A method and system for identifying buffalo milk based on oligosaccharide profile and logistic regression model

Abstract

The present application belongs to the technical field of dairy detection and traceability, and particularly relates to a buffalo dairy species identification method and system based on oligosaccharide spectrum and a logistic regression model. The method comprises: detecting the content of seven characteristic oligosaccharides (3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT and LNT II) in a to-be-detected milk sample; inputting the content data into a pre-trained logistic regression classification model to obtain a prediction probability; and outputting an identification result according to the prediction probability. The present application firstly determines that the combination of the seven oligosaccharides can be used as a stable and efficient identification marker, and a classification model is constructed based on this. The identification accuracy of the method can reach 100%, the stability is good, the method is suitable for rapid and objective species identification of raw milk and processed milk, and has important application value in the field of dairy product quality control and product traceability.

Description

Technical Field

[0001] This invention belongs to the field of dairy product testing and traceability technology, specifically relating to a method and system for identifying buffalo milk breeds based on oligosaccharide profiles and logistic regression models. Background Technology

[0002] Identifying the true breed of milk is a crucial technical step in ensuring dairy product quality, maintaining market order, and protecting consumer rights. Buffalo milk, due to its unique nutritional composition and rich functional components, typically commands a higher market value than ordinary Holstein milk, creating an economic incentive to sell Holstein milk as buffalo milk. Currently, traditional identification methods, such as sensory evaluation and protein or fatty acid profiling, suffer from limitations including high subjectivity, limited accuracy, and susceptibility to processing techniques. While DNA testing is accurate, it is costly, complex, and prone to DNA degradation in highly processed dairy products, making it unsuitable for rapid routine screening and large-scale sample testing.

[0003] Oligosaccharides are an important class of natural bioactive components in milk, and their composition and content are species- and variety-specific, making them potential biomarkers for variety identification. However, current research on bovine milk oligosaccharides largely remains at the level of species identification or analysis of single functional components. Although some studies have mentioned that buffalo milk may have higher levels of certain sialyl lactose or fucoidosyl lactose, systematic and quantitative comparative studies are lacking, and a set of characteristic indicators that can stably and efficiently distinguish buffalo milk from Holstein milk has not been screened from the complex oligosaccharide spectrum. Furthermore, how to transform these chemical indicators into operable identification models to achieve automated and high-precision discrimination remains a technical challenge that current technologies have not yet solved.

[0004] Therefore, developing a method for identifying dairy varieties based on stable chemical markers that is rapid, provides objective results, and has high accuracy is of urgent market demand and significant technological value for strengthening the supervision of the dairy industry, combating product adulteration, and enhancing the added value of specialty dairy products. Summary of the Invention

[0005] This invention aims to overcome the shortcomings of existing technologies by providing a method and system for identifying buffalo milk breeds based on oligosaccharide spectra and logistic regression models. It seeks to address the problems of low accuracy, high cost, and complex procedures in existing buffalo milk breed identification methods, and to provide a rapid, accurate, and stable identification solution.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for distinguishing between buffalo milk and Holstein milk includes the following steps:

[0008] S1. Detect the content of seven characteristic oligosaccharides in the milk sample to be tested. The seven characteristic oligosaccharides are: 3-fucosylated lactose (3FL), 3'-sialylated lactose (3SL), 3'-sialylated N-acetyllactosamine (3SLN), 6'-sialylated lactose (6SL), Lewis-X trisaccharide (Le-XT), lactose-N-neotetrasaccharide (LNnT), and lactose-N-trisaccharide (LNT II).

[0009] S2. Standardize the content of the seven oligosaccharides obtained from the test;

[0010] S3. Input the standardized content data of the seven characteristic oligosaccharides into the trained logistic regression classification model to obtain the predicted probability P that the milk sample to be tested is buffalo milk.

[0011] S4. Output the identification result based on the predicted probability P: if P ≥ 0.5, it is determined to be Holstein milk; if P < 0.5, it is determined to be buffalo milk.

[0012] To further explain, in step S1, the content of the seven characteristic oligosaccharides is detected by high performance liquid chromatography-tandem mass spectrometry.

[0013] To further explain, the high performance liquid chromatography-tandem mass spectrometry method uses a hydrophilic interaction column, with an aqueous solution containing ammonium formate and acetonitrile as the mobile phase for gradient elution, and the mass spectrometry uses electrospray positive ion mode and multiple reaction monitoring mode.

[0014] To further explain, in step S2, the standardization process involves standardizing the content xi of each oligosaccharide using the following formula: ,in, This represents the original content value of the j-th feature oligosaccharide in the i-th sample; The average content of the j-th characteristic oligosaccharide; Let $\frac{j}{j}$ be the standard deviation of the $j$-th characteristic oligosaccharide. These are the standardized feature values.

[0015] To further explain, the logistic regression classification model calculates the probability value using the following formula. :

[0016]

[0017] in, It is the probability that the i-th sample is a Holstein cow (label=1), and e is the natural constant (≈2.718).

[0018] z is the value of the linear combination, and its calculation formula is as follows:

[0019]

[0020] in, This is the linear output of the i-th sample; The model intercept; is the weighting coefficient for the j-th characteristic oligosaccharide; Let be the standardized value of the oligosaccharide content of the j-th characteristic; This represents the number of characteristic oligosaccharide types.

[0021] To further explain, the training process of the logistic regression model includes:

[0022] Step A1: Collect milk samples from known breeds of buffalo and Holstein cows to form a training set;

[0023] Step A2: Detect the content data of the characteristic oligosaccharides for each sample in the training set;

[0024] Step A3: Perform Z-score standardization on the feature oligosaccharide content data of the training set to obtain the standardized feature values;

[0025] Step A4: Based on the standardized feature values ​​and their corresponding variety labels, the logistic regression model is trained using gradient descent to learn the model parameters, which include the intercept b and the weight coefficient of each feature oligosaccharide. .

