Intermediate infrared rapid batch detection method for leucine content in milk
Through mid-infrared spectroscopy technology and standard normal variable + partial least squares regression model, combined with the selection of characteristic bands, the problems of high cost, time-consuming and low accuracy of leucine detection in milk are solved, and the brightness of milk is achieved Fast and accurate detection of ammonium content, suitable for dairy cow performance measurement and milk quality detection.
Patent Information
- Application Number
- CN202411023909.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-19
AI Technical Summary
The existing leucine content detection methods in milk are costly, time-consuming, and have low accuracy and sensitivity, making it difficult to achieve rapid batch use.
Using mid-infrared spectroscopy technology, the prediction results of leucine content are output by collecting MIR data in a specific band and using standard normal variable + partial least squares regression model. The characteristic band selection uses adjacency constraint hierarchical clustering + manual selection method , reduces the complexity of the model and multicollinearity problems, and improves the accuracy and efficiency of detection.
It realizes fast, accurate and low-cost detection of the leucine content in milk. The sample measurement time takes only 10-15 seconds, which improves the detection efficiency and is suitable for dairy cow performance measurement and milk quality detection.
Smart Images

Figure BDA0004968022870000031 
Figure BDA0004968022870000051 
Figure BDA0004968022870000061
Abstract
Description
Technical Field
[0001] The invention belongs to the field of dairy cow performance measurement and milk quality detection, and particularly relates to a mid-infrared rapid batch detection method for leucine content in milk. Background Art
[0002] Milk is rich in various nutrients and bioactive ingredients. It is the most ideal natural food for infants, the elderly and other people with low physiological status. It is known as "white blood". Animals obtain protein to obtain amino acids, especially essential amino acids that are crucial to animal growth and development, such as leucine, isoleucine, valine, etc. The types and contents of essential amino acids in milk are rich, and the composition is similar to that of human milk. It is a relatively high-quality protein source for children and adults.
[0003] The molecular formula of leucine (Leu) is C6H 13 NO2 is the highest essential amino acid in milk and one of the three branched-chain amino acids. Existing literature shows that leucine plays an important role in the growth and development of animals and the processing of dairy products, such as regulating the body's immunity, regulating mammalian skeletal muscle protein metabolism (promoting muscle protein synthesis to achieve the purpose of repairing muscles), promoting fat oxidation and decomposition, etc. In addition, studies have found that 1 mol of leucine can produce 40 mol of ATP after complete oxidation, and the efficiency of oxidation function is higher than that of other amino acids. Therefore, during special physiological periods such as hunger, lactation, stress and exercise, leucine can also be used as a body energy source. According to the existence form of amino acids, the amino acids in milk can be divided into hydrolyzed amino acids and free amino acids. Leucine exists in both hydrolyzed amino acids and free amino acids in milk. The content used in the present invention is the leucine content in free amino acids in milk.
[0004] There is no national standard for the determination of leucine in free amino acids in dairy products. Common methods include amino acid automatic analyzer analysis, high performance liquid chromatography, ultraviolet spectrophotometry, etc. However, these methods have problems such as high cost and long time consumption, so they are very limited and difficult to use in batches quickly in production practice. Mid-infrared spectroscopy (MIR) technology is a detection technology that uses the absorption, scattering, reflection and transmission characteristics of light to conduct qualitative or quantitative analysis of the object to be tested. The use of mid-infrared spectroscopy technology can realize the simultaneous detection of multiple nutrients in milk. Compared with traditional detection methods, this detection method has the advantages of being fast, efficient, accurate, environmentally friendly, and able to achieve non-destructive detection.
[0005] At present, research on mid-infrared detection of milk leucine content has begun, but there are problems such as low accuracy and sensitivity. The accuracy of infrared spectroscopy detection results will be affected by various factors.
[0006] The purpose of the present invention is to solve the existing problems, establish a rapid detection technology for leucine (the sum of free leucine and hydrolyzed leucine) in milk, and provide technical guarantee for the development of China's dairy industry. Summary of the invention
[0007] The purpose of the present invention is to provide a mid-infrared rapid batch detection method for leucine content in milk. The method is simple, rapid and has a high accuracy rate compared with the true value.
