Marker combination for raw material identification of white tea, quality fingerprint evaluation model and application thereof

Abstract

The application relates to the technical field of tea detection and quality evaluation, and discloses a marker combination for white tea raw material identification, a quality fingerprint evaluation model and application of the marker combination and the quality fingerprint evaluation model. The combination comprises kaempferol-3 O -galactoside, quercetin-3 O -glucoside, rutin and myricetin-3 O -galactoside, kaempferol, quercetin, myricetin, procyanidin B2, citric acid and quinic acid. After centrifugation and grading dilution, the UPLC-MS / MS-MRM method with a single needle analysis time of 5 min is used for detection, the original sample concentration is obtained by using an external standard method, a 10-index quality fingerprint model is constructed, and a matching relationship with a preset reference fingerprint is obtained. In the examples, the combined out-of-balance accuracy and the macro-average F1 value of the four types of reference fingerprint models are both 98.15%. The application can be used for quality comparison, reference matching and extraction process tracking; after matrix adaptation and model verification, the application can be used for raw material source matching of beverage production intermediate liquid.

Description

Technical Field

[0001] This invention relates to the field of tea testing and quality evaluation technology, specifically to the combination of markers for identifying white tea raw materials, a quality fingerprint evaluation model, and their applications. Background Technology

[0002] White tea is a lightly fermented tea, and its quality is closely related to the variety of fresh leaves, the grade of harvest, the withering and drying process, and the transformation during storage. For beverage development, the evaluation of white tea raw materials should not be limited to traditional appearance, liquor color, aroma, taste, and infused leaves. Further attention should be paid to non-volatile components in the tea liquor that can affect bitterness, astringency, acidity, body, aftertaste harmony, and stability. Flavonol glycosides are generally associated with bitterness, astringency, and aftertaste structure; flavonol aglycones can reflect the flavonoid skeleton level in the raw material and certain processing and transformation characteristics; organic acids such as citric acid and quinic acid are related to acidity, freshness, and overall taste balance; and dimeric catechins such as proanthocyanidins B2 are related to astringency structure, body, and the risk of aftertaste bitterness.

[0003] Currently, companies routinely evaluate white tea raw materials, extracts, and intermediate liquids used in beverage production primarily through sensory evaluation and conventional physicochemical indicators such as tea polyphenols, amino acids, and caffeine. Sensory results are easily influenced by the experience of the evaluators and the condition of the samples. Furthermore, conventional indicators are often total or single-category indicators, making it difficult to further break down the contributions of specific molecules such as flavonol glycosides, flavonol aglycones, organic acids, and proanthocyanidins to sample differences. When comparing white tea samples from different raw materials, different white tea product categories, different extraction conditions, or different production stages, relying solely on the above methods makes it difficult to form a unified and verifiable molecular quality fingerprint, and also makes it difficult to extend raw material information to subsequent sample process tracking and source matching.

[0004] If existing decentralized detection pathways are adopted, different categories of substances often require separate pretreatment, separate method establishment, or batch injection. Due to the significant differences in polarity, retention behavior, and ionization response among flavonol glycosides, organic acids, proanthocyanidins, and flavonoid aglycones, decentralized methods will lead to longer detection times, increased sample usage, and amplified cross-batch errors. This is detrimental to cross-sectional comparison and continuous tracking of white tea raw materials, extracts, beverage base liquids, and production intermediate liquids under a unified detection standard.

[0005] Therefore, existing technologies still lack a method to rapidly quantify key quality components such as flavonol glycosides, flavonol aglycones, proanthocyanidins, and organic acids under uniform sample preparation and detection standards, establish comparable quality fingerprints for white tea, and achieve quality evaluation, reference category matching, extraction process tracking, and raw material source matching within the model's applicable domain based on a pre-set reference quality fingerprint database. Summary of the Invention

[0006] In view of this, the technical problem to be solved by the present invention is to provide a combination of markers for the identification of white tea raw materials, a quality fingerprint evaluation model and its application. Through standardized sample preparation, simplified pre-centrifugation treatment, graded dilution, UPLC-MS / MS-MRM detection with a single-needle analysis time of 5 min, external standard quantification and index matrix processing, the same method and the same caliber are used to detect 10 quality markers across chemical categories. A reference quality fingerprint database is established and the evaluation model is trained, so that white tea samples can be compared in quality, referenced and matched and the extraction process can be tracked. After completing the corresponding matrix adaptation and model verification, the source of raw materials can be matched.

[0007] This invention provides a combination of markers for identifying white tea raw materials, including kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O -Galactoside, Kaempferol, Quercetin, Myricetin, Proanthocyanidins B2, Citric acid and Quinic acid.

[0008] In some embodiments, the 10 markers cover flavonol glycosides, flavonol aglycones, proanthocyanidins, and organic acids, and can characterize the differences in molecular composition of white tea samples from multiple chemical dimensions such as glycosylation degree, flavonoid skeleton, proanthocyanidin structure, and organic acid composition.

[0009] In some embodiments, the biomarker combination is obtained by comprehensively screening candidate differential metabolites based on differential trend, variable importance, standard availability, MRM response stability, chromatographic retention behavior, and 5-minute detection feasibility.

[0010] Under the same cross-validation process grouped by process conditions, the combined out-of-balance accuracy of the individual marker models ranged from 32.41% to 50.00%, while the accuracy of the 10-marker combination model of this invention was 98.15%. Furthermore, the normalized model weights of all 10 markers in the outer training layer were non-zero. These results indicate that the technical effectiveness of this invention stems from the combined utilization of multidimensional molecular information by this specific cross-category combination, rather than from any single marker.

[0011] The present invention also provides the application of the aforementioned marker combination in constructing a fingerprint database of white tea raw materials or a fingerprint evaluation model for the quality of white tea raw materials.

[0012] Compared to using a single flavor marker, a single chemical category marker, or other combinations of markers with varying numbers, the 10 cross-category flavor quality marker combinations described in this invention take into account information from different chemical categories such as flavonol glycosides, flavonol aglycones, proanthocyanidins, and organic acids, thus more comprehensively reflecting the molecular characteristic differences between white tea varieties. The quality fingerprint spectrum constructed based on these marker combinations exhibits higher category discrimination ability and model stability, improving the classification performance of white tea raw material identification models and the accuracy of determining the variety of white tea samples.

[0013] Experimental results show that the model classification performance obtained by the 10 marker combinations provided by this invention under cross-validation conditions is better than the model constructed using a single marker or other marker combinations.

[0014] In some embodiments, the applications include quality comparison of the test samples, reference category matching, sample consistency evaluation, process tracking, or raw material source matching.

[0015] In some embodiments, the white tea raw material quality fingerprint evaluation model can be defined based on the white tea raw material variety, white tea product category, known raw material source, processing stage, or a combination thereof; the white tea raw material in this embodiment includes at least one of Zhenghe Dabai White Peony, Zhenghe Dabai Shoumei, Fuding Dahao White Peony, and Fuding Dahao Shoumei.

