A method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm

Abstract

This application relates to the field of liquor analysis and testing technology, and in particular to a method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with the PLS-R algorithm. The method includes steps such as Maotai-flavor liquor sample collection, ultra-high performance liquid chromatography-high resolution mass spectrometry data acquisition, data processing and compound annotation, differential compound screening, vintage identification model construction, and vintage identification of the sample to be tested. This application can achieve accurate vintage identification of Maotai-flavor liquor, and is simple to operate, has a short detection cycle, good repeatability, and objective and controllable identification results. It fills the gap in the incomplete detection of non-volatile vintage characteristic substances in existing technologies, effectively solves the technical problem of vintage identification of Maotai-flavor liquor, and has important application value for curbing market counterfeiting, protecting consumer rights, and promoting the healthy development of the industry.

Description

Technical Field

[0001] This invention belongs to the field of food analysis and testing technology, specifically a method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with the PLS-R algorithm. Background Technology

[0002] With the rise of consumption upgrades and the craze for collecting Maotai-flavor liquor, aged liquors, with their richer taste and higher quality, have become a core high-end category in the liquor market and are highly sought after by consumers. However, limited by the huge premium of aged liquors, some unscrupulous merchants have engaged in fraudulent activities such as falsely labeling storage years and blending new liquor to imitate old liquor, leading to frequent irregularities in the market, seriously damaging consumer rights and disrupting the normal order of the industry. Therefore, establishing a scientific, accurate, and quantifiable method for identifying the age of Maotai-flavor liquor has become an urgent need to ensure market credibility and promote the healthy development of the industry.

[0003] Currently, the main techniques for identifying vintage wines include sensory evaluation, gas chromatography, gas chromatography-mass spectrometry, spectroscopy, and stable isotope mass spectrometry. While sensory evaluation can quickly and roughly determine the vintage, it heavily relies on the experience and subjective judgment of the tasters, resulting in poor repeatability and making it difficult to establish a unified and objective identification standard. Gas chromatography and spectroscopy mainly determine the vintage by analyzing volatile flavor components or spectral characteristics combined with predictive models, but these methods have significant limitations in detecting non-volatile or thermally unstable substances and require a large amount of sample data to support model construction, thus limiting their applicability. Stable isotope technology estimates the vintage based on the decay patterns of radioactive isotopes during storage, but its accuracy is greatly affected by sample background and environmental factors, making it difficult to meet the requirements of high-precision identification.

[0004] During long-term storage, Maotai-flavor liquor undergoes a series of complex chemical transformations, including Maillard reaction, esterification reaction, and redox reaction, generating a large number of non-volatile or thermally unstable substances, such as organic acids, higher fatty acids and their esters, amino acids and short peptides, and Maillard reaction intermediates (e.g., 5-hydroxymethylfurfural). The content of these substances changes regularly with storage time, containing rich information about the vintage. However, current research has not systematically incorporated these non-volatile components into the vintage identification system, making it difficult to fully reveal their dynamic changes during storage and to construct a high-precision vintage identification model.

[0005] Ultra-high performance liquid chromatography-high resolution mass spectrometry (UHPLC-MS / MS), with its advantages of high resolution, high sensitivity, and high quality and precision, can simultaneously detect and accurately identify multiple components in complex matrices, and has received increasing attention in the field of baijiu research in recent years. Although some researchers have used UHPLC-MS / MS to analyze key differentiating components among baijiu from different years, related research is still at the preliminary screening stage, lacking in-depth and systematic analysis of the changes in non-volatile components over time, and has not yet formed a high-precision vintage identification model based on a combination of multiple biomarkers. Summary of the Invention

[0006] The purpose of this invention is to provide a method for identifying the vintage of Maotai-flavor liquor based on ultra-high performance liquid chromatography-high resolution mass spectrometry combined with a chemometric model, thereby overcoming many shortcomings of existing Maotai-flavor liquor vintage identification techniques.

[0007] To achieve the above-mentioned objectives, the technical solution of the present invention is as follows: A method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm, comprising the following steps: S1. Sample Collection of Maotai-flavor Baijiu: Samples of Maotai-flavor Baijiu from multiple manufacturers with different storage years were collected. Multiple replicate samples were set up for each year, and each sample was taken independently. The sample solution was directly injected for analysis without the need for complicated sample pretreatment. The multiple manufacturers covered different enterprises in the core production area of ​​Maotai-flavor Baijiu. The different storage years included storage periods of 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 9 years, and 12 years. The diversity of sample sources ensured the universality and reliability of the method developed for identifying the year of Maotai-flavor Baijiu from different manufacturers and sources. S2. Data acquisition by ultra-high performance liquid chromatography-high resolution mass spectrometry: Non-targeted metabolomics data acquisition of sample solutions was performed using an ultra-high performance liquid chromatography-high resolution mass spectrometry system. S3. Data Processing and Compound Annotation: Import the raw mass spectrometry data into the data processing software for peak detection, peak alignment, gap filling and normalization. Based on the accurate molecular weight, retention time, isotope distribution and secondary mass spectrometry fragment information, compare with the mass spectrometry database to annotate the compounds. S4. Screening of year-related differential compounds: Multivariate statistical analysis was used to screen differential compounds related to the storage year. A principal component analysis model was constructed to remove outliers. An orthogonal partial least squares-discriminant analysis model (OPLS-DA) was constructed. Variable projection importance (VIP) was used for screening. After FDR correction, year-characteristic differential compounds were obtained. S5. Construction of the year identification model: The peak area of ​​the compounds with the year-specific differences is used as the input variable, and the partial least squares regression algorithm is used to construct the year identification model of Maotai-flavor liquor. S6. Identification of the year of the liquor sample to be tested: Data is collected from the liquor sample to be tested, the peak area of ​​the characteristic difference compound of the year is extracted, and after normalization and standardization, it is input into the liquor year identification model mentioned above, and the predicted year of the sample to be tested is output.