[0026] Further explanation: The model parameters include: intercept b = -1.545637; weight coefficient for 3-fucoyllactose (3FL) = -0.479375; weight coefficient for 3-sialyllactose (3SL) = 1.524488; weight coefficient for 3-sialyl-N-acetyllactosamine (3SLN) = 0.768616; weight coefficient for 6-sialyllactose (6SL) = -1.387730; weight coefficient for Lewis X trisaccharide (Le-XT) = -0.628087; weight coefficient for lactose-N-trisaccharide (LNT II) = 0.313290; weight coefficient for lactose-N-neotetrasaccharide (LNnT) = -2.133230; the linear combination value z is obtained from the above model parameters:

[0027] .

[0028] This embodiment also provides an identification system for implementing any of the methods described above, comprising:

[0029] The data acquisition module is used to acquire the content data of seven characteristic oligosaccharides in the milk sample to be tested;

[0030] The data processing module is used to standardize the detection data;

[0031] The model calculation module stores the trained logistic regression classification model and is used to receive the processed data and calculate the prediction probability P.

[0032] The result output module is used to output the identification result based on the predicted probability.

[0033] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0034] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the methods described above.

[0035] The present invention has the following beneficial effects:

[0036] This invention, based on systematic glycomics research, discovered a novel set of characteristic oligosaccharide combinations (3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, LNT II) capable of stably distinguishing between buffalo milk and Holstein milk. Large-scale sample validation and in-depth statistical analysis demonstrated that the content of these seven oligosaccharides differs significantly between the two types of milk, especially 3SL and 3SLN, with statistically significant differences (Cohen's d absolute value > 1.6), serving as key discriminant markers. The logistic regression model constructed based on this, with its linear combination parameters and regression coefficients learned from a large number of real samples using a rigorous gradient descent optimization algorithm, possesses solid scientific data support. Validated on the test set, the model achieves a classification accuracy of 100%, an average accuracy of 99.0% with 5-fold cross-validation, and an area under the ROC curve (AUC) of 1.0, fully demonstrating the scientific rigor, high accuracy, and strong robustness of this invention.

[0037] This invention combines the detection of characteristic oligosaccharides with machine learning models to construct a complete and easily industrialized identification solution. On the detection side, this invention provides a standardized LC-MS / MS quantitative detection method. This method features simple pretreatment, high sensitivity, and strong specificity, enabling stable and accurate quantification of seven characteristic oligosaccharides. On the model application side, this invention solidifies all complex model parameters (intercept, regression coefficients, standardized mean, and standard deviation). Users only need to input the original content of the seven oligosaccharides in the sample to obtain a rapid and objective identification conclusion through standardization and model calculation. The entire process is simple to operate, provides objective results, and is cost-effective. Furthermore, this invention also protects the dedicated system, computer-readable storage medium, and electronic equipment used to implement this method, covering various scenarios from laboratory testing to field applications, possessing extremely high industrial application value and market transformation potential. Attached Figure Description

[0038] Figure 1 The OPLS-DA score diagram and orthogonal substitution test diagram of oligosaccharides under positive and negative ion modes;

[0039] Figure 2 Cluster analysis diagram of oligosaccharides under two ionic modes;

[0040] Figure 3 Volcano plot showing the difference in oligosaccharide content between buffalo milk and Holstein milk;

[0041] Figure 4 The changes in oligosaccharide content and percentage of different oligosaccharides in milk from three buffalo breeds throughout the entire lactation period were investigated.

[0042] Figure 5 PCA results were used to screen the abundance of oligosaccharides in two types of milk;

[0043] Figure 6 For liquid phase spectrum;

[0044] Figure 7 A validation plot of the standardized coefficients and confusion matrix for the logistic regression model;

[0045] Figure 8 This is a logistic regression ROC curve.

[0046] Figure 9 This is a graph showing the results of the five-fold cross-validation.

[0047] Figure 10 Stacked diagram of oligosaccharide composition for different bovine colostrum contents;

[0048] Figure 11 PCA analysis results of oligosaccharides in four types of bovine colostrum;

[0049] Figure 12Radar chart showing the composition of oligosaccharide content in normal bovine milk (data standardized).

[0050] Figure 13 PCA analysis results of oligosaccharides in two types of bovine milk;

[0051] Figure 14 This is a flowchart illustrating the method for differentiating buffalo milk from Holstein milk according to the present invention.

[0052] Figure 15 This is a schematic diagram of the structural block of the identification system of the present invention. Detailed Implementation

[0053] The present invention will be further described in detail below with reference to specific embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the contents of this invention, those skilled in the art can make various modifications and substitutions to the described specific embodiments based on the principles and ideas of the invention, and these equivalent modifications and substitutions also fall within the scope defined by the claims of this invention.

[0054] Example 1: Determination and Verification of Characteristic Oligosaccharide Combinations

[0055] This embodiment aims to illustrate how to identify the seven characteristic oligosaccharide combinations ultimately used to construct the identification model from a large number of potential differentials.

[0056] 1. Sample preparation and grouping

[0057] Twenty-four milk samples from Nilirafi buffaloes in mid-lactation (100-200 days of lactation) and twenty-four milk samples from Holstein cows in mid-lactation were collected from Guangxi province. To reduce individual variability and improve sample representativeness, the buffalo milk samples were evenly distributed according to lactation days and then mixed in equal volumes to form six mixed buffalo milk samples (SN group); four Holstein milk samples were randomly combined to form six mixed Holstein milk samples (HN group). Basal milk composition was measured immediately after collection, and all samples were stored at -80℃ for later use.