[0016] It should be noted that the specific evaluation task of the model is determined by the label definition of the reference quality fingerprint category. A variety identification model can be constructed when the raw material variety is used as the label; a product category determination model can be constructed when the white tea product category is used as the label; and a joint reference fingerprint matching model can be constructed when a combination of labels—raw material variety, white tea product category, or a known source—is used. Different evaluation tasks should be trained, have their model parameters selected, and be validated separately; the deployment coefficients of the four types of joint models in this embodiment cannot be directly applied.

[0017] This invention provides a method for detecting the aforementioned combination of markers, comprising the following steps: Step 1: Take the extract of the white tea sample to be tested, and use UPLC-MS / MS multiple reaction monitoring mode to simultaneously detect it in the same chromatographic-mass spectrometry method and the same detection sequence in one injection to obtain the peak area of ​​each marker; Step 2: Establish standard curves for each marker using the external standard method, and calculate the concentration of each marker based on its peak area.

[0018] In some embodiments, the synchronous detection time in step 1 is less than or equal to 5 minutes. In actual detection, the single-needle analysis time of the UPLC-MS / MS method is less than or equal to 5 minutes; for white tea quality markers whose concentrations in the same dilution fall within the corresponding standard curve range, they are detected synchronously in a single injection.

[0019] In some embodiments, the UPLC-MS / MS multiple reaction monitoring mode includes chromatographic separation using a reversed-phase column, with a mobile phase A of 0.1% formic acid in methanol / water = 2 / 98 (V / V) and a mobile phase B of acetonitrile, employing a gradient elution program, and simultaneously detecting kaempferol-3- in the same chromatographic-mass spectrometry method and the same detection sequence under a multiple reaction monitoring mode with switching between positive and negative ions from an electrospray ionization source. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O -Galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid and quinic acid, and the peak areas of each marker were obtained.

[0020] In some embodiments, the reversed-phase column is an ACQUITY BEH T3 column (2.1 mm × 100 mm, 1.7 μm or a reversed-phase column with equivalent separation performance).

[0021] In some embodiments, the chromatographic conditions include: a flow rate of 0.4 mL / min, a column temperature of 40 °C, and an injection volume of 1 μL; The gradient elution procedure includes: From 0 to 1 min, mobile phase A was 98% and mobile phase B was 2%. Within 1-3 minutes, mobile phase A decreased from 98% to 5%, while mobile phase B increased from 2% to 95%. 3-4 min, mobile phase A is 5%, mobile phase B is 95%; 4.01~5 min, the mobile phase A was restored to 98% and the mobile phase B to 2%.

[0022] In some embodiments, the tandem mass spectrometry employs a triple quadrupole tandem mass spectrometer equipped with an electrospray ionization source, and uses positive and negative ion switching modes and multiple reaction monitoring modes for detection.

[0023] In some embodiments, the extract of the white tea sample to be tested is centrifuged to obtain a supernatant, and the supernatant is then graded and diluted according to the concentration range of each target analyte.

[0024] In some embodiments, step 2, the calculation further includes obtaining the content based on the peak area of ​​each marker, and obtaining the concentration of each marker based on the content multiplied by the dilution factor.

[0025] This invention provides a method for constructing a fingerprint evaluation model for the quality of white tea raw materials, comprising the following steps: Step 1: Obtain different white tea raw materials and prepare extracts; Step 2: Detect kaempferol-3- in the extract from Step 1 using the detection method described above. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - Concentrations of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid; Step 3: Perform logarithmic transformation and standardization on the concentration data obtained in Step 2 to obtain the quality fingerprint feature vector of the extract, and form a quality fingerprint spectrum based on different extracts; Step 4: Establish a database based on the quality fingerprint spectra of different extracts obtained in Step 3, and use the database to train and obtain a quality fingerprint evaluation model for white tea raw materials.

[0026] In some embodiments, in step 1, the different white tea raw materials include at least one of Zhenghe Da Bai Bai Mu Dan, Zhenghe Da Bai Shou Mei, Fuding Da Hao Bai Mu Dan, and Fuding Da Hao Shou Mei; The preparation method includes mixing white tea raw materials with water, and extracting the extract to obtain an extract, wherein: The volume-to-mass ratio of the white tea raw material to water is (30~50) mL:1 g, the extraction temperature is 60~80℃, the extraction time is 10~30 min, and the ultrasonic power is 100~500W.

[0027] In some embodiments, the volume-to-mass ratio of the white tea raw material to water is 30mL:1g, 40mL:1g, or 50mL:1g, the extraction temperature is 60°C, 70°C, or 80°C, the extraction time is 10 min, 20 min, or 30 min, and the ultrasonic power is 100W, 300W, or 500W.

[0028] In some embodiments, the data preprocessing includes performing a log10 transformation on the effective positive concentration and standardizing it using the mean and standard deviation of the training data.

[0029] The quality fingerprint comprises feature vectors of 10 markers corresponding to each sample, processed using standardized data, and is used to establish a reference database in the form of a "sample × index" matrix. The quality fingerprint is used for standardized digital characterization of white tea samples and serves as input data for quality comparison, model training, reference matching, process tracking, and relative raw material source matching or production chain tracking within the model's applicable domain.

[0030] In some embodiments, for target substances whose concentrations fall within the linear range in the same dilution, they can be detected simultaneously in a single injection; sample solutions with different dilution ratios are detected in the same detection sequence using the same single-needle analysis method with a time of 5 minutes.

[0031] In some embodiments, the single-needle analysis time is less than or equal to 5 minutes.

[0032] In some embodiments, in step 4, the training method includes at least one of partial least squares discriminant analysis, orthogonal partial least squares discriminant analysis, support vector machine, random forest, and artificial neural network.

[0033] In some embodiments, partial least squares discriminant analysis is preferably used in step 4; under the condition of using the same 10 index inputs and completing the corresponding training and verification, orthogonal partial least squares discriminant analysis, support vector machine, random forest or artificial neural network may also be used.

[0034] In some embodiments, the database includes quality fingerprint feature vectors corresponding to multiple preset reference quality fingerprint categories under uniform sample preparation conditions, different extraction conditions, or different production stages.

[0035] In some embodiments, the model is grouped and cross-validated according to process condition number, so that different categories of samples under the same process condition can enter the training or validation cycle at the same time.

[0036] In some embodiments, the model is used for quality comparison, reference matching, sample consistency evaluation, extraction process tracking, production process monitoring, or relative raw material source matching of white tea samples consistent with a reference quality fingerprint system, sampling method, and matrix type.