[0008] Another object of the present invention is to provide an application of a mid-infrared rapid batch detection method for leucine content in milk. In order to achieve the above object, the present invention adopts the following technical measures:
[0009] The mid-infrared rapid batch detection method for leucine content in milk comprises the following steps:
[0010] 1. The characteristic band of the infrared spectrum of the collected milk sample is: 929.778cm -1 -964.5cm -1 、972.216cm -1 -976.074cm -1 、983.79cm -1 -1014.654cm -1 、1057.092cm -1 -1060.95cm -1 、1095.672cm -1 -1130.394cm -1 、1161.258cm -1 -1195.98cm -1 、1242.276cm -1 、1304.004cm -1 -1342.584cm -1 、1354.158cm -1 -1385.022cm -1 、1439.034cm -1 -1473.756cm -1 、1493.046cm -1 、1504.62cm -1 、1520.052cm -1 cm -1 、1547.058cm -1 、1570.206cm -1 、1720.668cm -1 、1732.242cm -1-1743.816cm -1 、1921.284cm -1 and 1932.858cm -1 MIR data in;
[0011] 2. Substitute the MIR data obtained by the measurement into the standard normal variable + partial least squares regression model to output the prediction results of leucine content;
[0012] The principal component of the partial least squares regression is 10.
[0013] The protection scope of the present invention also includes:
[0014] Application of mid-infrared rapid batch detection method for leucine content in milk in milk quality detection
[0015] In the above application, preferably, the milk quality detection is leucine content detection.
[0016] The leucine content involved in the present invention refers to the total content of free leucine and hydrolyzed leucine in milk.
[0017] Compared with the prior art, the present invention has the following advantages:
[0018] 1. In terms of the selection of characteristic bands, we broke the common algorithm for feature selection and used the adjacency constrained hierarchical clustering (HAC) + manual selection method. The characteristic waves finally selected not only improved the prediction performance of the model, but also overcame the multicollinearity problem of the spectrum and reduced the complexity of the model (the input variables were reduced from 1060 points to 86 points), ultimately saving the time cost, storage space cost and electricity cost of the model operation.
[0019] 2. The optimal preprocessing and algorithm combination for establishing the leucine model was selected, the optimal parameters were determined, and the accuracy of the model was improved.
[0020] 3. It realizes the rapid, accurate and low-cost detection of leucine content in raw milk and realizes rapid batch detection. The detection time of each sample is only 10-15 seconds, which improves the detection efficiency and has strong practicality. It can be widely used in dairy cow performance determination and milk quality detection. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 (a) The infrared spectrum of the untreated milk sample and the average (± standard deviation) spectrum (b) are shown.
[0022] Figure 2 This is the mid-infrared spectrum of the milk sample after SNV pretreatment.
[0023] Figure 3 This is the correlation diagram between the actual value and predicted value of the model milk data and the fitted straight line diagram. DETAILED DESCRIPTION
[0024] The technical solutions described in the present invention, unless otherwise specified, are all conventional solutions in the art; the reagents or materials described, unless otherwise specified, are all from commercial channels.
[0025] 1. Experimental Materials
[0026] The experimental materials were 317 Chinese Holstein cows from 7 dairy farms in 6 regions of my country. A milk sample was collected from each cow. The milk sample collection was completed using an automatic milking device. The milk room was first wiped with a sterilized towel, and then the udder was disinfected with an iodine-glycerol mixed solution. After squeezing out the first three handfuls of milk, milk samples were collected during the entire milking process. Each milk sample was 40 ml and divided into new cylindrical sampling bottles with a diameter of 3.5 cm and a height of 9 cm. They were numbered in sequence, and bronopol preservative was immediately added to each sampling bottle. It was shaken slowly to fully dissolve. Ice bags (2-4°C) were placed around the milk samples on the way back to prevent deterioration. Spectral data were collected immediately after the samples arrived at the laboratory. Spectral data were collected twice for all milk samples.