[0058] 2. Non-targeted oligosaccharide profiling (Discovery Phase)

[0059] 2.1 Oligosaccharide extraction:

[0060] Take 500 μL of each mixed emulsion sample, dilute with 500 μL of pure water, and centrifuge at 4℃ and 14000 rcf for 20 minutes. Carefully aspirate the intermediate aqueous phase, add 2 volumes of pre-cooled ethanol, mix well, and incubate at -80℃ for 1 hour. Then, freeze-dry under vacuum. Before analysis, reconstitute with 100 μL of acetonitrile-water solution (1:1, v / v), centrifuge again at 4℃ and 14000 rcf for 20 minutes, and take the supernatant for mass spectrometry analysis.

[0061] 2.2 Instrumental analysis conditions:

[0062] Chromatographic conditions: Samples were separated using a Vanquish LC ultra-high performance liquid chromatography (UHPLC) system with a HILIC column; column temperature 45℃; flow rate 0.4 mL / min; injection volume 2 μL; mobile phase composition A: 10 mM ammonium formate + 0.1% FA aqueous solution, B: acetonitrile; gradient elution program as follows: 0-25 min, solution B linearly decreased from 75% to 55%; 25-29 min, solution B linearly decreased from 55% to 15%; 29-40 min, solution B maintained at 15%; 40-40.1 min, solution B linearly decreased from 15% to 75%; 40.1-55 min, solution B maintained at 75%; throughout the analysis, samples were placed in an autosampler at 4℃. To avoid the influence of instrument signal fluctuations, samples were analyzed sequentially in a random order. QC samples were inserted into the sample queue to monitor and evaluate the stability of the system and the reliability of the experimental data.

[0063] Orbitrap Exploris mass spectrometry conditions: After separation by a Vanquish LC ultra-high performance liquid chromatography (UHPLC) system, the samples were analyzed by mass spectrometry using an Orbitrap Exploris 480 mass spectrometer (Thermo), with detection performed in both positive and negative electrospray ionization (ESI) modes. ESI source parameters were: spray voltage 3600 V (ESI+) / 3200 V (ESI-), sheath gas flow rate 40 L / min, ion transfer tube temperature 320℃, and nebulization temperature 450℃. Detection was performed in full-scan / data-dependent two-stage scan (Full-MS / dd-MS2) mode, with the top 10 MS1 ions used to obtain MS / MS spectra. Collision energies (CEs) were set to normalized levels 20, 40, and 60. The primary mass-to-charge ratio scan range was 300–2000.

[0064] 2.3 Data Processing and Differential Screening:

[0065] Raw mass spectrometry data were extracted, aligned, and had retention time corrected using XCMS software. The detected chromatographic peaks were compared with a self-built oligosaccharide database to complete structural identification. First, data quality was assessed; the high overlap in peak response intensity and retention time of the QC samples indicated good instrument stability. Subsequently, orthogonal partial least squares discriminant analysis was used to analyze the SN and HN groups, and oligosaccharides with significant content differences were screened based on the criteria of variable projection importance (VIP) value > 1.0 and t-test p-value < 0.05.

[0066] 3. Preliminary screening results and determination of candidate biomarkers

[0067] Non-targeted analysis identified 47 oligosaccharides, with 46 detected in buffalo milk and 42 in Holstein milk. The OPLS-DA model clearly distinguished between the two groups of samples. Figure 1 The permutation test showed that the model was not overfitting. Based on the criteria of VIP>1 and p<0.05, a total of 32 differentially oligosaccharides were screened, of which 26 were found in higher concentrations in buffalo milk and 6 in higher concentrations in Holstein milk. Clustering heatmap of differentially oligosaccharides (…) Figure 2 The data shows that buffalo milk and Holstein milk samples can be clearly divided into two clusters, and the differences in oligosaccharide profiles among individual buffalo milk samples are greater than those among Holstein milk samples. (Volcano plot) Figure 3 The results visually demonstrate the fold change (FC) and statistical significance of various oligosaccharides, with A4021 (FC=2047.30) and DFLNHa (FC=1615.80), which are extremely abundant in buffalo milk, being particularly prominent.

[0068] Based on significant differences, the known prebiotic and immunomodulatory functional activities of oligosaccharides (such as the anti-inflammatory effect of fucoidan and the brain-development-promoting effect of sialyl lactose), and the commercial availability of standards, nine oligosaccharides with significant differences were selected as candidate characteristic sugars for subsequent quantitative method construction and validation. These include: 2'-fucosylvose (2FL), 3'-fucosylvose (3FL), 3'-sialyl lactose (3SL), 6'-sialyl lactose (6SL), 3'-sialyl-N-acetyllactosamine (3SLN), lactose-N-trisaccharide (LNT II), lactose-N-neotetrasaccharide (LNnT), Lewis-X trisaccharide (Le-XT), and HT trisaccharide (HT).

[0069] 4. Targeted quantitative validation and core combination determination of candidate biomarkers

[0070] To verify the actual identification ability of the above nine candidate oligosaccharides, we used them as target analytes and established a highly sensitive and specific liquid chromatography-tandem mass spectrometry (LC-MS / MS) quantitative detection method (see Example 2 for detailed methodological validation), and then tested an expanded range of samples.

[0071] A total of 48 samples were collected, including colostrum, peak-stage milk, mid-stage milk, and late-stage milk from three buffalo breeds (Nilalafell, Mediterranean, and Mora), as well as colostrum and regular milk from Holstein cattle. The established LC-MS / MS method was used for detection. Results showed that FL and HT were present in the vast majority of buffalo milk and all Holstein milk samples at levels below the method's limit of quantitation or were not consistently detectable, thus they could not be used as reliable quantitative identification indicators. However, seven oligosaccharides—3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, and LNT II—were stably and accurately quantified in all samples, and the data quality met the analytical requirements (see details). Figure 4 ).