[0037] This invention provides a method for identifying white tea varieties, comprising the following steps: S1. Take a sample of white tea to be tested and use the detection method described above to detect kaempferol-3- in the sample. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - Concentrations of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid; S2, logarithmically transform and standardize the concentration data obtained in S1 to obtain the quality fingerprint feature vector of the white tea sample to be tested, and form a quality fingerprint spectrum. S3. Input the quality fingerprint spectrum described in S2 into the white tea raw material quality fingerprint evaluation model constructed by the construction method to obtain the matching relationship between the white tea sample to be tested and the white tea raw material preset in the white tea raw material quality fingerprint evaluation model, and determine the variety of the white tea sample to be tested based on the matching relationship.

[0038] In some embodiments, in S1, the white tea sample to be tested includes at least one of the following: white tea standard brewing liquid, white tea extract, white tea concentrate, white tea beverage base liquid, white tea beverage production intermediate liquid, and white tea products adapted to the matrix. In step S3, the matching relationship includes at least one of relative matching score, maximum matching category, quality fingerprint difference, or consistency result.

[0039] In some embodiments, the white tea sample to be tested is subjected to the same data processing method as in the model building stage to obtain a quality fingerprint spectrum.

[0040] In some embodiments, the model output results include the relative matching score between the test sample and each preset reference quality fingerprint category, the maximum matching category, and the quality fingerprint difference or consistency results; when the reference quality fingerprint category corresponds to a known raw material source, the results are used for matching or tracing the raw material source of the test sample.

[0041] This invention provides the application of the aforementioned marker combination, the aforementioned detection method, the aforementioned construction method to construct a white tea raw material quality fingerprint evaluation model, or the aforementioned identification method in white tea quality analysis.

[0042] In some embodiments, the white tea quality analysis includes white tea raw material identification and / or screening, batch consistency evaluation of white tea raw materials or their products, monitoring of the production process of white tea raw materials or their products, matching of white tea raw material sources, and white tea beverage research and development.

[0043] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention uses kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - A specific combination of 10 cross-chemical quality markers, consisting of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid, was established. Under the same cross-validation process grouped by process conditions, the combined out-of-balance accuracy of the 10-marker combination model was 98.15%, higher than the 32.41%–50.00% of the individual marker models, indicating that the combination can form a molecular quality fingerprint with comprehensive discriminative power.

[0044] 2. This invention constructs a quality fingerprint feature vector and establishes a reference quality fingerprint database by quantitatively detecting 10 markers in white tea samples. Based on this, a quality fingerprint evaluation model is trained to output the matching degree, difference, or consistency results between the tested sample and the reference fingerprint, providing a unified, quantitative, and verifiable quality fingerprint evaluation basis for white tea raw materials and extracts. When extended to beverage base liquids and intermediate production liquids, corresponding matrix adaptation and model verification should be completed.

[0045] 3. This invention uses UPLC-MS / MS-MRM mode to achieve simultaneous detection of 10 different chemical markers in a single injection, and combines external standard method for quantitative analysis. This avoids the problems of long detection cycle and large cross-batch error caused by building methods and detecting different types of substances separately. It realizes rapid and simultaneous detection of multiple key flavor markers in white tea samples, and improves detection efficiency and data consistency.

[0046] 4. This invention employs grouped cross-validation based on process condition numbers. The combined out-of-division balance accuracy and macro-average F1 score of the fixed 4 latent variable model are both 98.15%, and the lowest single-fold balance accuracy is 95.00%. Further sensitivity analysis based on unique process parameter signatures yields a result of 99.07%. These results demonstrate that the quality fingerprint evaluation model still exhibits good reference fingerprint matching stability under extraction process conditions that were not used for training.

[0047] 5. When the reference quality fingerprint database is established from samples of known raw material sources and different processing stages, and subsequent samples complete the corresponding matrix adaptation and model verification, this invention can extend the molecular fingerprint of the raw material to the extract, beverage base liquid or production intermediate liquid for relative raw material source matching and production chain tracking; for finished products with complex composition or significantly exceeding the coverage of training data, no forced matching conclusion is output. Attached Figure Description

[0048] Figure 1 The chemical structural formulas of 10 flavor quality markers are shown, wherein formulas 1 to 10 are kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O -Galactoside, Kaempferol, Quercetin, Myricetin, Proanthocyanidins B2, Citric acid and Quinic acid; Figure 2 Representative quantitative ion chromatograms of 10 flavor quality markers under the detection conditions of this invention's UPLC-MS / MS-MRM are shown. Figure 3 The 10-index reference quality fingerprint is formed after the standard brewing liquid of four representative white teas is transformed and standardized by log10; Figure 4 The average quality fingerprints of 10 indicators for four representative white tea extracts under 27 experimental conditions are shown. The fingerprints are used to evaluate the retention of reference molecular characteristics under different process conditions. The 27 experimental conditions correspond to 25 unique combinations of process parameters. Figure 5 The figure shows the score distribution of the first two latent variables of the final determined four-latent variable white tea quality fingerprint evaluation model on all 108 white tea extract samples. ZHDB_WP represents Zhenghe Da Bai Bai Mudan, ZHDB_SM represents Zhenghe Da Bai Shoumei, FDDH_WP represents Fuding Dahao Bai Mudan, and FDDH_SM represents Fuding Dahao Shoumei. These four categories serve as the four preset reference quality fingerprints in this embodiment. Circles represent Bai Mudan, and squares represent Shoumei. This figure is used to show the overall distribution of samples in the first two latent variable spaces and is not used as an independent evaluation basis for the model's out-of-range predictive performance. Figure 6 The results show a comparison of the combined out-of-fold balance accuracy (BA) of each single marker model with the final determined white tea quality fingerprint evaluation model with 4 latent variables and 10 indicators under the same 5-fold cross-validation process grouped by process number. Figure 7 The final determined white tea quality fingerprint evaluation model with 4 latent variables and 10 indicators is shown in the combined out-of-fold confusion matrix of the 5-fold cross-validation grouped by process number; Figure 8 The mean and standard deviation of the normalized model coefficients weights of 10 markers in the 5 outer layer training compromises of the final determined 4 latent variable white tea quality fingerprint evaluation model are shown. Figure 9 The mean and standard deviation of the 5-fold equilibrium accuracy of each white tea quality fingerprint evaluation model under the same outer layer grouping when the number of latent variables is 1 to 10 are shown, and the final determined 4-latest variable model is marked. Detailed Implementation

[0049] This invention provides a combination of markers and their application in the identification of white tea varieties. Those skilled in the art can refer to this document and appropriately modify the process parameters to achieve the desired results. It should be particularly noted that all similar substitutions and modifications are obvious to those skilled in the art and are considered to be included in this invention. The methods and applications of this invention have been described through preferred embodiments. Those skilled in the art can clearly modify or appropriately change and combine the methods and applications described herein without departing from the content, spirit, and scope of this invention to realize and apply the technology of this invention.

[0050] Terminology Explanation: 1) Quality fingerprint evaluation: refers to forming a molecular quality fingerprint based on the quantitative composition of 10 quality markers in white tea samples, and evaluating the degree of matching, difference or consistency between the sample to be tested and one or more preset reference quality fingerprints; the evaluation is not the same as continuous sensory quality score.

