentioned method is used for detecting the lactoferrin content in milk.

[0014] Compared with the prior art, the advantages of the present invention are as follows:

[0015] 1. In terms of the selection of characteristic bands, instead of using the commonly used algorithm to screen features, an artificial manual selection + multiple traversal method is used. Finally, more characteristic bands and wave points are selected for modeling, with a wider range. 437 characteristic wave points are screened out for the establishment and optimization of the lactoferrin prediction model.

[0016] 2. The spectral MIR of the second measurement of the same milk sample is selected for modeling, improving the accuracy of the model established from the data of the first spectral measurement.

[0017] 3. The optimal combination of data pre - processing methods and model algorithms is screened out, and the optimal parameters are determined, improving the accuracy of the model.

[0018] 4. It can quickly, accurately and at low cost detect the lactoferrin content in milk products, realizing rapid batch detection. The measurement time for each sample is only 10 - 15 seconds, improving the detection efficiency, with strong practicability, and will be widely applied to the determination of dairy cow performance and the detection of milk quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the mid - infrared spectrogram (a) and the average spectrogram (b) of the untreated milk sample.

[0020] Figure 2 It is the mid - infrared spectrogram of the milk sample after SNV pre - processing.

[0021] Figure 3 It is the overall spectrogram (a) of six selected characteristic bands and the enlarged spectrogram (b) of each characteristic band.

[0022] Figure 4 It is the correlation and fitting straight - line graph between the true value and the predicted value of the model milk data. DETAILED DESCRIPTION OF THE INVENTION

[0023] The technical solutions described in the present invention are conventional solutions in the art unless otherwise specified; the reagents or materials are from commercial channels unless otherwise specified.

[0024] 1. Experimental materials

[0025] The test materials were sourced from Chinese raw milk and pasteurized fresh milk. Milk samples were collected over a two-year period (August 2019 - August 2021). The samples were representative as they included all four seasons. There were a total of 289 samples. 40 ml of each milk sample was collected and dispensed into new cylindrical sampling bottles with a diameter of 3.5 cm and a height of 9 cm. They were numbered in sequence, and 2-bromo-2-nitro-1,3-propanediol preservative was immediately added to each sampling bottle. The bottles were gently shaken to dissolve the preservative completely, and spectral data collection was carried out immediately. Spectral data of all milk samples were collected twice.

[0026] 2. Mid-infrared spectroscopy measurement and collection

[0027] The sample was poured into a cylindrical sample tube with a diameter of 3.5 cm and a height of 9 cm, and water-bathed in a 42 °C water bath for 15 - 20 min. The MilkoScan 7RM milk composition detector from FOSS, Denmark, was used. The solid optical fiber probe was inserted into the liquid, and the sample was mixed evenly before scanning. TM 7RM milk composition detector, the solid optical fiber probe was inserted into the liquid, and the sample was mixed evenly before scanning.

[0028] 3. Detection method for the true (reference) value of lactoferrin in milk

[0029] The lactoferrin content in the sample was determined with reference to T / TDSTIA 006 - 2019 Determination of Lactoferrin in Milk and Dairy Products - Liquid Chromatography Method.

[0030] Example 1:

[0031] Selection of the lactoferrin prediction model algorithm:

[0032] Since ridge regression is one of the effective algorithms for linear regression, when establishing a regression equation, ridge regression adds regularization constraints to achieve the effect of solving overfitting. There are two types of regularization, namely l1 regularization and l2 regularization. The advantages of l2 regularization compared to l1 regularization are: (1) It can perform cross-validation. (2) It implements stochastic gradient descent. Ridge regression is a linear regression model with l2 regularization added. It retains the advantages of linear regression, meets the requirements for model establishment, and the results are relatively stable. It is one of the effective algorithms. Therefore, this embodiment selects this algorithm as a candidate. The partial least squares regression algorithm is one of the very effective algorithms for multi-feature samples. Each sample in the mid-infrared spectrum data corresponds to 1060 wave points, which is representative of multi-feature samples. At the same time, the partial least squares regression algorithm rarely exhibits overfitting. Therefore, many researchers of mid-infrared spectra choose to use the partial least squares regression algorithm to establish models. Thus, this algorithm is also used as a candidate algorithm for modeling.

[0033] This application uses two algorithms, namely ridge regression (Ridge) and partial least squares regression (PLSR), to establish two models respectively, compares and analyzes the prediction capabilities of these two models, and screens out a suitable algorithm.

[0034] Example 2:

[0035] Selection of the number of mid-infrared spectrum measurements:

[0036] All samples used in this application have been subjected to spectral acquisitions twice in a row. The purpose is to screen out the most effective MIR data for modeling by comparing the effects of different measurement times of the same sample and the three types of MIR data obtained (the first time, the second time, and the average of the two times) on the modeling accuracy. Since some researchers believe that the MIR measured at different times may affect the modeling accuracy, in this embodiment, the first time, the second time, and the average of the two spectral MIRs after removing the water absorption band are respectively modeled, and the accuracy of the models is compared and analyzed. The results are shown in the following table:

[0037] Comparison results of the Ridge algorithm:

[0038]

[0039] Comparison results of the PLSR algorithm:

[0040]

[0041] After comprehensively considering the comparison results of the two algorithms, the second spectral MIR is finally selected for model establishment.