[0072] Principal component analysis (PCA) was performed on the contents of the above nine oligosaccharides in buffalo milk and Holstein milk. The results are as follows: Figure 5 As shown, the sum of the principal component contributions of the two samples reached 91.86%. The two milk samples were completely separated in the principal component space with no overlapping areas. It can be clearly seen that there is a significant difference in the composition of the two samples, indicating that the combination of these 7 oligosaccharides can effectively distinguish between buffalo milk and Holstein milk, and can be used as a characteristic marker for subsequent quantitative detection and identification model construction.

[0073] In summary, 32 differentially expressed oligosaccharides were initially screened using non-targeted glycomics, from which 9 candidate sugars with clearly defined functions were selected. After validation using targeted quantitative methods and large-sample testing, 3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, and LNT II were ultimately identified as a core characteristic combination that can be stably detected and exhibits significant and systematic differences in content between buffalo milk and Holstein milk. This combination forms the material basis for subsequently constructing a high-precision mathematical identification model.

[0074] Example 2: LC-MS / MS Quantitative Detection Method for Seven Characteristic Oligosaccharides

[0075] This embodiment details the LC-MS / MS method for detecting the seven core oligosaccharides, which is crucial for obtaining model input data.

[0076] 1. Preparation of standard solutions

[0077] Accurately weigh appropriate amounts of the following standards: 2'-fucosyllactose (2FL), 3'-fucosyllactose (3FL), 3'-sialyllactose (3SL), 6'-sialyllactose (6SL), 3'-sialyl-N-acetyllactosamine (3SLN), lactose-N-trisaccharide (LNT II), lactose-N-neotetrasaccharide (LNnT), Lewis-X trisaccharide (Le-XT), and H-trisaccharide (HT). Dissolve and dilute to volume with ultrapure water to prepare single standard stock solutions. Based on the purity and response of each standard, accurately transfer appropriate volumes of the single stock solutions, mix them, and dilute with ultrapure water to prepare mixed standard intermediate solutions containing all nine target analytes. Finally, through serial dilution, a series of mixed standard working solutions with concentrations (0.05, 0.1, 0.2, 0.4, 0.5, 1.0, 5.0 μg / mL) are obtained for plotting standard curves. (This step covers nine sugars and aims to establish a broad screening method. Subsequent sample testing results showed that 2FL and HT were present in low and unstable levels in buffalo milk and Holstein milk, so they were not used in the final core identification model, but the method itself has detection capabilities.)

[0078] 2. Sample pretreatment steps

[0079] Take 1 mL of the raw milk sample to be tested, add 1 mL of ultrapure water, and vortex to mix thoroughly. Centrifuge at 12000 rpm for 10 minutes at 4°C, and carefully remove the upper fat layer. Accurately pipette 0.5 mL of the intermediate aqueous phase, add 1 mL of pre-cooled anhydrous ethanol (-20°C), vortex to mix, and place in a -20°C refrigerator to precipitate overnight (at least 12 hours). The next day, centrifuge at 12000 rpm for 10 minutes at 4°C, collect all the supernatant, and dry under a gentle nitrogen stream. Redissolve the residue with 1.0 mL of the initial mobile phase ratio (acetonitrile-water solution), vortexing to aid dissolution. Filter the solution through a 0.22 μm microporous membrane and transfer to an LC-MS / MS vial for analysis.

[0080] 3. Liquid Chromatography Conditions

[0081] The analysis was performed using an ultra-high performance liquid chromatography system.

[0082] Chromatographic column: ACQUITY UPLC BEH Amide Column (130 Å, 1.7 μm, 2.1 mm × 100 mm) or equivalent hydrophilic column.

[0083] Mobile phase: Phase A is an aqueous solution containing 10 mmol / L ammonium formate and 0.1% formic acid; Phase B is acetonitrile.

[0084] The gradient elution procedure is shown in Table 1 below:

[0085]

[0086] Flow rate: 0.3 mL / min; column temperature: 40 ℃; injection volume: 1 μL.

[0087] 4. Mass spectrometry conditions

[0088] A triple quadrupole mass spectrometer equipped with an electrospray ionization (ESI) source was used.

[0089] Ion source parameters: Ion source: Electrospray ionization source (ESI source); Scanning mode: Multiple reaction monitoring (MRM); Electrospray voltage (IS): 5500 V; Ion source temperature (TEM): 500 ℃; Curtain gas (CUR): 45 psi; Nebulizing gas (GS1): 50 psi; Auxiliary gas (GS2): 50 psi.

[0090] Scanning mode: Multiple reaction monitoring (MRM). For each target oligosaccharide, the declustering voltage (DP) and collision energy (CE) were optimized, and two pairs of characteristic ion pairs were selected (one pair for quantification and one pair for qualitative analysis). Detailed MRM mass spectrometry parameters are shown in Table 2.

[0091] Table 2 Mass spectrometry analysis parameters of nine oligosaccharides

[0092]

[0093] Note 1: These are quantitative ions.

[0094] Note 2: The monitoring ion pairs and other measurement conditions provided by the method are recommended conditions. Each laboratory should make appropriate adjustments according to the specific conditions of the instrument it is equipped with. In the case of measurement interference from the sample matrix, other monitoring ion pairs may be selected.

[0095] 5. Methodological Validation Results

[0096] To demonstrate that this method is applicable to the accurate quantification of target oligosaccharides in complex milk matrices, we conducted comprehensive methodological validation.

[0097] 5.1 Linearity Range and Limit of Detection: The seven oligosaccharides showed good linearity within the concentration range of 0.05–5.0 μg / mL, with correlation coefficients (R²) greater than 0.99. The limits of detection (LOD) and quantitation (LOQ) of the method were determined using signal-to-noise ratios (S / N) ≥ 3 and ≥ 10, respectively. The LOQs of all target analytes were below 0.2 μg / mL, fully meeting the requirements for actual sample analysis. Specific LOD / LOQ data are shown in Tables 3 and 4.