[0051] 2) Reference Quality Fingerprint Category: This refers to a predefined reference category based on the variety of white tea raw materials, the type of white tea product, the known source of raw materials, the processing stage, or a combination thereof. In this embodiment, the four reference quality fingerprints use a combined label representing the combination of raw materials and the type of white tea product, namely Zhenghe Da Bai Bai Mudan, Zhenghe Da Bai Shoumei, Fuding Da Hao Bai Mudan, and Fuding Da Hao Shoumei.

[0052] 3) White tea sample liquid to be tested: refers to the liquid sample formed by brewing, soaking, extracting, blending, concentrating or diluting white tea raw materials, including but not limited to white tea standard brewing liquid, white tea extract, white tea beverage base liquid, concentrated white tea soup, white tea beverage production intermediate liquid and its dilution.

[0053] 4) White tea standard brewing liquid: refers to the white tea raw material brewing liquid prepared in accordance with the national standard tea sensory evaluation method (GB / T 23776) or the enterprise's internal equivalent evaluation method, under the conditions of uniform tea sample quantity, uniform boiling water volume, uniform brewing time and uniform filtration method, to eliminate the interference of differences in brewing conditions on the comparison of raw materials.

[0054] 5) Rapid synchronous quantitative detection: refers to the use of a unified centrifugation pretreatment and a single-needle analysis time of 5 min for UPLC-MS / MS-MRM method to perform necessary fractional dilutions of the same test sample solution according to the concentration range of the target analyte, and to complete the quantification of multiple target analytes in the same batch, using the same method and the same caliber in the same detection sequence; it does not mean that all original samples only need to be injected once.

[0055] 6) UPLC-MS / MS: Ultra-high performance liquid chromatography-tandem mass spectrometry, suitable for highly selective qualitative and quantitative analysis of trace or micro-scale target compounds in complex food matrices.

[0056] 7) MRM: Multiple Reaction Monitoring mode, which selectively detects parent ion-daughter ion pairs, improving the specificity of target identification and quantification in complex tea liquor matrices.

[0057] 8) Ten flavor quality markers: including flavonol glycosides (kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O-Galactosides, flavonol aglycones (kaempferol, quercetin, myricetin), proanthocyanidins (proanthocyanidin B2), and organic acids (citric acid, quinic acid).

[0058] 9) Quality fingerprint matrix: refers to the "sample × index" matrix formed by the quantitative results of 10 target substances of one or more white tea sample liquids to be tested according to the unified unit, sample number and data structure, and logarithmic transformation, standardization, reference matching, clustering or discriminant analysis is performed according to the specific evaluation task.

[0059] 10) Relative raw material source matching or production chain tracing: This refers to comparing the quality fingerprints of the white tea extract to be tested, the matrix-adapted beverage base liquid, or the intermediate production liquid with a reference quality fingerprint database established from known raw material sources and verified by corresponding samples to obtain a relative source matching relationship. Its applicable scope is limited by the reference database and the applicable domain of the model, and is not equivalent to general geographical origin tracing that is not limited by the sample range.

[0060] The reagents, standards, and raw material samples used in this invention are all commercially available. The invention is further illustrated below with reference to specific examples.

[0061] Example 1: Basis for determining the combination of 10 specific white tea quality fingerprint markers

[0062] This embodiment illustrates the technical source and determination basis of a specific combination of 10 white tea quality fingerprint biomarkers. These 10 biomarkers originated from a candidate component library established by the applicant through previous non-targeted metabolomics research on white tea. Based on this candidate library, and comprehensively considering the significance of quality characterization, raw material source and trends in white tea product category differences, chemical category complementarity, availability of standards, MRM response stability, chromatographic retention behavior, and feasibility of 5-minute targeted detection, the aforementioned 10 biomarker combinations were ultimately determined. This screening was not based solely on a single statistical indicator, but rather to obtain a fixed cross-category indicator system that can be used for white tea quality fingerprint library construction, evaluation, and reference matching.

[0063] When determining the above 10 biomarker combinations, information such as the importance of projected reference variables, inter-group difference trends, changes in product grade and maturity, differences in representative raw material combinations, quality correlation, and inter-sample stability were considered. Previous studies have shown that components such as flavonol glycosides, organic acids, proanthocyanidins, and flavonol aglycones are associated with white tea grade, maturity, representative raw material combinations, and taste differences.

[0064] Based on the aforementioned candidate components, this invention further refines the technical scope by considering the practical feasibility of targeted detection: the target analyte should have available standards, identifiable chromatographic retention behavior, stable response in the corresponding positive or negative ion MRM mode, a linear range suitable for the concentration levels of white tea standard brewing liquid and extract, and the ability to complete batch detection using the same 5-min short gradient method. Considering the above criteria, chemical category coverage, and methodological feasibility, kaempferol-3- was ultimately determined. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - A combination of 10 specific flavor quality markers consisting of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid.

[0065] This invention does not achieve its technical effect solely by increasing the number of detection indicators. Comparative Example 1 compared a model with 10 single-marker models and a specific 10-indicator quality fingerprint evaluation model under identical data preprocessing, grouping, and prediction rules. The combined out-of-bounds balance accuracy of each single marker was only 32.41%–50.00%, while the specific combination reached 98.15%. Furthermore, the normalized weights of the 10 markers in the 5 outer-layer training tradeoffs were all non-zero and exhibited high consistency in ranking, indicating that the model actually utilized complementary information from multiple chemical categories.

[0066] Example 2: Method for Rapid Detection of 10 Quality Markers and Construction of Quality Fingerprint Evaluation Model

[0067] 1. Preparation of Standard White Tea Brewing Solution

[0068] White tea raw material samples covering core production areas, representative varieties, and typical white tea product categories were collected. The samples included representative raw materials from the Zhenghe core production area: Zhenghe Da Bai Bai Mu Dan and Zhenghe Da Bai Shou Mei; and representative raw materials from the Fuding core production area: Fuding Da Hao Bai Mu Dan and Fuding Da Hao Shou Mei. These four categories served as representative white tea samples for the reference quality fingerprint category in this embodiment. Each sample was brewed according to the national standard sensory evaluation method for tea, ensuring uniform tea sample quality, boiling water volume, brewing time, brewing container, and filtration method to obtain a standard white tea brewing solution.

[0069] The establishment of standard brewing solutions serves two technical purposes: first, it standardizes the raw material evaluation conditions upfront, ensuring that differences between samples primarily stem from the raw materials themselves; second, it establishes a raw material-level reference quality fingerprint, serving as a benchmark for subsequent quality comparison and matching of extracts, and providing a methodological basis for the application of beverage base liquids or intermediate production liquids after completing appropriate matrix adaptation and model validation. Standard brewing solutions can also be used in parallel with traditional sensory evaluation.