[0008] Preferably, in step S2, an HSS T3 column is used for chromatographic separation. The column has an inner diameter of 2.1 mm, a length of 100 mm, a particle size of 1.8 μm, a column temperature of 40 °C, a mobile phase A of 0.1% formic acid aqueous solution, a mobile phase B of methanol, a flow rate of 0.3 mL per minute, and an injection volume of 1 μL.

[0009] Preferably, in step S2, the mobile phase gradient elution program is set as follows: during the 0-1 minute period, the volume ratio of mobile phase B is 2%; during the 1-15 minute period, the volume ratio of mobile phase B increases linearly from 2% to 98%; during the 15-17 minute period, the volume ratio of mobile phase B remains at 98%; during the 17-17.1 minute period, the volume ratio of mobile phase B decreases from 98% to 2%; and during the 17.1-20 minute period, the volume ratio of mobile phase B remains at 2%. Preferably, in step S2, the stationary phase of the HSS T3 chromatographic column is a trifunctional bonded silica gel packing material, which has good retention characteristics for both polar and non-polar compounds in the wine sample. It can achieve the synchronous separation of the two types of substances under the same chromatographic conditions, effectively avoiding the problem of missed detection of certain characteristic compounds of certain years due to excessive polarity differences.

[0010] Preferably, in step S2, an electrospray ionization source is used as the ionization method, and data is acquired simultaneously in positive and negative ion modes. The first-stage mass spectrometry scan range is set to 70-800 Da with a resolution of 120,000, the second-stage scan uses a data-dependent acquisition mode with a resolution of 15,000, and collision-induced dissociation uses a step collision energy mode with collision energies set to 20%, 40%, and 60%, respectively. The simultaneous acquisition mode of positive and negative ions allows for the acquisition of complete mass spectrometry data in both positive and negative ion modes in a single injection, significantly improving analytical throughput. The step collision energy mode, by fragmenting at three different collision energies, can obtain multi-stage mass spectrometry information of the same compound at different fragmentation energies, thereby improving the accuracy and coverage of compound structure resolution and reducing the probability of false positive interference.

[0011] Preferably, the spray voltage for positive ion mode is set to 3.5kV, the spray voltage for negative ion mode is set to 2.5kV, the temperature of the ion transmission tube is set to 320℃, the sheath gas flow rate is set to 50 Arb, the auxiliary gas flow rate is set to 10 Arb, and the purge gas flow rate is set to 1 Arb.

[0012] Preferably, in step S3, the acquired raw mass spectrometry data is imported into Compound Discoverer software for peak detection, peak alignment, gap filling, and normalization. The compound is annotated based on the accurate molecular weight, retention time, isotope distribution, and secondary mass spectrometry fragment information compared with the mass spectrometry database. The mass spectrometry database includes the mzCloud database, ChemSpider database, LipidSearch database, and a locally built database. The error range of the accurate molecular weight is controlled within 5 ppm to ensure the accuracy of the compound qualitative results.

[0013] Preferably, in step S4, a multivariate statistical analysis method is used to screen differential compounds related to the storage year, specifically including the following steps: First, an unsupervised principal component analysis model is constructed to observe the inter-group separation trend of samples from different years and remove outliers; then, an orthogonal partial least squares-discriminant analysis model is constructed, and the reliability of the model is verified by permutation test; the number of permutation tests is not less than 200 to ensure that the model does not overfit; differential compounds are screened after FDR correction using VIP>1 as the screening criterion; the orthogonal partial least squares-discriminant analysis model uses supervised data dimensionality reduction technology, which can maximize the extraction of variable information related to year grouping and effectively remove orthogonal variables that are irrelevant to the year; the variable projection importance value reflects the degree of contribution of each variable to the year classification, and the larger the variable projection importance value, the greater the contribution of the variable to the year distinction; after screening by the joint screening criteria, the number of year-characteristic differential compounds finally obtained is 40.