[0098] Table 3. Detection Limits (LODs) of Nine Oligosaccharide Detection Methods by Liquid Chromatography-Tandem Mass Spectrometry

[0099]

[0100] Table 4. Limit of Quantitation (LOQ) Results of Nine Liquid Chromatography-Tandem Mass Spectrometry (LC-MS / MS) Detection Methods for Oligosaccharides

[0101]

[0102] 5.2 Precision: Milk samples with the same spiking concentration were injected six times consecutively, and the relative standard deviation (RSD) of the peak area of ​​each target analyte was calculated. The results are shown in Table 5. The RSDs of the seven oligosaccharides ranged from 0.8% to 4.4%, indicating good repeatability of the method.

[0103] Table 5 Precision test results of nine oligosaccharides

[0104]

[0105] 5.3 Accuracy (Recovery Rate): Spiked recovery experiments were conducted at three levels (low, medium, and high) in a blank milk matrix, with six replicates for each level. As shown in Table 6, the average recovery rates of the seven target oligosaccharides ranged from 82.6% to 102.6%, and the RSDs met the requirements for biological sample analysis, demonstrating the accuracy and reliability of the method.

[0106] Table 6 Results of Recovery Rate Tests for Nine Oligosaccharides

[0107]

[0108] 5.4 Specificity: At the target oligosaccharide peak position, the blank matrix sample showed no interfering peaks, and the positive sample exhibited good peak shape and satisfactory resolution. A typical chromatogram is shown below. Figure 6 See Table 7.

[0109] Table 7. Specificity Tests for Nine Oligosaccharide Detection Methods by Liquid Phase Tandem Mass Spectrometry

[0110]

[0111] This method was applied to detect actual samples of buffalo milk and Holstein milk. Since 2FL and HT among the nine oligosaccharides could not be consistently detected due to their low content, seven characteristic oligosaccharides (3FL, 3SL, 6SL, 3SLN, LNT II, ​​LNnT, and Le-XT) were selected for further research. Therefore, this embodiment established and validated an efficient, sensitive, and reliable LC-MS / MS method for the simultaneous quantification of seven characteristic oligosaccharides (3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, and LNT II) in milk. This method features simple pretreatment, good chromatographic separation, high mass spectrometry selectivity, and all methodological indicators meet the requirements for quantitative analysis, providing a solid and reliable data source for the subsequent construction of a high-precision breed identification model.

[0112] Example 3: Construction and Evaluation of a Logistic Regression Classification Model

[0113] This embodiment details how to use the large-scale sample data obtained in Embodiment 2 to construct, train, and validate a high-precision, robust logistic regression classification model as a model for identifying buffalo milk and Holstein milk varieties.

[0114] 1. Data preparation and feature sources:

[0115] The data used in model construction came from targeted detection of known milk samples. The detection method strictly followed the LC-MS / MS quantitative method established in Example 2 of this invention.

[0116] 1.1 Sample composition: A total of 104 mid-lactation milk samples with clear breed identification were collected, including 62 buffalo milk samples (including three breeds of buffalo: Nereid, Mora, and Mediterranean, with lactation days of 50–210 days) and 42 Holstein milk samples (lactation days of 50–200 days).

[0117] 1.2 Characteristic Variables: The content (unit: μg / mL) of seven characteristic oligosaccharides (3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, LNT II) was measured in each sample. These seven sugars are stable and detectable core differential markers screened based on previous non-targeted omics analysis and quantitative methodology validation (Example 2).

[0118] 1.3 Target Variables: The breed information is encoded as a binary variable, where "buffalo milk" is encoded as 1 and "Holstein milk" is encoded as 0.

[0119] 2. Data Preprocessing and Feature Analysis

[0120] Before building the model, the raw content data were preprocessed and statistically analyzed to verify its potential as a classification feature and to meet the model input requirements.

[0121] 2.1 Data Splitting: All 104 samples were randomly divided into training and test sets at a ratio of approximately 4:1. Stratified sampling was used to ensure that the proportion of the two classes was consistent with the original dataset. There were 62 buffalo milk samples and 42 Holstein milk samples.

[0122] 2.2 Feature Standardization: Due to the significant differences in the numerical range of oligosaccharide content, to avoid the adverse effects of unit weighting on model training, all feature data (i.e., the content of each sugar) need to be z-score standardized. The standardization formula is:

[0123]

[0124] z = (x − μ) / σ, where Let $\frac{j}{i}$ be the original content value of the $j$-th oligosaccharide characteristic of the $i$-th sample. Let be the mean content of the j-th feature oligosaccharide in all training samples. Let be the standard deviation of the j-th characteristic oligosaccharide. These are the standardized feature values. After standardization, the mean of each feature is 0, and the standard deviation is 1.

[0125] 3. Construction, training, and evaluation of logistic regression classification models

[0126] 3.1 Model Selection and Construction: A binary logistic regression model was chosen as the classifier. Its core principle is to map the linear combination of features to the probability that a sample belongs to the "buffalo" class using the Logit function. The standardized features are then linearly combined with their weights to obtain the log-odds.

[0127]

[0128] in, The linear output of the i-th sample. The intercept (bias term) of the model. : The weighting coefficient of the j-th feature oligosaccharide Let be the standardized value of the oligosaccharide content of the j-th characteristic; : Number of features (7 oligosaccharides in this case: 3FL, 3SL, 6SL, 3SLN, LNT II, ​​LNnT, Le-XT).

[0129] Then, the Sigmoid function is used to... Convert the probability that the sample belongs to Holstein milk. :

[0130]

[0131] in, It is the probability that the i-th sample is a Holstein cow (label=1), and e is the natural constant (≈2.718).

[0132] 3.2 Model Training and Parameter Estimation

[0133] The 104 samples were randomly divided into a training set (83 samples) and a test set (21 samples). Using the training set data, the log loss function was minimized using gradient descent, and the model parameters were iteratively optimized. and ).