[0070] 2. Sample pretreatment

[0071] Take 1.5 mL of the standard white tea brewing solution and place it in a 2 mL centrifuge tube. Centrifuge at 10000 r / min for 3 min, and collect 1.0 mL of the supernatant. Based on the actual concentration range of the samples in this embodiment, fractional dilution was used before injection: In the standard brewing solution shown in Table 6, kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - Galactoside and proanthocyanidin B2 were diluted 2-fold, citric acid was diluted 50-fold, and quinic acid was diluted 100-fold. Kaempferol, quercetin, and myricetin could be injected directly. In the extracts shown in Table 8, the aforementioned four flavonol glycosides and proanthocyanidin B2 were diluted 5-fold, and citric acid and quinic acid were diluted 100-200-fold depending on the concentration of the process sample. The three flavonol aglycones could be injected directly. If individual samples still exceeded the linear range of 1-2500 μg / L after dilution, they were further diluted by a fixed factor, and the result was multiplied by the total dilution factor in the calculation. This pretreatment does not use solid-phase extraction, liquid-liquid extraction, or complex purification steps, which can reduce sample loss and operational errors.

[0072] In a preferred embodiment, standards or quality control samples can be inserted into the sample detection sequence for system suitability monitoring. For example, target compounds with stable responses, good peak shapes, and stable retention times, such as rutin, can be selected as representative quality control indicators. The instrument status can be determined by changes in retention time, peak area, response factor, or calculated concentration. This step is not the only limiting condition of the present invention, but it can enhance the verifiability of the long-sequence sample detection process.

[0073] 3. Chromatographic separation conditions

[0074] Reversed-phase chromatography was preferred for separation, such as an ACQUITY BEH T3 column (2.1 mm × 100 mm, 1.7 μm) or an equivalent column with comparable separation performance. Mobile phase A was a 2 / 98 (V / V) solution of 0.1% formic acid in methanol / water, and mobile phase B was acetonitrile; the flow rate was 0.4 mL / min; the column temperature was 40 ℃; and the injection volume was 1 μL. The gradient elution program is shown in Table 1, and the single-injection analysis time was 5 min.

[0075] Table 1 Gradient elution program for high-throughput detection in UPLC-MS / MS

[0076] This short gradient design can cover target compounds with significant retention differences, such as polar organic acids, flavonol glycosides, proanthocyanidins, and flavonol aglycones, within 5 minutes. Compared with long gradient qualitative methods, this method is more suitable for batch testing, quality fingerprint library construction, and cross-sectional comparison of white tea standard brewing liquid and extract. When used for beverage base liquid or production intermediate liquid, corresponding matrix adaptation and methodological validation should be completed.

[0077] 4. Mass spectrometry detection conditions and target analyte combinations

[0078] Detection was performed using a triple quadrupole tandem mass spectrometer equipped with an electrospray ionization source, employing positive and negative ion switching modes and multiple reaction monitoring (MRM) mode. Among these, kaempferol-3- O -Galactoside, Quercetin-3- O - Glucosides and rutin are used in a positive ion mode; Myricetin-3- O -Galactosides, kaempferol, quercetin, myricetin, proanthocyanidins B2, citric acid, and quinic acid are detected using negative ion mode. Ion source parameters can be set as follows: capillary voltage 3.0 kV, ion source temperature 150 ℃, cone backflush gas flow rate 50 L / h, desolvation gas temperature 500 ℃, and desolvation gas flow rate 900 L / h. Depending on the instrument model and the response of the target analyte, the ion source parameters, cone voltage, collision energy, and residence time can be equivalently adjusted without altering the fundamental nature of this invention.

[0079] This invention selects 10 flavor quality markers as the target combination, specifically including kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O -Galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid. These target compounds cover flavonol glycosides, flavonol aglycones, proanthocyanidins, and organic acids. Their molecular composition characteristics related to the flavor and quality of white tea can be characterized from multiple chemical dimensions, including the flavonoid skeleton, degree of glycosylation, proanthocyanidin polymerization structure, and organic acid composition. Basic information on the target compounds is shown in Table 2, MRM parameters are shown in Table 3, and chemical structural formulas are shown in [Table 3]. Figure 1 .

[0080] Representative quantitative ion chromatograms of 10 flavor quality markers under the UPLC-MS / MS-MRM conditions of this invention are shown below. Figure 2 As shown. By Figure 2As can be seen, within the 5-minute analysis period, each target analyte can obtain a clear response signal in the corresponding positive or negative ion MRM mode, with retention times mainly ranging from approximately 0.66 to 2.94 minutes. The quantitative specificity of this method mainly stems from the selective monitoring of the precursor ion-daughter ion pair in MRM mode, rather than simply relying on complete baseline separation of chromatographic peaks. For target analytes whose concentrations fall within the linear range in the same dilution, they can be detected simultaneously in a single injection; for target analytes with significantly different concentration magnitudes, they are injected separately into the same detection sequence after fractional dilution as described in Part 2 of this embodiment, and quantification is completed using the same 5-minute chromatographic-MRM method.

[0081] Table 2 Basic Information of 10 Flavor Quality Markers

[0082] Table 3. UPLC-MS / MS MRM parameters of 10 flavor quality markers

[0083] Note: Represents quantitative ion pairs

[0084] 5. Establishment of standard curve and quantification using external standard method

[0085] Ten target analyte standard solutions were prepared, and an external standard regression equation was established between peak area and concentration, as shown in Table 4. Samples were diluted according to the dilution ratios described in Part 2 of this embodiment before injection. The target analyte concentration was calculated by substituting the peak area of ​​the corresponding quantitative ion pair into the standard curve and multiplying it by the corresponding total dilution ratio to obtain the original sample concentration. For target analytes that still exceeded the upper limit of linearity after dilution, further dilutions were performed and re-detected until the response fell within the linear range of 1-2500 μg / L. This quantitative logic is directly applicable to the white tea standard brewing solution and white tea extract in this embodiment; when extended to white tea concentrate, beverage base liquid, or production intermediate liquid, the matrix effect, recovery rate, and applicable dilution ratios should be verified.

[0086] Table 4. Linear range, external standard regression equation, limit of detection, and limit of quantitation for 10 flavor quality markers.

[0087] 6. Data Processing and Quality Evaluation Rules

[0088] The external standard method quantitative results of 10 target compounds in the white tea sample solution were compiled into a "sample × 10 index" matrix. Samples with known origins, white tea product categories, or processing stages were used to establish a reference quality fingerprint database; the test samples were used for comparison with the reference fingerprints. The effective positive concentration of each target compound was logarithmically calculated. 10The data is standardized according to the mean and standard deviation of the training data and then input into the partial least squares discriminant model. The model output is the relative matching score or matching category between the test sample and each preset reference quality fingerprint, which can be used for quality comparison, reference fingerprint consistency evaluation, process tracking, or relative raw material source matching; the score is not a continuous sensory quality score.