[0014] Preferably, in step S5, the peak areas of the 40 compounds with different characteristics of different years are used as input variables, and a partial least squares regression algorithm is used to construct a model for identifying the year of Maotai-flavor liquor. The specific steps for constructing the model are as follows: the sample data of Maotai-flavor liquor from different years are randomly divided into training set, validation set and test set in proportions of 70%, 15% and 15% respectively; the model is trained using the training set data; the model parameters are optimized using cross-validation, and the number of cross-validation iterations is not less than 5; the model is validated using the validation set data; the accuracy of the calibrated model is comprehensively evaluated by the calibration root mean square error, validation root mean square error, calibration coefficient and validation coefficient; the predictive performance of the model is independently evaluated using the test set data.

[0015] Furthermore, the method also includes a step of establishing a database of year-characteristic compounds after the year identification model construction step and before the year identification step of the sample to be tested: the standard spectra, retention times, precise molecular weights, secondary mass spectrometry fragment information and quantitative information of the 40 core marker combinations in sauce-flavored liquor samples from multiple manufacturers with different storage years are entered into the database to form a standard reference database; the standard reference database can serve as a standard basis for rapid year comparison and identification, which significantly improves detection efficiency in practical applications.

[0016] In the process of identifying the year of the liquor sample to be tested, the sample of sauce-flavored liquor to be tested is taken, and the sample is processed and data is acquired in accordance with the aforementioned sample pretreatment steps and ultra-high performance liquid chromatography-high resolution mass spectrometry data acquisition steps. The peak areas of the 40 compounds with different characteristics of the year are extracted, and after data preprocessing such as normalization and standardization, the data is input into the year identification model. The model outputs the predicted year of the sample to be tested.

[0017] Beneficial effects: Compared with the prior art, the present invention has the following characteristics: 1. The technical solution of this invention addresses the objective laws governing the complex chemical transformation processes, such as Maillard reaction, esterification reaction, and redox reaction, that continuously occur in Maotai-flavor liquor during long-term storage. It utilizes ultra-high performance liquid chromatography-high resolution mass spectrometry (UHPLC-HDMS) to comprehensively detect non-volatile components in the liquor, analyze the changes in these components over time, screen for characteristic markers closely related to storage years, and reveal the dynamic changes in the content of organic acids, higher fatty acids and their esters, amino acids and short peptides, and Maillard reaction intermediates over storage time. These non-volatile components are incorporated into the vintage identification system, and a high-precision vintage prediction model is constructed based on a combination of multiple markers. This fills the gap in the incomplete detection of non-volatile vintage characteristic substances in existing technologies, effectively solving the technical challenge of vintage identification for Maotai-flavor liquor, curbing market counterfeiting, protecting consumer rights, and ensuring the healthy development of the industry.

[0018] 2. This invention achieves accurate screening of 40 compounds with distinct year characteristics through a dual verification mechanism combining orthogonal partial least squares-discriminant analysis and variable projection importance, avoiding false positives that may occur with a single screening criterion. The partial least squares regression year identification model constructed based on these 40 compounds, validated on a test set, shows a root mean square error of approximately 1.056 years on the validation set. Both the calibration and validation coefficients of the model exceed 0.9, and the prediction error for over 97% of the test samples is controlled within ±1 year. High-precision year prediction of Maotai-flavor liquor can be achieved using only these 40 markers.

[0019] 3. The technical solution of this invention adopts a direct injection analysis method, which eliminates the need for complex extraction, concentration or derivatization of samples, greatly simplifies the pretreatment process, shortens the single sample analysis cycle, and increases the detection throughput. The entire analysis process and data processing are automatically completed by instruments and chemometrics software, without relying on the subjective experience of tasters, eliminating the interference of human factors on the identification results. The identification results are objective, controllable and have good repeatability, providing a reliable technical means for the scientific identification of aged Maotai-flavor liquor. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method for identifying the vintage of soy sauce-flavored baijiu according to the present invention; Figure 2 This is a fitting graph of the predicted value and reference value of the PLS-R year recognition model in an embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0022] like Figure 1 As shown, a method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm is presented. The method includes the following seven main steps: Maotai-flavor liquor sample collection, LC-HRMS method research, DDA data collection of liquor samples, data processing, OPLS-DA model establishment, PLS-R regression prediction, and vintage prediction of liquor samples.

[0023] In the collection of samples of Maotai-flavor liquor, to ensure the broad applicability and reliability of the established vintage identification method, the samples covered multiple enterprises in the core production area of ​​Maotai-flavor liquor. Maotai-flavor liquors with an alcohol content of 53±1% vol and different storage years were selected as the research objects. Specifically, the selected storage years included eight different storage periods: 1 year, 2 years, 3 years, 4 years, 5 years, 6 years, 9 years, and 12 years. Forty bottles were collected from each year, and each bottle was independently sampled to form an independent sample, totaling 40 samples per year, for a total of 240 original samples collected from the eight years. In addition, during the sample collection process, a quality control sample was interspersed every 10 samples, for a total of 28 quality control samples. The original samples and quality control samples totaled 268 samples. The quality control samples were used to monitor the stability of the instrument throughout the data collection process, ensuring the repeatability and reliability of the collected data. The above sample collection strategy fully considers the possible differences in processes and raw materials between different manufacturers, so that the final vintage identification model can be applied to sauce-flavored liquor samples from different companies and batches, rather than being effective only for samples from a single source.