[0134] 3.2.1 Define the loss function

[0135] For a training set containing N samples, the log loss function is defined as:

[0136]

[0137] Where: N: total number of samples, : No. The true labels of each sample (0 / 1) This is the probability that the model predicts the sample to be of the positive class (Holstein milk). This function quantifies the difference between the model's prediction and the true value, and our goal is to minimize this function value.

[0138] 3.2.2 Calculate the gradient

[0139] To minimize the loss function, we need to calculate the partial derivatives (i.e., gradients) of the loss function with respect to each parameter $w_j$ and $b$. The gradient calculation formula is derived as follows:

[0140]

[0141]

[0142] The gradient indicates the direction in which the loss function increases the fastest at the current parameter values. Therefore, by updating the parameters in the opposite direction of the gradient, we can gradually reduce the loss.

[0143] 3.2.3 Iterative parameter update

[0144] Set learning rate (In this embodiment, we take) =0.01), and iteratively update all parameters according to the following rules until the loss function converges or the preset maximum number of iterations is reached:

[0145]

[0146]

[0147] Through the above iterative optimization process, the final set of optimal model parameters is shown in Table 8.

[0148] Table 8 Model Parameter Table

[0149]

[0150] 3.2.4 Final Model Output: After the model performance is evaluated and confirmed to meet the standards, the intercept obtained from training on the full training set (83 samples) and the mean and standard deviation of each feature required for standardization are solidified as the final variety identification model. This model can be represented as follows:

[0151]

[0152]

[0153] P

[0154] When inputting the content of seven sugars in a new sample to be tested, the linear combination value must first be calculated. Then, substitute the values ​​into the model formula to calculate the probability P. Set the discrimination threshold to 0.5. If P ≥ 0.5, it is judged as Holstein milk; if P < 0.5, it is judged as buffalo milk.

[0155] 4. Comprehensive validation of model performance

[0156] To ensure the reliability and generalization ability of the model, we adopted a variety of strategies, including manual verification, confusion matrix analysis, and cross-validation.

[0157] 4.1 Manual verification

[0158] To verify the accuracy of the model calculations, we manually verified the calculations on the samples in the test set according to the above formula.

[0159] Verification Case 1: Taking a typical buffalo milk sample as an example, the calculation process and results are shown in Table 9. The verification results of all test samples are consistent with the actual categories, with an accuracy rate of 100%.

[0160] The calculation process is shown in Table 9:

[0161]

[0162] Detailed calculation instructions:

[0163] (1) Calculation of standardized values:

[0164] 3FL: =(0.816-1.-1.208) / 1.469 = -0.267

[0165] 3SL: = (18.717 - 38.469) / 14.871 = -1.328

[0166] 6SL: =(6.799-19.364) / 15.565=-0.807

[0167] 3SLN: =(0.768−1.588) / 0.826 = -0.993

[0168] LNT II: =(47.311−47.392) / 25.387= -0.003

[0169] LNnT: =(0.778−0.386) / 0.320 = 1.226

[0170] Le-XT: =(0.439-0.409) / 0.521 = 0.057

[0171] (2) Calculation of contribution value

[0172] 3FL: (-0.479375)x(-0.266851) = +0.127922

[0173] 3SL: (1.524488)x(-1.328260) = -2.024916

[0174] 6SL: (-1.387730)x(-0.807296) = +1.120309

[0175] 3SLN: (0.768616)x(-0.992545) = -0.762886

[0176] LNT II: (0.313290)x(-0.003190) = -0.000999

[0177] LNnT:(-2.133230)x(1.226129) = -2.615615

[0178] Le-XT: (-0.628087)x(0.056928)= -0.035756

[0179] (2) Calculation of linear combination value:

[0180] =−1.545637+(0.127922)+(1.120309)+(-2.024916)+(-0.762886)+(−0.000999)+(-2.615615)+(−0.035756)=−5.737578

[0181] (4) Probability calculation:

[0182]

[0183] (5) Judgment result: Due to 0.003212 < 0.5, therefore it is determined to be buffalo milk, which is consistent with the actual category.

[0184] Verification Case 2: Taking a typical Holstein milk sample as an example:

[0185] Input values ​​(μg / mL): 3FL=0.252, 3SL=46.777, 6SL=11.610, 3SLN=5.722, LNT II=33.49, LNnT=0.050, Le-XT=0.141

[0186] Judgment result: The calculated value is P≈0.772, which is greater than 0.5. Therefore, it is judged to be Holstein milk, which is consistent with the actual category.

[0187] The calculations were performed on all 21 samples in the test set, and the results were consistent with the actual categories, achieving an accuracy of 100%. Partial verification results for the training set are shown in the table below. Verification method: Ten training samples were randomly selected, and the calculation steps of Case 1 were repeated to verify the consistency between the predicted results and the actual categories.

[0188] Table 10 Sample Verification Display

[0189]

[0190] 4.2 Confusion Matrix and ROC Curve Analysis

[0191] All 104 samples were used as the test set for back-substitution testing, and confusion matrices and receiver operating characteristic (ROC) curves were plotted. The results showed that all samples were correctly classified, the model accuracy was 100%, and the area under the ROC curve (AUC) was 1.0, indicating that the model has perfect discriminative ability (see...). Figures 7-8 ).

[0192] 4.3 Five-fold cross-validation

[0193] To further evaluate the model's generalization ability, 5-fold cross-validation was performed on all 104 samples. The data were randomly divided into 5 groups, with 4 groups used for training and 1 group for testing each time, repeated 5 times. The average accuracy of cross-validation was 99.0%, and the standard deviation was 2.24%, indicating that the model has good stability and generalization ability, and no obvious overfitting was observed (see Table 11).

[0194] Table 11 Validation results of the model's five-fold difference

[0195]

[0196] As shown above, the cross-validation results are: average accuracy (100.0% + 100.0% + 100.0% + 100.0% + 100.0%) / 5 = 99.0%, standard deviation is 2.24%, which is consistent with the model cross-validation results, proving that the model has good stability and generalization ability.