[0089] The reference category labels determine the meaning of the model output. In this embodiment, a joint four-category label of "representative raw material combination + white tea product category" is used. Therefore, the model output is the relative matching result of the four joint reference quality fingerprints, rather than an independent pure variety conclusion, an independent production area conclusion, or a white tea quality score.

[0090] Example 3: Establishment of Reference Quality Fingerprint for White Tea

[0091] Establishment of reference quality fingerprints for standard brewing solutions of four representative types of white tea

[0092] To verify the applicability of the method of this invention in the quality fingerprint evaluation of white tea, this embodiment selects representative samples of Zhenghe Dabai and Fuding Dahao to construct a 2×2 reference system of "representative raw material combination × product type", specifically including Zhenghe Dabai Bai Mudan, Zhenghe Dabai Shoumei, Fuding Dahao Bai Mudan, and Fuding Dahao Shoumei (as shown in Table 5). The above four categories together constitute the four types of reference quality fingerprints in this embodiment. Since the production area and variety are interdependent in this embodiment, the relevant results are used for reference quality fingerprint matching and are not used to independently separate the production area effect and the variety effect.

[0093] Standard brewing solutions were prepared for all four types of white tea raw materials according to the national standard sensory evaluation method, and 10 quality markers were quantified using the UPLC-MS / MS method of this invention. The standard brewing solutions were used to establish a reference quality fingerprint for the raw materials, aiming to eliminate differences in extraction processes and ensure that the differences between samples primarily reflect the molecular composition characteristics of the four types of reference samples in this embodiment, providing a reference for matching the sources of subsequent extracts and beverage production samples.

[0094] It should be noted that the standard brewing solution has been quantitatively analyzed using the results shown in Table 6. Figure 3 The heatmap method was used for application verification. Since each type of standard brewing solution in this embodiment serves only as a reference sample point at the raw material end, the sample size is insufficient for meaningful cross-validation modeling. Therefore, the standard brewing solution was not used as the training / validation dataset for the PLS-DA model. The PLS-DA validation was conducted on 108 groups of extracts to examine the fingerprint retention capability of 10 index combinations under different extraction condition perturbations.

[0095] Table 5. Four Types of Joint Reference Quality Fingerprints and Sample Settings

[0096] Table 6. Quantitative results of 10 flavor and quality markers in the standard brewing solutions of four types of white tea (unit: μg / L)

[0097] Example 4: Validation of the fingerprint evaluation model for the quality of 108 groups of white tea extracts

[0098] To further verify the technical basis of the quality fingerprint evaluation model of this invention, which extends from standard white tea brewing liquid to white tea extract, beverage base liquid, and intermediate production liquid, white tea extract was prepared using the above four types of reference white tea raw materials under a four-factor, three-level process matrix. The four factors are extraction temperature, extraction time, water-to-tea ratio, and ultrasonic power, with levels of 60, 70, and 80 °C, 10, 20, and 30 min, 30, 40, and 50 mL / g, and 100, 300, and 500 W, respectively (Table 7). Twenty-seven experimental condition numbers were set for each type of reference sample, resulting in 108 sets of white tea extract data for the four types of samples; the 27 experimental condition numbers corresponded to 25 unique combinations of process parameters. This embodiment is not used to select the optimal process, but rather to verify whether the reference quality fingerprint can still be recognized by the model under different process perturbations.

[0099] Table 7. Process parameter settings for 27 experimental conditions

[0100] Note: Experiment numbers 1, 11 and 22 are repeated settings with the same central conditions (70 ℃, 20 min, 40 mL / g, 300 W), so the 27 experimental condition numbers correspond to 25 unique combinations of process parameters.

[0101] The aforementioned 108 sets of data represent 108 sets of extract detection data generated from four types of reference white tea raw materials under 27 experimental condition numbers, and not 108 independent batches of raw materials. This embodiment primarily verifies the adaptability of the quality fingerprint evaluation model to process condition disturbances not involved in training, i.e., it can still match the extract to the corresponding reference quality fingerprint under internal process disturbances. This result provides a methodological basis for source matching of beverage base liquids and production intermediate liquids, but it is not yet used to prove the general external traceability capability for samples with unidentified raw material batches, samples from different years, or samples not included in the reference database.

[0102] Based on the detection results of each type of sample under 27 experimental conditions, the average concentration of 10 biomarkers was calculated, and a 10-index average quality fingerprint of the four types of white tea extracts was constructed. The results are as follows: Figure 4As shown, although the concentration of the target analyte in the extract varies with the process intensity, under the strict grouping condition of simultaneously including four types of samples under the same process number in the training or validation set, the combined out-of-compromise balanced accuracy (BA) of the recommended 4-latent variable model is still 98.15%, with the lowest fold being 95.00%. Therefore, the "fingerprint retention" referred to in this embodiment means that the model can still match the sample to the corresponding reference quality fingerprint under process conditions that are not involved in modeling, rather than that the concentration of each target analyte is unaffected by the process. It should be noted that the 27 condition_ids contain 25 unique combinations of process parameters, of which 3 condition_ids correspond to the same central process parameter. To eliminate the possible influence of the same process parameter combination distributed in different folds, samples with identical temperature, time, water-to-tea ratio, and ultrasonic power were further combined into the same process signature for strict grouping sensitivity analysis. The results show that the combined out-of-compromise balanced accuracy of the 4-latent variable model is 99.07%, with the lowest fold still being 95.00%, indicating that the main conclusion of this embodiment regarding the combined fingerprint retention capability is not caused by repeated combinations of process parameters.

[0103] Table 8. Average concentrations of 10 biomarkers in four types of white tea extracts (mean values ​​of 27 experimental conditions, unit: μg / L)

[0104] Comparative Example 1: Evaluation of a Single Biomarker

[0105] The purpose of establishing the data model in this invention is not to replace UPLC-MS / MS detection or traditional sensory evaluation, nor to consider accuracy as the sole technical effect. Rather, it is to verify whether a specific combination of 10 quality markers can form a molecular quality fingerprint that can still be used for reference matching under perturbations in extraction conditions. The model input X only includes the quantitative concentrations of K3OGal, Q3OGlu, Rutin, M3OGal, Kaf, Quc, Myr, PCB2, CA, and QA; sample_class serves only as a supervisory label for the four classes of joint reference quality fingerprints; sample number, source description, white tea product category, condition_id, as well as temperature, time, water-to-tea ratio, and ultrasonic power are not included in X.

[0106] This embodiment employs Partial Least Squares Discriminant Analysis (PLS-DA). All latent variable candidate models share the same identical outer GroupKFold five-fold partition, where groups = condition_id; samples from the same process number are simultaneously included in the training or validation set. Within each fold, the logarithm of X is first taken. 10The standard scaler is fitted only during the training fold, and the corresponding validation fold is transformed. Within the training fold, a onehotencoder is fitted to the sample_class. Subsequently, a PLSRegression(scale=False) is fitted, and the class is obtained by taking the argmax of each row of the prediction matrix Y_hat. Therefore, validation fold information is not involved in fitting the standardized parameters or PLS coefficients.