[0024] In the LC-HRMS method study, researchers conducted a systematic methodological study on the analytical method of ultra-high performance liquid chromatography-high resolution mass spectrometry, determined the optimal chromatographic separation conditions and mass spectrometry detection parameters, and provided a stable and reliable technical platform for subsequent large-scale sample data acquisition.

[0025] Chromatographic separation was performed using a Waters ACQUITY Premier HSS T3 column. The stationary phase of this column is trifunctional bonded silica gel with a particle size of 1.8 micrometers. The column length is 100 mm and the inner diameter is 2.1 mm. This column model was chosen because the HSS T3 stationary phase, after trifunctional bond modification, exhibits significantly enhanced retention of polar compounds while maintaining good retention characteristics for nonpolar compounds. The chemical composition of Maotai-flavor liquor contains both highly polar organic acids, amino acids, Maillard reaction products, and other non-volatile substances, as well as highly nonpolar substances such as higher fatty acid esters. Using the HSS T3 column allows for the simultaneous separation of polar and nonpolar substances under the same chromatographic conditions. This avoids the problem of certain vintage characteristic compounds being eluted prematurely due to weak retention or difficult elution due to excessively strong retention caused by large polarity differences, thus ensuring good peak shape and resolution for vintage characteristic compounds of different polarity ranges.

[0026] The mobile phase of liquid chromatography consists of two phases: mobile phase A is an aqueous solution containing 0.1% formic acid, and mobile phase B is pure methanol. The addition of formic acid plays two important roles: first, as an acidic additive, formic acid can inhibit the ionization of organic acid compounds, improving the ionization efficiency of these substances in the electrospray ionization source; second, formic acid can form a buffer system with water in the mobile phase, maintaining the stability of the mobile phase pH, which is beneficial for the reproducibility of retention times for acidic compounds. Methanol, as the organic phase, has strong elution power and good solvent properties, effectively dissolving nonpolar compounds and eluting them from the stationary phase of the column.

[0027] The gradient elution program is set as follows: In the initial stage of data acquisition, from 0 to 1 minute, the proportion of mobile phase B is maintained at 2%. At this time, the column is in an initial equilibrium state with a low organic phase ratio, mainly used to separate highly polar water-soluble compounds, while ensuring sufficient retention of polar components in the sample at the column inlet. From 1 to 15 minutes, the proportion of mobile phase B increases linearly from 2% to 98%. This stage is the main gradient elution stage. As the organic phase ratio gradually increases, the interfacial tension between the stationary phase and the mobile phase continuously decreases, achieving gradient elution from polar compounds to moderately polar compounds and then to non-polar compounds on the column. This allows compounds of different polarity ranges to be separated according to their polarity. The organic phases elute from the column in descending order of size and are then detected by the mass spectrometer. From 15 to 17 minutes, the proportion of mobile phase B is maintained at 98%. This high organic phase concentration elutes all residual nonpolar compounds from the column, ensuring thorough cleaning at the end of each analysis cycle and providing clean chromatographic conditions for the next sample. From 17 to 17.1 minutes, the proportion of mobile phase B rapidly decreases from 98% to 2%. This stage is the column rebalancing phase, where the rapid reduction of the organic phase restores the stationary phase to its initial state. From 17.1 to 20 minutes, the proportion of mobile phase B is maintained at 2%, ensuring the column reaches complete equilibration before the next sample injection. The total analysis time for this gradient elution program is 20 minutes. Combined with a flow rate of 0.3 mL / min and an injection volume of 1 μL, this approach ensures thorough separation of various compounds in complex samples while keeping the analysis cycle for individual samples within a reasonable range, achieving an optimal balance between resolution and analytical throughput.

[0028] The column temperature for liquid chromatography was set at 40℃. This temperature condition was determined after methodological optimization, which can reduce the potential degradation of thermally unstable compounds by high temperatures while ensuring chromatographic resolution. The spray voltage for mass spectrometry was set to 3.5 kV in positive ion mode and 2.5 kV in negative ion mode. These different spray voltage settings are to accommodate the different voltage requirements of the two ion modes. Positive ion mode typically requires a higher spray voltage to achieve better ionization efficiency, while negative ion mode can achieve stable ion generation at a lower spray voltage.

[0029] In the DDA data acquisition of baijiu samples, an electrospray ionization (ESI) source was used as the ionization technique to convert compound molecules eluted from liquid chromatography into gaseous ions, which were then introduced into the mass spectrometer for analysis. The working principle of the ESI source is that under the action of a high electric field, droplets ejected from the tip of the spray needle undergo Coulomb explosions, generating charged droplets. Subsequently, under the action of a dry gas, the solvent in the droplets gradually evaporates. During solvent evaporation, the droplet volume shrinks, leading to a continuous increase in surface charge density. When the Coulomb repulsion exceeds the surface tension of the droplet, the droplet breaks down, ultimately producing individual gaseous ions that enter the mass analyzer. This ionization process is a soft ionization technique, enabling target compounds to be ionized and enter mass spectrometry analysis with almost no fragmentation. It is particularly suitable for analyzing complex sample matrices like baijiu, which contain a large number of polar and thermally unstable compounds.