[0042] Example 3:

[0043] Establishment of a method for detecting the content of lactoferrin in milk by mid-infrared spectrum:

[0044] 1. Division of the Modeling Dataset

[0045]

[0046] In the division of the modeling dataset in this embodiment, 80% is the training set and 20% is the test set. The ratio of the training set to the test set is 4:1. At the same time, the training set is also called the cross-validation set, and 10-fold cross-validation is performed during the process of training the model.

[0047] 2. Screening of the Preprocessing Method for Modeling MIR Data

[0048] Effective feature screening is a basic operation for processing spectral data, aiming to eliminate noise and lay a good foundation for feature extraction. There are mainly three types of effective feature screening: feature extraction, feature preprocessing, and feature dimensionality reduction. In this embodiment, five processing methods, namely SG (convolution smoothing), MSC (multiplicative scatter correction), SNV (standard normal variate transformation), diff1 (first-order difference), and diff2 (second-order difference), are mainly used to perform feature preprocessing on the spectral data.

[0049] 3. Manual Selection Process and Determination of the Modeling Feature Bands

[0050] There are many methods for selecting feature bands, mainly including two types: algorithm-based feature selection and manual feature selection. The principle of algorithm-based feature selection mainly comes from the correlation between each wave point and the reference value. Its advantages are fast speed and high efficiency, but its disadvantage is that it ignores the cooperative effect between adjacent wave points and the thinking is relatively single; the advantage of manual feature selection is that the role of the bands (i.e., adjacent wave points) can be strengthened during the selection process, and at the same time, more of the original information state of the spectrum can be retained during the process of improving the model, with stronger inclusiveness and generalization ability, and the selected bands are accurate. The disadvantage is that the selection speed is slow and the efficiency is low.

[0051] In this embodiment, the method of manual selection is used to select the feature bands, and the selection steps are as follows:

[0052] (1) Determine the basic algorithm. As can be seen from Example 2, the overall effect of the partial least squares regression algorithm is better. Therefore, the partial least squares regression algorithm is finally selected as the lactoferrin prediction algorithm.

[0053] (2) Determine the best preprocessing combination. The mid-infrared spectrum of the sample is removed from the water absorption band, and the preprocessing in Example 3 is carried out and compared. Finally, the SNV preprocessing method is selected (the results are as follows in the table, Figure 2 ).

[0054]

[0055] The manual selection process of the modeling feature bands is as follows:

[0056] (1) Remove 1593.35 cm -1 -1709.1 cm -1 and 3059.39 cm -1 -3641.95 cm -1 Two spectral regions related to water absorption.

[0057] (2) Remove 4500 cm -1 -5011.54 cm -1 section of the spectral region because this region is completely outside the mid-infrared absorption range.

[0058] (3) Divide the remaining region into six sections. The region less than 1593.35 cm -1 is the first section, and the region greater than 3641.95 cm -1 is the last section. The band between 1709.1 cm -1 and 3059.39 cm -1 is evenly divided into four sections.

[0059] (4) Taking 50 wave points as a group, using the partial least squares regression algorithm, first add or subtract a group of wave points at both ends of the critical point of the first band to find the optimal effect, and perform a similar operation on the second band based on this. Finally, after all six bands have completed one round of operation, it is considered that the first traversal is completed.

[0060] (5) After the first traversal is completed, perform the second, third, or more manual traversals until all wave points no longer change, which is the optimal characteristic band.

[0061] Finally, after seven rounds of screening, the optimal result is obtained, as shown in the table:

[0062]

[0063] The finally selected characteristic band results are: 945.21 cm -1 -1373.45 cm -1 、1755.39 cm -1 -1797.83 cm -1 、2010.02 cm -1 -2318.66 cm -1 、2384.24 cm -1 -2600.29 cm -1 、2731.46 cm -1 -2924.36 cm -1 and 3622.66 cm -1 -4112.63 cm -1 , with a difference of two wave points allowed before and after each section ( Figure 3)。It was found that the synergistic effect of multiple bands and multiple wave points in the model enabled the model to achieve the optimal effect, indicating that there were more bands and wave points related to lactoferrin, with a wide range, and the number of characteristic wave points was 437.

[0064] 4. Screening and determination of model parameters

[0065] The model parameters include the parameters of the preprocessing method and the parameters of the algorithm. In this model, the data preprocessing method is SNV (Standard Normal Variate Transformation), and SNV does not require parameters; the main parameter is the parameter of the partial least squares regression algorithm (PLSR): the number of principal components (n_component). The comparison results of parameter selection are as follows:

[0066]

[0067] According to the comparison results, the number of principal components (n_component) was finally selected as 16.