[0107] The comparison scale is set to "single marker model" and "10-indicator combination model". The single marker model inputs only one target at a time, using one latent variable due to the limitation of the number of features; the 10-indicator combination model uses four latent variables selected through nested grouped cross-validation; the remaining logarithms... 10 The transformation, training standardization, grouping, and prediction rules are completely consistent. Figure 6 The combined out-of-range Balanced Accuracy of the individual markers shown ranged from 32.41% to 50.00%, while that of the 10-marker combination was 98.15%, indicating that the technical effect comes from the combination of chemical categories rather than any single marker (Table 9).

[0108] Table 10 further shows that the normalized weights of the 10 markers are all non-zero, and the Spearman correlation coefficients of the weight ranking of the five outer training layers have an average of 0.890 and a minimum of 0.818, indicating that the quality fingerprint evaluation model is not driven solely by a single marker. The above weights are only used to explain the relative information utilization in the model and are not equivalent to the causal contribution rate of each compound to the sensory quality of white tea.

[0109] Table 9. Selection and Leakage-Free Cross-Validation Results of the Recommended Four-Latent Variable White Tea Quality Fingerprint Evaluation Model

[0110] Table 10 Normalized weights of 10 biomarkers for the recommended 4 latent variable white tea quality fingerprint evaluation model

[0111] Note: In each outer training fold, the absolute values ​​of the output coefficients for the four categories are first averaged, and then the 10 markers are normalized to a total of 100%. The values ​​in the table are the mean ± standard deviation of the five outer training folds. The Spearman correlation coefficients for the weight ranking between folds have a mean of 0.890 and a minimum of 0.818. These weights are used to explain the model contribution in this embodiment and are not equivalent to the true contribution rate or causal effect of a single compound on the flavor quality of white tea.

[0112] Final recommended quality fingerprint evaluation model formula and sample evaluation steps: After completing the model selection using nested grouped cross-validation, all 108 sets of data were used to establish the final 4-latent variable PLS-DA quality fingerprint evaluation model in this embodiment, following the same log10 transformation and standardization rules. The standardized parameters were then incorporated into the deployment model coefficients and intercepts, forming a fixed scoring formula that can be directly used to evaluate the samples to be tested. In the following general model, K represents the number of pre-defined reference quality fingerprint categories; in this embodiment, K=4.

[0113] make:

[0114] Among them, C i For the first i The effective positive concentrations of the biomarkers were measured according to the method of this invention and converted to the original samples, with the units uniformly expressed as μg / L. i =1,…,10.

[0115] For a white tea quality fingerprint evaluation task containing K preset categories, the first... k The model score for the class is:

[0116] in, The intercept of the k-th class is... For the i-th marker, the corresponding to the th k The overall coefficient of the deployment model of the class. k =1, ..., K. In this embodiment, K=4, and the four classes are FDDH_SM, FDDH_WP, ZHDB_SM, and ZHDB_WP, respectively. Their deployment model coefficients and intercepts are shown in Table 11. The maximum matching reference quality fingerprint of the test sample is:

[0117] The deployment model's comprehensive coefficients and intercepts in this formula already include the standardized correction information from the StandardScaler during the training phase. Therefore, when calculating the test samples, it is only necessary to unify the original concentrations of the 10 biomarkers to μg / L and perform logarithmic calculations. 10 The conversion does not require re-execution of StandardScaler processing. The model score represents the relative matching degree between the tested sample and the set K-class reference quality fingerprint, and is not directly interpreted as a probability value or continuous sensory quality score; depending on the label definition of the evaluation task, the matching results can be used for quality category comparison, batch consistency analysis, process tracing, or source matching.

[0118] When any target analyte fails to yield a valid quantitative result greater than 0, or fails to fall within the corresponding standard curve range even after adjusting the dilution factor, or when the matrix type, processing method, and target analyte concentration distribution of the sample significantly exceed the coverage of the training data in this embodiment, the above scoring formula should not be directly applied. Instead, the sample should be re-tested, the dilution factor adjusted, or the model expanded and re-validated using new samples. For evaluation tasks where the category system or number of categories changes, the corresponding K-classification model should be retrained and the corresponding coefficients determined; the four-category deployment coefficients in Table 11 should not be directly applied.

[0119] Table 11. Coefficients of the four reference fingerprint score functions (log) of the final recommended four-latent variable white tea quality fingerprint evaluation model. 10 (Concentration scale)

[0120] Note: The coefficients in Table 11 have been merged when using all 108 sets of data to build the final model, for log... 10 The concentration is calculated using the mean and scale employed by StandardScaler. Therefore, when using this deployment formula, simply take the logarithm of the original sample concentration, which is then standardized to μg / L. 10 Then, the values ​​are directly substituted without executing StandardScaler separately. If the original model object is called, StandardScaler should be executed first according to the training process before prediction. The two calculation methods should not be mixed. The calculation results of the two methods are consistent; Table 11 shows the deployment formula after completing the cross-validation selection and is not used for backcalculation or to replace the aforementioned performance trade-offs.

[0121] When evaluating the samples to be tested, the quantitative results of 10 markers are first obtained according to the UPLC-MS / MS-MRM method of this invention, and the units are standardized; then, log10 is taken and substituted into the four scoring functions in Table 11; the category corresponding to the highest score is taken as the matching result in the four types of reference quality fingerprints in this embodiment. FDDH_SM, FDDH_WP, ZHDB_SM, and ZHDB_WP correspond to the reference quality fingerprints of Fuding Dahao Shoumei, Fuding Dahao Bai Mudan, Zhenghe Dabai Shoumei, and Zhenghe Dabai Bai Mudan, respectively. When evaluating beverage base liquid or production intermediate liquid, the matching result is interpreted as a relative raw material source matching result only if the matrix has been included in the reference database and independent model validation has been completed.

[0122] The above evaluation is applicable only to samples consistent with the detection methods, concentration units, matrix matching rules, and four types of reference quality fingerprint systems used in this embodiment. This embodiment directly verifies the reference fingerprint matching ability of the four types of white tea extracts under different extraction process conditions. For beverage base liquids, intermediate production liquids, different raw material batches, different years, or other production areas, varieties, and white tea product categories not included in the reference system, independent reference samples should be added and external verification should be conducted before evaluating their predictive ability. The results are not directly extended to general geographical origin traceability without the limitation of reference databases, nor are they equivalent to continuous sensory quality scores.

[0123] In beverage production applications, a phased reference quality fingerprint can be established first using standard brewing solutions, extracts, and corresponding beverage base solutions or intermediate production solutions from known raw material sources. Then, a validated model is used to compare the relative matching relationships between the test sample and each reference fingerprint, which is used for raw material source matching and production process verification. If fruit juice, milk components, flavorings, or other ingredients that significantly alter the distribution of the matrix and target substances are added to the formula, a reference fingerprint under the corresponding formula matrix should be established, and the model should be retrained and validated. The four types of extract models in this embodiment should not be directly applied.