[0030] The mass analyzer employs an electrostatic field orbital ion trap, which boasts extremely high mass resolution and accuracy, enabling high-resolution detection over a wide mass range. Specific parameter settings are as follows: Primary mass spectrometry uses full-scan mode for data acquisition, with a scan range of m / z 70 to 800 Da. This mass range covers the vast majority of small molecule compounds in baijiu, including organic acids, amino acids, short peptides, Maillard reaction products, and lower fatty acid esters. The primary mass spectrometry mass resolution is set to 120,000. This extremely high resolution completely separates different compounds with similar mass numbers in the mass spectrum, avoiding peak identification errors caused by mass number overlap. Automatic gain control is set to Standard mode to optimize ion filling time, maintaining the ion quantity in the ion trap at an optimal level. The mass error tolerance is set to 5 ppm. Compound matching within this error range ensures the accuracy of the compound qualitative results. The secondary mass spectrometry employs a data-dependent acquisition mode. This mode works by first detecting all precursor ion signals in the primary mass spectrometer, then selecting several precursor ions with higher signal intensities according to a preset priority rule for fragmentation and acquiring their secondary mass spectra. In this embodiment, the secondary mass spectrometry scan range is set to 70 to 800 Da m / z, the resolution is set to 15000, and collision-induced dissociation uses a step-collision energy mode, with the three collision energies set to 20%, 40%, and 60%, respectively. The design principle of the step-collision energy mode is that different compounds exhibit different fragmentation behaviors at different collision energies. Low collision energies mainly lead to mild fragmentation such as neutral loss, while high collision energies cause the compound to fragment into smaller fragment ions. By performing fragmentation at three different collision energies, multi-level mass spectrometry information of the same compound at different fragmentation energies can be obtained, thereby achieving a more comprehensive characterization of the compound structure and reducing the probability of qualitative errors caused by incomplete fragmentation information at a single collision energy. Throughout the mass spectrometry data acquisition process, data is acquired simultaneously in both positive and negative ion modes. This setup allows for the acquisition of complete mass spectrometry information in both positive and negative ion modes with a single injection, significantly improving analytical efficiency and expanding the coverage of detectable compounds.

[0031] When collecting large batches of sample data, the injection order of samples is randomized, with a quality control sample interspersed between every 10 wine samples. This randomization design aims to eliminate systematic errors that may be introduced by the injection order. For example, instrument sensitivity may slowly drift as analysis time increases. Randomization distributes these time-related systematic errors throughout the dataset, allowing them to be effectively eliminated through normalization in subsequent data processing. Repeated analysis of the quality control samples is used to assess the instrument's stability throughout the data acquisition process. If significant deviations are observed in the quality control sample test results, the instrument's condition needs to be checked and necessary maintenance performed.

[0032] In data processing, the raw mass spectrometry data were acquired and managed by Xcalibur software, and then imported into Compound Discoverer 3.4 software for systematic data processing.

[0033] The data processing workflow includes the following key steps: First, the blank background subtraction operation. Since the matrix of the Maotai-flavor liquor sample is relatively complex, direct injection analysis will introduce a large amount of matrix background signal. The blank background subtraction operation compares the sample data with the blank solvent data to identify and subtract signals from background sources such as solvent, system pipeline and injection bottle, thereby improving the detection sensitivity of target compounds. The second step is peak detection. This operation uses the software's built-in peak recognition algorithm to search for signal peaks in the raw mass spectrometry data that meet the expected peak shape characteristics. It also sets an appropriate detection threshold based on parameters such as signal-to-noise ratio and peak width to ensure that the detected peak signals are real and reliable. Secondly, peak alignment is performed. During continuous analysis lasting tens of hours, the retention time of liquid chromatography may experience slight drift. Peak alignment compares and corrects the retention time offset of the same compound in different samples, ensuring that the retention time of the same compound in different samples corresponds to a unified time window, thereby ensuring the accuracy of subsequent quantitative analysis. Finally, there are gap filling and normalization. Gap filling is used to handle trace components that are not detected due to low signal intensity, and normalization is used to eliminate quantitative deviations caused by differences in injection volume or fluctuations in ionization efficiency between samples.

