Wine classification and evaluation methods

Abstract

To quickly provide wine classification / evaluation means based only on wine ingredients without referring to price information and origin / brand information for wine classification.SOLUTION: It has been confirmed that a wide variety of wines can be properly classified by performing clustering processing for each of basic tastes, such as umami, sweetness, sourness, and bitterness, and then combining evaluations of clusters for the basic tastes.SELECTED DRAWING: None

Description

[Technical Field]

[0001] The present invention relates to a method and system for classifying and evaluating wine. [Background technology]

[0002] The components that make up wine are diverse, including amino acids, organic acids, nucleosides, and nucleotides, and recent advances in analytical technology have led to the proposal of various analytical methods. For example, a method for analyzing wine using food metabolomics using an analyzer that employs an ion-pair-free LC / MS / MS method, which enables simultaneous analysis of hydrophilic metabolite components, has been reported (e.g., a triple quadrupole mass spectrometer LCMS). TM (See the Shimadzu-8060NX pamphlet.) However, in the case of wine, it is said that the composition of precisely analyzed components does not directly reflect the taste of the wine. Also, although gas chromatography has been used to measure aroma components in wine, it is known that components with high peak intensities (alcohols, fatty acids) do not necessarily contribute significantly to the sensory experience (see, for example, Non-Patent Document 1).

[0003] Recent wine component analysis technologies include a mobile terminal (see, for example, Patent Document 1) for use in a wine taste evaluation system, which includes a control unit configured to match a user's taste data with reference taste data related to the taste of wine stored in the wine information memory, quantify the degree of match between the user's taste data and the reference taste data, and display the quantified data on a display. However, because the user's taste data is the subject of matching, it is difficult to objectively evaluate the wine itself. Also, a wine probe and a method for measuring wine characteristic quantities from one such probe (see, for example, Patent Document 2) have been proposed, which can express measurements encoded in an RGB framework in a framework known as hue, saturation, and brightness (HSL) and / or CIELAB. While the method provides the user with data such as ripeness, body, or acidity, it does not provide an evaluation of the wine's taste. Furthermore, in a computer system that suggests wines to pair with food (see, for example, Patent Document 3), "taste" is evaluated from multiple perspectives related to wine characteristics such as "sweetness," "saltiness," "acidity," "astringency," and "maturity," but there is no specific instruction on how each of these tastes is separated and evaluated.

[0004] In addition, while wine evaluation has recently become possible through precise component analysis, including the metabolomics mentioned above, there is also an accumulation of traditional evaluations based on origin and brand, and pricing is often determined in conjunction with these traditional evaluations. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. WO2016 / 148124 Brochure [Patent Document 2] Special Publication No. 2019-526811 [Patent Document 3] Patent No. 6424300 [Non-patent literature]

[0006] [Non-Patent Document 1] Biotechnology Vol. 98 No. 12 (2020) 674-678 Summary of the Invention [Problem to be solved by the invention]

[0007] The object of the present invention is to quickly provide a classification and evaluation means for classifying wines based solely on the ingredients contained in the wines, without referring to price information or information on origin and brand, and which is highly consistent with the evaluations of wine drinkers. [Means for solving the problem]

[0008] The present inventors have been investigating methods for classifying and evaluating wine taste. First, they attempted to cluster wines by evaluating all of the components that make up wine, including amino acids, organic acids, nucleosides, and nucleotides, using recently developed metabolomics technology. However, they were unable to identify common taste characteristics among the wines belonging to each cluster generated by clustering. Therefore, they selected components that make up each of the commonly known basic tastes, such as umami, sweetness, sourness, and bitterness, and performed clustering for each basic taste. They found that they could achieve classifications that closely matched the taste ratings of wine experts. Furthermore, they confirmed that a wide variety of wines could be appropriately classified by combining evaluations, such as by calculating the sum of the ratings (values) of the cluster groups for each basic taste. They also confirmed that such classifications correspond to the enhanced taste sensations experienced when wine is served with meat or fish dishes, due to the synergistic effect of wine components and the taste-affecting components contained in the food. This led to the completion of the present invention.