[0027] Sample information statistics table
[0028]
[0029] 2. Mid-infrared spectrum measurement and collection
[0030] The sample was poured into a cylindrical sample tube with a diameter of 3.5 cm and a height of 9 cm, and then placed in a water bath at 42°C for 15-20 min. The sample was scanned using the MilkoScan TM The 7RM milk composition detector extends the solid fiber optic probe into the liquid, mixes the sample and then scans it.
[0031] 3. Detection method of the true (reference) value of leucine content in milk
[0032] 3.1 Instruments, equipment and reagents
[0033] Fully automatic amino acid analyzer (German Sykam S433D); syringe filter; 0.22μm nylon filter membrane; electric heating blast drying oven; nitrogen blower; electronic analytical balance; vortex oscillator; centrifuge; injection bottle; hydrolysis tube; parallel evaporator; volumetric flask.
[0034] Hydrochloric acid (HCl): concentration ≥36%, high purity; phenol (C6H5OH); nitrogen: purity 99.9%; sodium citrate (Na3C6H5O7·2H2O); sodium hydroxide (NaOH): high purity; mixed amino acid standard solution; amino acid mixed standard solution; ninhydrin solution; buffer A, buffer B, buffer C, regeneration solution D; LiOH×H2O
[0035] 3.2 Experimental methods
[0036] 3.2.1 Collection of mid-infrared spectra
[0037] MilkoScanTM FT+ was used for spectrum collection. The specific collection steps were as follows: milk samples were placed in batches in a 45°C electric constant temperature water bath for preheating for 5 minutes. The preheated milk samples were placed on the detection rack and shaken up and down several times to mix the milk colloidal solution evenly. The detection rack was placed on the detection track, the bottle cap was opened, and the tests were performed in sequence. After the spectrum was collected, the milk samples were stored at 4°C for the subsequent determination of the amino acid content in the milk.
[0038] 3.2.2 Determination of total amino acid content in milk (sample reference value)
[0039] (1) Preparation of standard solution
[0040] Mixed amino acid standard stock solution (1 μmol / mL): Accurately weigh individual amino acid standards (accurate to 0.00001 g) in the same 50 mL beaker, dissolve with 8.3 mL 6 mol / L hydrochloric acid solution, accurately transfer to a 250 mL volumetric flask, dilute to volume with water, and mix well.
[0041] Mixed amino acid standard working solution (100nmol / mL): Accurately pipette 1.0mL of mixed amino acid standard stock solution into a 10mL volumetric flask, add pH 2.2 sodium citrate buffer solution to the scale, mix well, and use it as the standard liquid for the machine.
[0042] (2) Milk sample processing
[0043] Pipette about 1 ml of milk sample into a hydrolysis tube, add 10 mL to 15 mL of 6 mol / L hydrochloric acid solution and 3 to 4 drops of phenol into the hydrolysis tube. Place the hydrolysis tube in a cryogen, freeze for 3 to 5 minutes, connect it to the vacuum pump's exhaust pipe, evacuate (close to 0 Pa), and then fill it with nitrogen. Repeat the evacuation and nitrogen filling process 3 times, and seal or tighten the screw cap while filling it with nitrogen.
[0044] Place the sealed hydrolysis tube in an electric blast thermostat or hydrolysis furnace at 110℃±1℃. After hydrolysis for 22 hours, take it out and cool it to room temperature.
[0045] Open the hydrolysis tube, filter the hydrolyzate into a 50mL volumetric flask, rinse the hydrolysis tube with a small amount of water several times, transfer the washing liquid into the same 50mL volumetric flask, and finally make up to the scale with water and oscillate to mix.
[0046] Accurately pipette 1.0mL of filtrate into a 15mL or 25mL test tube, and dry under reduced pressure using a test tube concentrator or parallel evaporator at 40℃~50℃. Dissolve the dried residue with 1mL~2mL of water, dry under reduced pressure, and finally evaporate to dryness.