[0034] During the compound annotation stage, the structure of compounds is identified by comparing precise molecular weight, retention time, isotopic distribution, and secondary mass spectrometry fragment information with the mass spectrometry database. Precise molecular weight matching uses a mass precision standard with an error not exceeding 5 ppm. Matching within this error range ensures the accuracy of molecular formula deduction. The retention time of the target analyte deviates from the retention time parameter in the local mass spectrometry library by ≤0.2 min or ±2.5% (not exceeding 0.5 min) to increase the accuracy of compound retrieval. Isotopic distribution comparison utilizes the specific proportional relationships of naturally occurring isotopes in the elemental composition for verification. For example, compounds containing chlorine atoms will exhibit a characteristic M+2 isotope peak (intensity approximately 32% of the M peak), and compounds containing bromine atoms will exhibit an M+2 isotope peak (intensity approximately 98% of the M peak). These isotopic characteristics can effectively distinguish compounds with similar mass numbers but different elemental compositions. Secondary mass spectrometry fragment information comparison involves comparing the actually collected fragment ions with the theoretical or experimental fragment ions of standard compounds in the database under the same fragmentation conditions. Compound confirmation is based on the degree of matching of the fragment ions. The mass spectrometry databases used in this embodiment include the mzCloud database (which contains a vast amount of experimental mass spectrometry data of known compounds), the ChemSpider database (which integrates compound information from multiple chemical databases), the LipidSearch database (a database specifically for searching lipid compounds), and a locally built database (containing mass spectrometry data of baijiu-related compounds accumulated by our research team in previous work). This multi-database joint search design maximizes the coverage of various compounds that may be present in baijiu, improving the coverage of compound annotations.

[0035] After the above data processing, a complete dataset containing information such as compound name, retention time, precise molecular weight, and peak area is obtained. This dataset is exported as a CSV file and then imported into SIMCA 14.1 statistical software for subsequent multivariate statistical analysis.

[0036] In the OPLS-DA model building, an unsupervised principal component analysis (PCA) model is first constructed to conduct preliminary exploratory analysis of the dataset. PCA is an unsupervised data dimensionality reduction method that does not require pre-defined sample grouping information; it only seeks the principal components that can best explain the data variance based on the data's internal structure. PCA models can observe the distribution patterns of samples from different years in the principal component space. If samples from different years exhibit a certain clustering trend, it indicates the existence of systematic differences related to the year in the data. If all samples are mixed together and indistinguishable, it suggests that there may be no significant year-related features or that the data quality is problematic. Furthermore, PCA can be used to identify and remove outliers, which may be caused by operational errors during sample collection, storage, or analysis. These outliers can interfere with the subsequent construction and evaluation of supervised models. In this embodiment, the PCA results show that the eight different years of Maotai-flavor liquor samples exhibit a clear year-related clustering trend in the principal component space, and no serious outliers requiring removal were found, indicating that the overall quality of the sample data is good and it can proceed to the next step of supervised modeling analysis.

[0037] Building upon principal component analysis, an orthogonal partial least squares-discriminant analysis (OPLS-DA) model is further constructed. OPLS-DA is a supervised multidimensional data dimensionality reduction technique that utilizes the year grouping information of known samples as classification labels to establish a quantitative relationship between the data matrix and the classification labels. The core idea of ​​OPLS-DA is to decompose the data matrix into predictive components related to the classification labels and orthogonal components unrelated to the classification labels. The predictive components are used to explain the variation information related to the year grouping, while the orthogonal components are used to describe systematic variation unrelated to the year grouping (such as differences between different manufacturers, differences between analysis batches, etc.). By separating these two types of variation, OPLS-DA can more clearly reveal the contribution of year-related feature variables.

[0038] The reliability of the OPLS-DA model was verified using a permutation test. The permutation test involves randomly shuffling the year labels of the samples, then rebuilding the OPLS-DA model using the shuffled labels and calculating the corresponding model parameters. The parameters of these randomized models are then compared with those of the original model. If the original model truly captured the real year-related variations in the data, its predictive ability should significantly decrease after the labels were randomly shuffled. This is reflected in the calibration and validation coefficients of the model after the permutation test being significantly lower than those of the original model. In this embodiment, the permutation test was performed no fewer than two hundred times to ensure sufficient confidence in the statistical test results. The results of the permutation test show that the calibration and validation coefficients of the original OPLS-DA model are significantly higher than those of the randomized model, indicating that the model did not overfit and that the model parameters are realistic and reliable.

[0039] After the model was validated, differential compounds were screened based on the VIP values ​​of the OPLS-DA model. The projected values ​​of important variables reflect the contribution of each variable to the year classification; the larger the projected value of an important variable, the more significant the difference between samples from different years, and the greater its contribution to year differentiation. In this embodiment, VIP > 1 was used as the screening criterion, and after FDR correction, 40 differential compounds with year characteristics were finally screened. The chemical types of these 40 differential compounds cover multiple categories, including organic acids, higher fatty acids and their esters, amino acids and short peptides, and Maillard reaction intermediates, fully reflecting the molecular basis of the complex chemical transformation processes such as Maillard reaction, esterification reaction, and redox reaction during the long-term storage of Maotai-flavor liquor.

[0040] In PLS-R regression prediction module 6, the peak area data of 40 selected compounds with different characteristics based on year are used as independent variables, and the actual storage year of Maotai-flavor liquor is used as the dependent variable. A partial least squares regression algorithm is employed to construct a year identification model. Partial least squares regression is a regression analysis method in multivariate statistics specifically designed to handle quantitative relationships between multiple independent variables and a single dependent variable. It can still provide stable regression coefficient estimates even when multicollinearity exists among independent variables, making it particularly suitable for data analysis scenarios like metabolomics, where there are many variables and complex correlations between them.