[0197] Example 3: Statistical differences and practical application verification of seven characteristic oligosaccharides in distinguishing between buffalo milk and Holstein milk.

[0198] This embodiment primarily elucidates the discriminative power of the selected features from a statistical perspective and verifies their identification effect in actual samples.

[0199] 1. Statistical analysis: In-depth statistical analysis was performed on the data of seven characteristic oligosaccharides from 104 samples in Example 2.

[0200] 1.1 Significance Test (t-test / ANOVA): The p-values ​​of each oligosaccharide in the two types of milk were calculated to determine whether the differences were significant. The results showed that the p-values ​​of all seven oligosaccharides were less than 0.05, indicating statistically significant differences. Among them, except for LNT II, ​​the differences of the other six sugars were extremely significant (see Table 12).

[0201] 1.2 Effect Size Analysis (Cohen's d): Effect sizes were calculated to measure the magnitude of the differences. The effect sizes of the six sugars other than LNT II, ​​which were >1 or < -1, were considered "maximum effects," indicating their importance for classification and also serving as specific markers for buffalo milk (see Table 12).

[0202] Table 12 Summary of Significance Difference Tests

[0203]

[0204] 1.3 Descriptive statistics: The mean and standard deviation of each oligosaccharide were compared in the two types of milk. For example, the content of 3SL in Holstein milk (mean 52.575 μg / mL) was 1.818 times that in buffalo milk (mean 28.913 μg / mL); while 3SLN was only detected in buffalo milk (mean 1.46 μg / mL) and was not detected at all in Holstein milk (see Table 14).

[0205] Table 14 Descriptive statistical comparison of oligosaccharide content in normal milk

[0206]

[0207] 2. Practical Application Verification

[0208] 2.1 Colostrum Identification: Analysis of colostrum samples from Holstein and buffalo (Niri, Mora, Mediterranean) revealed extremely high levels of 3SLN in buffalo colostrum (up to 942.459 μg / mL), a characteristic component. In Holstein colostrum, 3SL was absolutely dominant (89.80%). This characteristic can effectively distinguish colostrum from 3SL (see...). Figure 10 and Figure 11 ).

[0209] 2.2 Identification of Standard Milk: Based on the oligosaccharide content of the two types of standard milk, buffalo milk still had a significantly higher total oligosaccharide content than Holstein milk. In Holstein milk, 3SL was the dominant oligosaccharide; while in buffalo milk, the proportions of LNT II, ​​3SL, and 6SL were similar, while the contents of 3FL, LNnT, and Le-XT were all relatively low. PCA analysis based on the seven oligosaccharides showed that the sample points from both types of milk were completely separated in the principal component space with no overlap, proving that this characteristic combination can be effectively used for the identification of standard milk (see...). Figure 12 and Figure 13 ).

[0210] Example 4: A method for differentiating between buffalo milk and Holstein milk

[0211] See Figure 14 This embodiment provides a practical identification method flow based on the aforementioned model.

[0212] S401: Sample Testing. Upon receiving the milk sample to be tested, the contents of seven characteristic oligosaccharides (3FL, 3SL, 3SLN, 6SL, Le-XT, LNnT, LNT II) were quantitatively determined according to the LC-MS / MS method described in Example 2.

[0213] S402: Data Standardization. The original content of each sugar measured in step S401 is... (Unit: μg / mL), using the corresponding mean and standard deviation parameters provided in Table 2 of Example 3 of this invention, z-score standardization was performed to obtain the standardized eigenvalue xi. The standardization calculation formula is: .in, and The values ​​are the mean and standard deviation for each feature, specifically: 3FL: μ=1.208, σ=1.469; 3SL: μ=38.469, σ=14.871; 6SL: μ=19.364, σ=15.565; 3SLN: μ=1.588, σ=0.826; LNT II: μ=47.392, σ=25.387; LNnT: μ=0.386, σ=0.320; Le-XT: μ=0.409, σ=0.521.

[0214] S403: Model Prediction. The normalized eigenvalues ​​of the seven characteristic oligosaccharides calculated in step S402 are then used. This is then substituted into the logistic regression classification model constructed and trained in Embodiment 3 of the present invention. First, the linear combination value z is calculated.

[0215]

[0216] Subsequently, the probability value P of the test sample being Holstein milk was calculated using the Sigmoid function. .

[0217] S404: Result Determination and Output. Set the determination threshold to 0.5. Compare the probability P calculated in step S403 with the threshold. If P ≥ 0.5, the system determines the sample as "Holstein milk"; if P < 0.5, it determines it as "buffalo milk". Finally, output the determination result in a visual form (e.g., screen display, printed report) or electronic data form (e.g., output via data interface).

[0218] Example 5: A system for differentiating buffalo milk from Holstein milk

[0219] Please refer to Figure 15 This embodiment provides a system 500 for implementing the above method.

[0220] Data acquisition and input module 510: Used to acquire the raw content data of seven characteristic oligosaccharides in the milk sample to be tested. Specifically, this module can establish a communication connection with a liquid chromatography-mass spectrometry (LC-MS / MS) instrument and automatically receive the quantitative detection results generated by the instrument. In addition, this module can also provide a manual input interface, allowing users to manually enter historical detection data or third-party detection data.

[0221] Data preprocessing module 520: Connected to data acquisition and input module 510, it is used to standardize the received raw content data. This module internally stores fixed parameters, namely the mean (μ) and standard deviation (σ) of the seven characteristic oligosaccharides listed in Table 2 of Example 3. Its core function is to call these parameters and process them according to the formula... Perform standardized calculations and finally output standardized feature values.

[0222] Core discrimination module 530: Connected to data preprocessing module 520, its core is a pre-trained and solidified logistic regression classification model. This model encapsulates the intercept (b=-1) determined in Example 3. The module receives the standardized feature values ​​and automatically performs the matrix multiplication and logical function operations described in step S403 of Example 4, that is, calculates the linear combination value z and the probability P, and finally outputs the predicted probability P.