[0124] Model results show that under nested GroupKFold with an outer 5-fold and an inner 4-fold structure, and following the 1-SE rule, the number of latent variables selected in each outer training fold ranges from 3 to 5. The 4-latest variable appears 3 times, with both its mode and median being 4. Therefore, a model with a fixed 4-latest variable is ultimately recommended, rather than simply determining it based on the highest Balanced Accuracy of a particular latent variable or a single instance. This model achieves a combined out-of-fold Balanced Accuracy and Macro-F1 score of 98.15% under the same outer grouping, a 5-fold Balanced Accuracy mean of 98.00% with a standard deviation of 2.74%, and a lowest fold Balanced Accuracy of 95.00%. The confusion matrix is ​​shown below. Figure 7 As shown: FDDH_SM and FDDH_WP each had one sample that was misclassified from one another, while ZHDB_SM and ZHDB_WP each had 27 samples that were correctly classified. To maintain a test structure where each process number included samples of all four categories, the four categories of labels were randomly permuted 200 times within each condition_id. The mean Balanced Accuracy after permutation was 25.15%, with a 95% accuracy of 15.74%–37.06%, close to the four-category random level of 25%. One-sided permutation tests of the observed results... p The value is 0.0050.

[0125] Figure 5 Used to display the sample distribution of the first two latent variables in the recommended 4-latent variable model; Figure 6 Provide a strict out-of-comparison between a single marker and a combination of 10 indicators; Figure 7Give the external confusion matrix of the final model; Figure 8 Provide the weighted mean and inter-fold standard deviation; Figure 9 Provide the sensitivity results for 1-10 latent variables. Figure 6 It is evident that the Balanced Accuracy of any single marker does not exceed 50.00%, while the combination of 10 markers reaches 98.15%; Figure 9 As can be seen, the combined out-of-complement (BA) of 4 and 5 latent variables is the same, both at 98.15%. However, the mode of the nested 1-SE model is 4, so the 4-latest-variable model with lower complexity and more stable weight ranking is chosen. Considering nested grouped cross-validation, the 1-SE rule, model complexity, and out-of-complement stability, the 4-latest-variable model is ultimately recommended. The combined out-of-complement Balanced Accuracy and Macro-F1 of this model are both 98.15%. Conventional physicochemical or catechin indicators can still be used as indicators for extraction intensity, production process, or beverage process control. The technical positioning of the 10 biomarkers in this invention is supported by metabolomics evidence, can be quantified using the same method and caliber using a unified single-needle analysis time of 5 min UPLC-MS / MS-MRM method, and is a cross-category molecular quality fingerprint combination validated by a leak-free model in the process perturbation data formed by 27 experimental condition numbers.

[0126] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A combination of markers for identifying white tea raw materials, characterized in that, Including kaempferol-3- O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O -Galactoside, Kaempferol, Quercetin, Myricetin, Proanthocyanidins B2, Citric acid and Quinic acid.

2. The application of the marker combination described in claim 1 in constructing a fingerprint database of white tea raw materials or a fingerprint evaluation model for the quality of white tea raw materials.

3. The method for detecting the combination of markers as described in claim 1, characterized in that, Includes the following steps: Step 1: Take the extract of the white tea sample to be tested, and use UPLC-MS / MS multiple reaction monitoring mode to simultaneously detect it in the same chromatographic-mass spectrometry method and the same detection sequence in one injection to obtain the peak area of ​​each marker; Step 2: Establish standard curves for each marker using the external standard method, and calculate the concentration of each marker based on its peak area.

4. The detection method according to claim 3, characterized in that, In step 1, the synchronous detection time is less than or equal to 5 minutes.

5. A method for constructing a fingerprint evaluation model for the quality of white tea raw materials, characterized in that, Includes the following steps: Step 1: Obtain different white tea raw materials and prepare extracts; Step 2: Detect kaempferol-3- in the extract obtained in Step 1 using the detection method described in claim 3 or 4. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - Concentrations of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid; Step 3: Perform logarithmic transformation and standardization on the concentration data obtained in Step 2 to obtain the quality fingerprint feature vector of the extract, and form a quality fingerprint spectrum based on different extracts; Step 4: Establish a database based on the quality fingerprint spectra of different extracts obtained in Step 3, and use the database to train and obtain a quality fingerprint evaluation model for white tea raw materials.

6. The construction method according to claim 5, characterized in that, In step 1, the different white tea raw materials include at least one of Zhenghe Da Bai Bai Mu Dan, Zhenghe Da Bai Shou Mei, Fuding Da Hao Bai Mu Dan and Fuding Da Hao Shou Mei; The preparation method includes mixing white tea raw materials with water, and extracting the extract to obtain an extract, wherein: The volume-to-mass ratio of the white tea raw material to water is (30~50) mL:1g, the extraction temperature is 60~80℃, the extraction time is 10~30 min, and the ultrasonic power is 100~500 W.

7. The construction method according to claim 5, characterized in that, In step 4, the training method includes at least one of partial least squares discriminant analysis, orthogonal partial least squares discriminant analysis, support vector machine, random forest, and artificial neural network.

8. A method for identifying white tea raw materials, characterized in that, Includes the following steps: S1. Take a sample of white tea to be tested and detect kaempferol-3- in the sample using the detection method described in claim 3 or 4. O -Galactoside, Quercetin-3- O -Glucoside, Rutin, Myricetin-3- O - Concentrations of galactoside, kaempferol, quercetin, myricetin, proanthocyanidin B2, citric acid, and quinic acid; S2, logarithmically transform and standardize the concentration data obtained in S1 to obtain the quality fingerprint feature vector of the white tea sample to be tested, and form a quality fingerprint spectrum. S3. Input the quality fingerprint spectrum described in S2 into the white tea raw material quality fingerprint evaluation model constructed by the construction method described in any one of claims 5 to 7, obtain the matching relationship between the white tea sample to be tested and the white tea raw material preset in the white tea raw material quality fingerprint evaluation model, and determine the variety of the white tea sample to be tested based on the matching relationship.

9. The identification method according to claim 8, characterized in that, In S1, the white tea sample to be tested includes at least one of the following: white tea standard brewing liquid, white tea extract, white tea concentrate, white tea beverage base liquid, white tea beverage production intermediate liquid, and white tea products adapted to the matrix. In step S3, the matching relationship includes at least one of relative matching score, maximum matching category, quality fingerprint difference, or consistency result.

10. The application of the marker combination of claim 1, the detection method of claim 3 or 4, the construction method of any one of claims 5 to 7 to construct the white tea raw material quality fingerprint evaluation model, or the identification method of claim 8 or 9 in the quality analysis of white tea.

Images