[0041] During model construction, the 268 sample data points were randomly divided into training, validation, and test sets in a 70%:15%:15% ratio. The training set (188 samples) was used for parameter estimation and internal training; the validation set (40 samples) was used to evaluate the model's validation performance and optimize hyperparameters during training; and the test set (40 samples) was used for independent external evaluation of the model's predictive performance after training. The random partitioning process ensured a relatively balanced distribution of years among the training, validation, and test sets to avoid model bias caused by imbalanced sample distribution.

[0042] During the model training phase, cross-validation is used to optimize model parameters, with at least five iterations. Cross-validation works by further dividing the training set into multiple subsets. In each iteration, one subset is used as the validation set, while the remaining subsets are used as the training set for model training. The average of the validation results from multiple iterations is then used as the model's performance evaluation metric. Cross-validation effectively assesses the model's generalization ability to new samples, preventing overfitting of the training data. After optimization using cross-validation, the final PLS-R regression model is obtained.

[0043] The model's performance is comprehensively evaluated using four metrics: calibration root mean square error, validation root mean square error, calibration coefficients, and validation coefficients. Figure 2 As shown, the PLS-R year recognition model's predicted values ​​are fitted to the reference values. A two-dimensional coordinate system is constructed with the reference year as the x-axis and the model's predicted year as the y-axis. Blue scatter points represent training set samples, and red scatter points represent validation set samples. Data points are distributed along the diagonal of y=x. The table with a blue border in the upper left corner displays the model's key performance parameters, including a slope of 0.908, an intercept of 0.484, a root mean square error of 1.057 years, and a coefficient of determination of 0.90. The model's coefficient of determination of 0.90 indicates a high linear correlation between the predicted and actual year values, demonstrating excellent accuracy in the calibration model. The root mean square error of the model's cross-validation is 1.056 years. This error mainly stems from baseline differences between samples from different sources, indicating that there are indeed certain differences in the chemical composition of Maotai-flavor liquor produced by different companies. After verifying samples from the same source, the root mean square error of prediction was close to 0.5 years and the coefficient of determination was close to 1. This proves that after excluding the source difference factor, there is an extremely significant correlation between the content of 40 year-specific compounds and the storage year of the liquor, further verifying the effectiveness of the markers selected in this invention.

[0044] Hotelling T 2 The test and DModX methods were used to evaluate the model suitability of the test set samples. Hotelling T 2The test is used to determine whether the new sample falls within the normal distribution range of the training set samples in the multidimensional variable space. The DModX method is used to evaluate the residual distance between the new sample and the model calibration sample. The evaluation results show that all test set samples are within the applicable range of the model, and the model has good adaptability to the new sample.

[0045] In vintage prediction of liquor samples, for unknown vintage Maotai-flavor liquor samples, the samples are first processed and data is acquired according to the aforementioned sample pretreatment steps and ultra-high performance liquid chromatography-high resolution mass spectrometry data acquisition steps. Then, the peak area data of 40 vintage characteristic compounds in the sample are extracted. After data preprocessing such as normalization and standardization, the data is input into the constructed PLS-R vintage identification model. The model automatically calculates and outputs the predicted vintage value of the sample. At the same time, operators can further consult the detailed information of each marker in the vintage characteristic compound database as needed, including standard spectra, retention times, precise molecular weights, secondary mass spectrometry fragment information, and quantitative information, as a reference for rapid vintage comparison and identification.

[0046] Table 1. Age Prediction Results for Maotai-flavor Baijiu Table 1 lists the year prediction results for some test samples. As can be seen from the data in the table, among the 40 test samples, the prediction error of 97.1% of the samples is controlled within ±1 year. Only a few samples have slightly larger prediction errors (for example, one sample with a true year of 6 was predicted as 4.07 years). This deviation may be related to slight fluctuations in the source of raw materials or production process of the batch of samples. However, from a statistical point of view, the overall prediction error of the model is still within an acceptable range, indicating that the PLS-R year recognition model constructed in this invention has good prediction accuracy and generalization ability for the storage year of Maotai-flavor liquor.

[0047] Furthermore, to improve detection efficiency in practical applications, this embodiment also established a year-specific compound database after the year identification model was constructed. This database includes standard spectra, retention times, precise molecular weights, secondary mass spectrometry fragment information, and quantitative information for 40 core marker combinations from multiple manufacturers and different storage years of Maotai-flavor liquor samples, forming a complete standard reference database. In actual year identification, operators can not only use the PLS-R regression model for quantitative prediction but also compare the detection information of each marker in the sample with the information in the standard reference database to quickly determine whether the presence of each marker in the sample conforms to the chemical composition characteristics of Maotai-flavor liquor from a specific year, thus providing a dual guarantee of qualitative verification based on quantitative prediction.