[0009] That is, the present invention is as follows. [1] A method of classifying wines that meets the following criteria (a) to (f) in order. (a) a data acquisition step for acquiring component profile data of a reference wine; (b) a digitization step of quantifying the content of each component constituting the component profile data of the reference wine acquired in the data acquisition step of step (a) into a characteristic value; (c) a data extraction step of extracting characteristic values ​​of components constituting basic tastes from the characteristic values ​​of each component of the reference wine created in the quantification step of step (b); (d) a clustering step of forming n cluster groups for each basic taste by performing clustering by unsupervised learning for each feature value of the components constituting each basic taste extracted in the extraction step of step (c); (e) a group classification step of classifying the reference wines into groups based on the cluster groups for each basic taste formed in step (d); (f) a comment creation process for associating taste information and creating comments; [2] The classification method described in [1] above, characterized in that the basic tastes consist of umami and sweetness. [3] The classification method described in [1] above, characterized in that the basic tastes consist of umami, sweetness, and sourness. [4] The classification method described in [1] above, characterized in that the basic tastes consist of umami, sweetness, sourness, and bitterness. [5] The classification method according to any one of [2] to [4] above, characterized in that the components that constitute umami include one or more selected from glutamic acid, inosinic acid, aspartic acid, adenylic acid, and succinic acid. [6] The classification method according to any one of [2] to [4] above, characterized in that the components that constitute umami are glutamic acid and succinic acid. [7] A classification method according to any one of [2] to [4] above, characterized in that the components that constitute sweetness include one or more selected from proline, glycine, alanine, threonine, serine, ornithine, and asparagine. [8] The classification method according to any one of [2] to [4] above, characterized in that the components that constitute sweetness are proline and alanine. [9] The classification method according to [3] or [4] above, characterized in that the components constituting the sour taste include one or more selected from tartaric acid, malic acid, citric acid, succinic acid, lactic acid, glycyl glutamine acetate, seryl glutamine, and glycyl asparagine.

[10] The classification method according to [3] or [4] above, characterized in that the components that constitute the sour taste are malic acid, citric acid, succinic acid, and lactic acid.

[0010] The present invention is also as follows.

[11] A wine classification system that outputs comments about wine, including classification results using wine component profile data, comprising the following (A) to (F). (A) a data acquisition unit for acquiring component profile data of reference wine; (B) a quantification unit that quantifies the content of each component constituting the component profile data of the reference wine acquired in the data acquisition unit into a characteristic value; (C) a data extraction unit that extracts the quantified data of the components that constitute the basic tastes from the quantified profile data; (D) A clustering section that performs clustering using unsupervised learning for each of the quantified data of the components that make up each basic taste, thereby forming n clusters for each basic taste; (E) A group classification section that performs group classification of reference wines based on n clusters formed for each basic taste; (F) A comment creation unit that associates taste information and creates comments;

[12] The classification system according to

[11] above, characterized in that the basic tastes consist of umami and sweetness.

[13] The classification system according to

[11] above, characterized in that the basic tastes consist of umami, sweetness, and sourness.

[14] The classification system according to

[11] above, characterized in that the basic tastes consist of umami, sweetness, sourness, and bitterness.

[15] A classification system according to any one of

[12] to

[14] above, characterized in that the components that constitute umami include one or more selected from glutamic acid, inosinic acid, aspartic acid, adenylic acid, and succinic acid.

[16] The classification system according to any one of

[12] to

[14] above, characterized in that the components that constitute umami are glutamic acid and succinic acid.

[17] A classification system according to any one of

[12] to

[14] above, characterized in that the components constituting sweetness include one or more selected from proline, glycine, alanine, threonine, serine, ornithine, and asparagine.

[18] The classification system according to any one of

[12] to

[14] above, characterized in that the components that constitute sweetness are proline and alanine.

[19] The classification system according to

[13] or

[14] above, characterized in that the components constituting the sour taste include one or more selected from tartaric acid, malic acid, citric acid, succinic acid, lactic acid, glycyl glutamine acetate, seryl glutamine, and glycyl asparagine.

[20] The classification system according to

[13] or

[14] above, characterized in that the components that constitute the sour taste are malic acid, citric acid, succinic acid, and lactic acid.

[21] A method for collecting data for wine classification, which uses wine component profile data provided by a user as input data, classifies the wine according to any of the wine classification methods described in [1] to [4] above, and records the results as reference data. [Effects of the Invention]

[0011] According to the present invention, wines can be classified without being influenced by price, brand, or distribution volume. [Brief explanation of the drawings]