[0223] Result generation and output module 540: Connected to the core discrimination module 530, it receives the probability value P and compares it with a preset threshold of 0.5 to generate a clear identification conclusion of "buffalo milk" or "Holstein milk". This module supports multiple output methods, including but not limited to: displaying the conclusion on a graphical user interface (GUI), generating a test report (such as PDF format) containing raw data, standardized values, probabilities, and final conclusions, or transmitting the conclusion to a laboratory information management system (LIMS) or other external devices via a standard data interface.

[0224] Example 6: Storage Media and Electronic Devices

[0225] This embodiment provides a computer-readable storage medium (such as a USB flash drive, hard disk, or cloud storage space) on which a computer program is stored. When the program is executed by a processor (such as the CPU of a server, personal computer, or embedded device), it can control relevant hardware (such as an autosampler or mass spectrometer) or process existing data, thus fully implementing all the steps of the identification method described in Embodiment 4.

[0226] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory. When the processor executes the program, it implements the identification method described in Embodiment 4. This electronic device can be a dedicated testing device, a Laboratory Information Management System (LIMS) server, or a regular personal computer.

[0227] In summary, this invention, through a rigorous research path of "discovery-verification-application," proposes and verifies for the first time a high-precision identification scheme for buffalo milk and Holstein milk based on seven characteristic oligosaccharide content patterns. This scheme combines advanced LC-MS / MS detection technology with a robust logistic regression model, achieving automated and intelligent conversion from chemical information to breed identification. Compared with existing technologies, the identification method provided by this invention has clear markers, near-perfect accuracy, and strong anti-interference capabilities, offering a novel and powerful technical tool for authenticity identification and quality control in the dairy industry.

[0228] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A method for distinguishing between buffalo milk and Holstein milk, characterized in that, Includes the following steps: S1. Detect the content of seven characteristic oligosaccharides in the milk sample to be tested. The seven characteristic oligosaccharides are: 3-fucosylated lactose (3FL), 3'-sialylated lactose (3SL), 3'-sialylated N-acetyllactosamine (3SLN), 6'-sialylated lactose (6SL), Lewis-X trisaccharide (Le-XT), lactose-N-neotetrasaccharide (LNnT), and lactose-N-trisaccharide (LNT II). S2. Standardize the content of the seven oligosaccharides obtained from the test; S3. Input the standardized content data of the seven characteristic oligosaccharides into the trained logistic regression classification model to obtain the predicted probability P that the milk sample to be tested is buffalo milk. S4. Output the identification result based on the predicted probability P: if P ≥ 0.5, it is determined to be Holstein milk; if P < 0.5, it is determined to be buffalo milk.

2. The method according to claim 1, characterized in that, In step S1, the content of the seven characteristic oligosaccharides is detected by high performance liquid chromatography-tandem mass spectrometry.

3. The method according to claim 2, characterized in that, The high-performance liquid chromatography-tandem mass spectrometry method uses a hydrophilic interaction column and gradient elution with an aqueous solution containing ammonium formate and acetonitrile as the mobile phase. The mass spectrometry uses electrospray positive ion mode and multiple reaction monitoring mode.

4. The method according to any one of claims 1-3, characterized in that, In step S2, the standardization process involves standardizing the content xi of each oligosaccharide using the following formula: ,in, This represents the original content value of the j-th feature oligosaccharide in the i-th sample; The average content of the j-th characteristic oligosaccharide; Let $\frac{j}{j}$ be the standard deviation of the $j$-th characteristic oligosaccharide. These are the standardized feature values.

5. The method according to claim 1, characterized in that, The logistic regression classification model calculates the probability value using the following formula. : in, It is the probability that the i-th sample is a Holstein cow (label=1), and e is the natural constant (≈2.718). z is the value of the linear combination, and its calculation formula is as follows: in, This is the linear output of the i-th sample; The model intercept; is the weighting coefficient for the j-th characteristic oligosaccharide; Let be the standardized value of the oligosaccharide content of the j-th characteristic; This represents the number of characteristic oligosaccharide types.

6. The method according to any one of claims 1-5, characterized in that, The training process of the logistic regression model includes: Step A1: Collect milk samples from known breeds of buffalo and Holstein cows to form a training set; Step A2: Detect the content data of the characteristic oligosaccharides for each sample in the training set; Step A3: Perform Z-score standardization on the feature oligosaccharide content data of the training set to obtain the standardized feature values; Step A4: Based on the standardized feature values ​​and their corresponding variety labels, the logistic regression model is trained using gradient descent to learn the model parameters, which include the intercept b and the weight coefficient of each feature oligosaccharide. .

7. The method according to claim 6, characterized in that, The model parameters include: intercept b = -1.545637; weight coefficient for 3-fucoyllactose (3FL) = -0.479375; weight coefficient for 3-sialyllactose (3SL) = 1.524488; weight coefficient for 3-sialyl-N-acetyllactosamine (3SLN) = 0.768616; weight coefficient for 6-sialyllactose (6SL) = -1.387730; weight coefficient for Lewis X trisaccharide (Le-XT) = -0.628087; weight coefficient for lactose-N-trisaccharide (LNT II) = 0.313290; weight coefficient for lactose-N-neotetrasaccharide (LNnT) = -2.133230; the linear combination value z is obtained from the above model parameters: 。 8. An identification system for implementing the method according to any one of claims 1-7, characterized in that, include: The data acquisition module is used to acquire the content data of seven characteristic oligosaccharides in the milk sample to be tested; The data processing module is used to standardize the detection data; The model calculation module stores the trained logistic regression classification model and is used to receive the processed data and calculate the prediction probability P. The result output module is used to output the identification result based on the predicted probability.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method according to any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1-7.

Images