[0068] After comparative analysis, the best regression model for lactoferrin was: SNV + PLS-DA (n_component = 16) model. The correlation coefficients of the training set and the test set were 0.8740 and 0.9066 respectively; the root mean square errors of the training set and the test set were 3.2186 and 2.9400 respectively.

[0069] Example 4:

[0070] Application of the rapid batch detection method of mid-infrared spectrum MIR of lactoferrin in milk:

[0071] The established best model (SNV + PLS-DA, n_component = 16) was used to predict the lactoferrin content in another 5 milk samples (not one of the 289 experimental materials), and the prediction results were compared with the true values.

[0072] Model usage method:

[0073] 1. The characteristic bands in the infrared spectrum of the milk sample were collected as: 945.21 cm -1 -1373.45 cm -1 、1755.39 cm -1 -1797.83 cm -1 、2010.02 cm -1 -2318.66 cm -1 、2384.24 cm -1 -2600.29 cm -1 、2731.46 cm -1 -2924.36 cm -1 And 3622.66 cm -1-4112.63 cm -1 MIR data in

[0074] Meanwhile, the true value of lactoferrin in the same batch of milk was detected by liquid chromatography.

[0075] 2. Substitute the obtained MIR data into the SNV+PLS-DA (n_component = 16) model constructed in Example 3 to output the predicted results of lactoferrin content;

[0076] As can be seen from the following table, the lactoferrin content predicted by this model is very close to the true content ( Figure 4 ), so the accuracy of this model is relatively high and can be used to predict the lactoferrin content of milk.

[0077]

[0078] References

[0079] [1] Jin Liang. Functions and Applications of Lactoferrin [J]. Food Safety Guide, 2014(15): 46-47.

[0080] [2] Feng Li, Deng Daping, etc. Physiological Functions and Research Progress of Lactoferrin [J]. Chinese Journal of Radiation Hygiene, 2012, 21(01): 121-124.

[0081] [3] Liu Shuan, Li Yikun, etc. Research Progress on Biological Functions of Lactoferrin [J]. Chinese Journal of Animal Nutrition, 2020, 32(04): 1508-1515.

[0082] [4] Liang Junfang. Research on the Seasonal Variation Law of Lactoferrin in Cow Milk [J]. Journal of Agricultural Products Processing (Academic Edition), 2009(02): 75-76.

[0083] [5] Liu Xiuqing, Jiang Jindou, Tao Dali. Determination of Lactoferrin Mass Fraction in Milk by Ultrafiltration Concentration-HPLC Method and Functional Evaluation [J]. China Dairy Industry, 2016, 44(11): 53-56.

[0084] [6] Jia Yunhong, Song Xiaoqing, Yang Kai, etc. Determination of Lactoferrin Content in Infant Formula Milk Powder by High Performance Liquid Chromatography [J]. China Dairy Cattle, 2015(13): 49-51, 52.

[0085] [7] Zhang Junchao, Hu Suli. Common Detection Methods of Lactoferrin in Dairy Products and Related Foods [J]. Food Safety Guide, 2020(19): 40-42.

[0086] [8] Tian Rongrong, Bai Shasha, Xu Jiajia, et al. Analysis of the Contents of α-Lactalbumin, β-Lactoglobulin and Lactoferrin in Milks from Different Sources [J]. Science and Technology of Food Industry, 2020, 41(4): 311-315, 321.

[0087] [9] De Marchi, M., V. Toffanin, M. Cassandro, and M. Penasa. 2014. Invited review: Mid-infrared spectroscopy as phenotyping tool for milk traits. J. Dairy Sci. 97: 1171–1186.

[0088]

[10] Shaw RA, Kotowich S, Leroux M, Mantsch HH. Multianalyte serum analysis using mid-infrared spectroscopy. Ann Clin Biochem. 1998; 35(Pt 5): 624–32.

[0089]

[11] Kruse-Jarres JD, Janatsch G, Gless U, Marbach R, Heise HM. Glucose and other constituents of blood determined by ATR-FTIR-spectroscopy. Clin Chem. 1990; 36: 401–2.

Claims

1. A rapid batch detection method for lactoferrin content in milk by MIR, comprising the following steps: 1). The characteristic bands in the infrared spectrum of the milk sample are: 945.21 cm -1 -1373.45 cm -1 、1755.39 cm -1 -1797.83 cm -1 、2010.02 cm -1 -2318.66 cm -1 、2384.24 cm -1 -2600.29 cm -1 、2731.46 cm -1 -2924.36 cm -1 and the MIR data in 3622.66 cm -1 -4112.63 cm -1 ; 2), Substitute the obtained MIR data into the SNV+PLS-DA(n_component=16) model to output the predicted result of lactoferrin content.

2. The method according to claim 1, characterized in that: In step 1), there is a tolerance of two wave points before and after each segment.

3. Use of the method according to claim 1 for detecting lactoferrin content in milk.

Images