[0047] Add 1.0mL to 2.0mL of pH 2.2 sodium citrate buffer solution into the dried test tube to dissolve it. After oscillation and mixing, pipette the solution through a 0.22μm filter membrane and transfer it to the instrument injection bottle as the sample measurement solution for instrument measurement.
[0048] (3) Chromatographic conditions
[0049] Chromatographic column: sulfonic acid cationic resin; detection wavelength: 570nm and 440nm.
[0050] 4. Elimination of outliers and selection of valid samples
[0051] The outliers are removed according to the following criteria:
[0052] (1) Remove reference values and samples with null spectra,
[0053] (2) Remove samples with spectral Mahalanobis distance > 3
[0054] (3) Remove samples whose reference values are outside the range of mean ± 3 times standard deviation.
[0055] Finally, 4 abnormal samples were eliminated from 317 samples, and 313 samples were selected for model building and optimization. Figure 1 (a) The infrared spectrum of the untreated milk sample and the average (± standard deviation) spectrum (b) are shown.
[0056] Embodiment 1:
[0057] Selection of prediction model algorithm for leucine in milk:
[0058] The purpose of this application is to establish a quantitative prediction model for the leucine content in milk, using three common basic modeling algorithms.
[0059] Partial least squares regression (PLSR) is a multivariate data analysis method in statistics and machine learning, and is particularly suitable for dealing with situations where there is multicollinearity between dependent and independent variables. PLSR is the most basic and common machine learning algorithm. Its core idea is to find new orthogonal projection directions (principal components) so that the projected dependent and independent variables have the largest covariance, and then establish a prediction model. Its advantages mainly include the following three aspects: (1) It effectively overcomes the multicollinearity problem between the dependent and independent variables; (2) When the dimensions of both the independent and dependent variables are very high, PLS regression can extract the most important information and construct a prediction model through dimensionality reduction technology; (3) When the number of samples is small, a relatively ideal prediction effect can still be obtained. Therefore, this algorithm is selected as the candidate algorithm in this embodiment.
[0060] K-nearest neighbor algorithm (KNN) is an algorithm in supervised learning, which can be used for both classification and regression. The advantages of KNN algorithm are that it is simple and easy to use, can be used for nonlinear traits, has fast model training time and good prediction effect. It is one of the most basic and simplest algorithms in machine learning algorithms. Therefore, this embodiment selects this algorithm as the candidate algorithm.
[0061] Ridge Regression Ridge regression is a least squares method with L2 regularization. It abandons the unbiasedness of the least squares method and obtains the regression coefficient at the cost of losing some information and reducing precision. It is a more practical and reliable regression method. It fits data with outliers better than the least squares method. Therefore, this embodiment selects this algorithm as the candidate algorithm.
[0062] Embodiment 2:
[0063] Establishment of a method for detecting leucine content in milk using mid-infrared spectroscopy:
[0064] 1. Division of modeling data set
[0065]
[0066] In this embodiment, the total data set is divided into two parts, namely, a training set and a test set, with a data ratio of about 4:1. The training set is used to train the model, and the test set is used to test the model performance. The training set is also called a cross-validation set, and a 10-fold cross-validation is performed during the model training process.
[0067] 2. Screening of MIR data preprocessing methods for modeling
[0068] This embodiment uses trial and error to select appropriate spectral preprocessing techniques, such as first derivative (1D), second derivative (1D), standard normal variable (SNV) and Savitzky-Golsy convolution smoothing (SG). The strategy of trial and error is to first apply each preprocessing to each model, and then determine the most successful model based on performance parameters. SNV is mainly used to eliminate the effects of particle size and surface scattered light on the spectrum. Derivative processing is usually used to improve the resolution of spectral samples and sharpen absorption bands.