[0048] In summary, this invention utilizes ultra-high performance liquid chromatography-high resolution mass spectrometry (UHPLC-HDMS) to comprehensively detect non-volatile components in Maotai-flavor liquor. Orthogonal partial least squares-discriminant analysis (PLLS-DI) is used to accurately screen compounds with age-specific differences. A high-precision age identification model is constructed based on PLS regression, enabling rapid, accurate, and objective identification of the storage age of Maotai-flavor liquor. The entire analytical process requires no complex sample pretreatment, is simple to operate, has a short detection cycle, and good repeatability. It provides a reliable technical means for the scientific identification of aged Maotai-flavor liquor and has significant application value for regulating the market order of aged liquor and protecting consumer rights.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any simple modifications, equivalent changes and alterations made by any person skilled in the art to the above embodiments based on the technical essence of the present invention without departing from the scope of the technical solution of the present invention shall still fall within the scope of the technical solution of the present invention.

Claims

1. A method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm, characterized in that, Includes the following steps: S1. Sample collection of Maotai-flavor liquor: Samples of Maotai-flavor liquor from multiple manufacturers with different storage years were collected. Multiple duplicate samples were set up for each year. Each sample was taken independently. The sample solution was analyzed by direct injection. S2. Data acquisition by ultra-high performance liquid chromatography-high resolution mass spectrometry: Non-targeted metabolomics data acquisition of sample solutions was performed using an ultra-high performance liquid chromatography-high resolution mass spectrometry system. S3. Data Processing and Compound Annotation: Import the raw mass spectrometry data into the data processing software for peak detection, peak alignment, gap filling and normalization. Based on the accurate molecular weight, retention time, isotope distribution and secondary mass spectrometry fragment information, compare with the mass spectrometry database to annotate the compounds. S4. Screening of year-related differential compounds: Multivariate statistical analysis was used to screen differential compounds related to the storage year. A principal component analysis model was constructed to remove outliers. An orthogonal partial least squares-discriminant analysis model was constructed. Variable projection importance (VIP) was used for screening. After FDR correction, year-characteristic differential compounds were obtained. S5. Construction of the year identification model: The peak area of ​​the compounds with the year-specific differences is used as the input variable, and the partial least squares regression algorithm is used to construct the year identification model of Maotai-flavor liquor. S6. Identification of the year of the liquor sample to be tested: Data is collected from the liquor sample to be tested, the peak area of ​​the characteristic difference compounds of the year is extracted, and after normalization and standardization, it is input into the liquor year identification model to output the predicted year of the sample to be tested.

2. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 1, characterized in that, In step S2, an HSS T3 column is used for chromatographic separation. The column has an inner diameter of 2.1 mm, a length of 100 mm, a particle size of 1.8 μm, a column temperature of 40 °C, a mobile phase A of 0.1% formic acid aqueous solution, a mobile phase B of methanol, a flow rate of 0.3 mL per minute, and an injection volume of 1 μL.

3. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 2, characterized in that, In step S2, the mobile phase gradient elution program is set as follows: for the 0-1 minute period, the volume ratio of mobile phase B is 2%; for the 1-15 minute period, the volume ratio of mobile phase B increases linearly from 2% to 98%; for the 15-17 minute period, the volume ratio of mobile phase B remains at 98%; for the 17-17.1 minute period, the volume ratio of mobile phase B decreases from 98% to 2%; and for the 17.1-20 minute period, the volume ratio of mobile phase B remains at 2%.

4. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 2, characterized in that, The stationary phase of the HSS T3 column is trifunctional bonded silica gel packing.

5. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 1, characterized in that, In step S2, an electrospray ionization source is used as the ionization method, and data is collected simultaneously in positive ion mode and negative ion mode. The first-stage mass spectrometry scan range is set to 70-800 Da and the resolution is set to 120,000. The second-stage scan adopts a data-dependent acquisition mode with a resolution of 15,000. The collision-induced dissociation adopts a step collision energy mode with collision energies set to 20%, 40%, and 60%, respectively.

6. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 5, characterized in that, The spray voltage for positive ion mode was set to 3.5kV, and the spray voltage for negative ion mode was set to 2.5kV; the temperature of the ion transmission tube was set to 320℃, the sheath gas flow rate was set to 50 Arb, the auxiliary gas flow rate was set to 10 Arb, and the purge gas flow rate was set to 1 Arb.

7. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 1, characterized in that, In step S3, the mass spectrometry database includes the mzCloud database, ChemSpider database, LipidSearch database, and a locally built database, and the error range of the accurate molecular weight is controlled within 5 ppm.

8. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 1, characterized in that, In step S4, the reliability of the orthogonal partial least squares-discriminant analysis model is verified by permutation test, with no less than 200 permutation tests; differential compounds are screened after FDR correction using VIP>1 as the screening criterion.

9. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 8, characterized in that, The number of compounds with year-specific differences obtained by the aforementioned joint screening criteria was 40.

10. The method for identifying the vintage of Maotai-flavor liquor based on HRMS combined with PLS-R algorithm as described in claim 1, characterized in that, In step S5, the sample data is randomly divided into training set, validation set and test set according to the proportions of 70%, 15% and 15%, respectively. The model is trained using the training set data, and the model parameters are optimized using cross-validation, with no less than 5 iterations of cross-validation. The model is validated using the validation set data, and the predictive performance of the model is independently evaluated using the test set data.

Images