[0069] The most appropriate spectral preprocessing technique was selected by trial and error. The modeling bands used were those with the water region removed (except 1593.35 cm -1 -1709.1cm -1 and 3059.39cm -1 -3641.95cm -1 ) and no information area (3,500-5,000cm -1 ) The remaining spectral region contains a total of 520 wave points. The modeling algorithm used is the most commonly used PLSR algorithm. The best preprocessing method is determined by the results: SNV, Figure 2 This is the mid-infrared spectrum of the milk sample after SNV pretreatment.
[0070] The results are shown in the table:
[0071]
[0072] 3. Selection of modeling algorithms
[0073] In the screening process of the modeling algorithm, the modeling band used was the one that removed the water area (except 1593.35cm -1 -1709.1cm -1 and 3059.39cm -1 -3641.95cm -1 ) and no information area (3,500-5,000cm -1 ) The remaining spectral region contains a total of 520 wave points, and the spectral preprocessing technology used is SNV. The best preprocessing method is determined by the results: PLSR, and the results are shown in the table:
[0074]
[0075] 4. Screening of modeling feature bands (adjacent constraint hierarchical clustering combined with manual selection)
[0076] The feature band method includes two methods: algorithmic feature selection and manual feature selection. The basis for algorithmic feature selection is mainly the correlation between each wave point and the reference value. Although it is fast and efficient, it ignores the synergy between adjacent wave points. The advantage of manually selecting features is that the role of the band (i.e., adjacent wave points) can be strengthened in the selection process, and the original information state of the spectrum can be retained more in the process of improving the model, with stronger inclusiveness and generalization ability. However, the speed of manual feature selection is relatively slow and the efficiency is also low. The milk spectrum is a high-dimensional data with 1060 wave points. One of the notable features is that adjacent wave points have multicollinearity, that is, they may contain the same or similar information. If these wave points are used directly for modeling, a large amount of similar information will exist in the model, thereby affecting the modeling effect. Taking the above points into consideration, the present application uses manual feature selection based on the wave points after adjacency constrained hierarchical clustering (HAC). This method not only handles the multicollinearity characteristics of the spectrum, but also more fully retains the original information state of the spectrum to improve the accuracy and effect of modeling.
[0077] This embodiment selects feature bands by combining adjacency constrained hierarchical clustering (HAC) with manual selection, and the selection steps are as follows:
[0078] (1) Remove 1593.35cm -1 -1709.1cm -1 and 3059.39cm -1 -3641.95cm -1 Two spectral regions associated with water absorption.
[0079] (2) Remove 3,500-5,000cm -1 region of the spectrum because it is believed to contain very little valuable information.
[0080] (3) Spatial clustering: The remaining 520 spectral points are subjected to HAC analysis, and the optimal number of windows is determined based on the slope heuristic method. In this embodiment, HAC is implemented using the adjclust package in R, and the slope heuristic method for determining the number of windows is implemented using the Capushe function in the adjclust package. Finally, the HAC algorithm clusters the 520 points into 64 groups, with the number of points in each group ranging from 1 to 46. The results are shown in the table:
[0081]
[0082]
[0083] (4) Using the partial least squares regression algorithm, first increase or decrease the wave points at both ends of the critical point of the first group of bands by one wave point (increase one wave point for the lowest wave number and decrease one wave point for the highest wave number) to find the optimal effect, and then perform similar operations on the second group of bands based on this. If the model performance of the second group of bands is not improved or even reduced after similar operations, then remove this band from the model. Finally, after all 64 bands have completed a round of operations until the model accuracy does not change or improve, it is the optimal feature band.
[0084] The results of the model established by the optimal characteristic band obtained after 64 rounds of screening are shown in the table:
[0085]
[0086] The final selected characteristic band result is: 929.778cm -1 -964.5cm -1 、972.216cm -1 -976.074cm -1 、983.79cm -1 -1014.654cm -1 、1057.092cm -1 -1060.95cm -1 、1095.672cm -1 -1130.394cm -1 、1161.258cm -1 -1195.98cm -1 、1242.276cm -1 、1304.004cm -1 -1342.584cm -1 、1354.158cm -1 -1385.022cm -1 、1439.034cm -1 -1473.756cm -1 、1493.046cm -1 、1504.62cm -1 、1520.052cm -1 cm -1 、1547.058cm -1 、1570.206cm -1 、1720.668cm -1 、1732.242cm -1 -1743.816cm -1 、1921.284cm -1 and 1932.858cm-1 .
[0087] 5. Screening and determination of model parameters
[0088] Model parameters include parameters of the preprocessing method and parameters of the algorithm;
[0089] Since the model does not use the SNV preprocessing algorithm, it only needs to select the algorithm parameters. The parameters of the partial least squares regression algorithm are: principal component (n_component). The comparison of parameter selection results is as follows:
[0090]
[0091] According to the comparison results, the main component (n_component) is finally selected as 10.
[0092] After comparative analysis, the best regression model for leucine content in milk is: SNV+PLSR (n_component=10) model. The training set and test set determinations are 0.86 and 0.86 respectively; the training set and test set root mean square errors are 0.02 and 0.02 respectively; the training set and test set performance deviation ratios are 2.71 and 2.50 respectively. Figure 3 This is the correlation diagram between the actual value and predicted value of the model milk data and the fitted straight line diagram.
[0093] Embodiment 3:
[0094] Application of rapid batch detection method of mid-infrared spectroscopy MIR for leucine content in milk:
[0095] The established optimal regression model of total essential amino acids SNV+PLSR (n_component=10) was used to predict 4 randomly selected milk samples (not one of the 317 experimental materials), and the predicted results were compared with the true values.
[0096] How to use the model:
[0097] 1. The characteristic band of the infrared spectrum of the collected milk sample is: 929.778cm -1 -964.5cm -1 、972.216cm -1 -976.074cm -1 、983.79cm -1 -1014.654cm -1 、1057.092cm -1 -1060.95cm -1 、1095.672cm -1 -1130.394cm -1 、1161.258cm-1 -1195.98cm -1 、1242.276cm -1 、1304.004cm -1 -1342.584cm -1 、1354.158cm -1 -1385.022cm -1 、1439.034cm -1 -1473.756cm -1 、1493.046cm -1 、1504.62cm -1 、1520.052cm -1 cm -1 、1547.058cm -1 、1570.206cm -1 、1720.668cm -1 、1732.242cm -1 -1743.816cm -1 、1921.284cm -1 and 1932.858cm -1 MIR data in;
[0098] At the same time, a fully automatic amino acid analyzer was used to detect the true value of the leucine content in the same batch of milk.
[0099] 2. Substitute the MIR data obtained by the measurement into the SNV+PLSR (n_component=10) model constructed in Example 2 to output the prediction result of leucine content.
[0100] The results predicted by the model are close to the actual results (as shown in the following table), so the model has high accuracy and can be used to predict the leucine content in milk.
[0101]
Claims
1. A mid-infrared rapid batch detection method for leucine content in milk, comprising the following steps:
1. The characteristic band in the infrared spectrum of the collected milk sample is: 929.778 cm -1 -964.5 cm -1 、972.216 cm -1 -976.074cm -1 、983.79 cm -1 -1014.654cm -1 、1057.092 cm -1 -1060.95cm -1 、1095.672 cm -1 -1130.394cm -1 、1161.258 cm -1 -1195.98cm -1 、1242.276cm -1 、1304.004 cm -1 -1342.584cm -1 、1354.158 cm -1 -1385.022cm -1 、1439.034 cm -1 -1473.756cm -1 、1493.046cm -1 、1504.62cm -1 、1520.052cm -1 cm -1 、1547.058cm -1 、1570.206cm -1 、1720.668cm -1 、1732.242 cm -1 -1743.816cm -1 、1921.284cm -1 and 1932.858cm -1 MIR data in; 2. Substitute the MIR data obtained by the measurement into the standard normal variable + partial least squares regression model to output the prediction results of leucine content; The principal component of the partial least squares regression is 10.
2. Application of the detection method according to claim 1 in milk quality detection.
3. The use according to claim 2, wherein the milk quality detection is a leucine content detection.