[0012] [Figure 1]This graph shows the results of component analysis of 53 reference wines using a triple quadrupole mass spectrometer and LC / MS / MS method package / cell culture profiling, quantifying the peak values ​​detected, and clustering all detected components. [Figure 2] The following graph shows the results of clustering, in which umami, sweet, sour, and bitter components were extracted and classified into six clusters. [Figure 3] 10 is a graph showing the results of clustering performed to classify umami components into three clusters. [Figure 4] 10 is a graph showing the results of clustering sweet components into three clusters. [Figure 5] 10 is a graph showing the results of clustering performed to classify sour components into three clusters. [Figure 6] 10 is a graph showing the results of clustering performed to classify bitter components into three clusters. [Figure 7] (a) shows an example of a flowchart for the method of the present invention when n=3, and (b) shows an example of a flowchart for predicting the group of a test sample wine using the grouping of a reference wine as a training value. [Figure 8] 1 is a diagram showing an example of the configuration of a classification system 1 according to an embodiment of a wine classification method. [Figure 9] (a) Heat map correlation matrix for glutamic acid, aspartic acid, and succinic acid. (b) Graph showing the correlation between glutamic acid and aspartic acid for each cluster. [Figure 10] The graphs show the results of cluster analysis of glutamic acid and succinic acid, which are components that make up umami, using (a) the elbow method and (b) the silhouette method. [Figure 11](a) Scatter plot of umami taste, divided into seven clusters using the K-means method. (b) Scatter plots for red wine only and (c) for white wine and sparkling wine. (d) Scatter plots for white wine and sparkling wine, created based on the automatically assigned cluster numbers, reassigned to each cluster group (G) after checking the amount of glutamic acid and succinic acid in each cluster. [Figure 12] (a) A graph confirming the amount of glutamic acid in each cluster. (b) A graph confirming the amount of succinic acid in each cluster. [Figure 13] Correlation matrix in heat map format for proline, alanine, glycine, threonine, and serine. [Figure 14] (a-1) A scatter plot of paired plots where the Y-axis is alanine and the X-axis is proline, and the clusters are divided into three groups using the K-means algorithm. (a-2) A scatter plot of paired plots where the Y-axis is proline and the X-axis is alanine, and the clusters are divided into three groups using the K-means algorithm. (b) A graph showing the correlation between proline and alanine, labeled to confirm the distribution of each type of wine. Each cluster is displayed separately. [Figure 15] (a) A graph confirming the abundance of proline in each cluster. (b) A graph confirming the abundance of alanine in each cluster. [Figure 16] Correlation matrix in heat map format for malic acid, lactic acid, citric acid, and succinic acid. [Figure 17] The following shows a scatter plot of pair plots for each component when malic acid, lactic acid, citric acid, and succinic acid were divided into five clusters using the K-means method. [Figure 18] 1 is a graph showing the correlation between malic acid and lactic acid. [Figure 19](a) A graph confirming the amount of malic acid for each cluster. (b) A graph confirming the amount of lactic acid for each cluster. (c) A graph confirming the amount of citric acid for each cluster. (d) A graph confirming the amount of succinic acid for each cluster. [Figure 20] Correlation matrix in heat map format for proline, histidine, arginine, valine, methionine, tryptophan, phenylalanine, isoleucine, leucine, and lysine. [Figure 21] (a) A graph showing the amount of arginine in each cluster. [Figure 22] (a) is a graph showing the concentration of glutamic acid in red wine at which the umami effect can be expected; (b) is a graph showing the concentration of glutamic acid in white wine at which the umami effect can be expected. [Figure 23] (a) Graph showing the correlation between proline and alanine in red wine. (b) Graph showing the correlation between proline and alanine in white wine and sparkling wine. [Figure 24] (a) Graph showing the abundance of proline and (b) alanine in each cluster. DETAILED DESCRIPTION OF THE INVENTION

[0013] The present invention is not particularly limited as long as it comprises a wine classification method including: (a) a data acquisition step of acquiring component profile data of a reference wine; (b) a digitization step of quantifying the content of each component constituting the component profile data of the reference wine acquired in the data acquisition step of step (a) into a feature value; (c) a data extraction step of extracting feature values ​​of components constituting basic tastes from the feature values ​​of each component of the reference wine created in the digitization step of step (b); (d) a clustering step of forming n cluster groups for each basic taste by clustering by unsupervised learning for each feature value of the components constituting each basic taste extracted in the extraction step of step (c); and (e) a comment creation step of associating taste information with each group clustered for each basic taste and creating a comment.

[0014] FIG. 8 is a diagram showing an example of the configuration of a classification system 1 according to an embodiment of the wine classification method of the present invention. The classification system 1 is not particularly limited as long as it is a classification system equipped with a data acquisition unit 10, a digitization unit 20, a data extraction unit 30, a clustering unit 40, a group classification unit 50, and a comment creation unit 60, and may further include a component determination unit that determines the components that constitute the basic tastes from the digitized profile data, as necessary.

[0015] The data acquisition step (a) above is not particularly limited as long as it is a step of acquiring component profile data of reference wines, and the reference wines may be wines that can be selected from all wines on the market. The number of reference wines is not particularly limited as long as they are a number that can be classified into n clusters, but examples include 50 or more types, preferably 100 or more types, more preferably 200 or more types, even more preferably 1000 or more types, and even more preferably 2000 or more types of wine, and the maximum number can be all types of wine that may be on the market.

[0016] In the data acquisition step (a) above, known component analysis methods can be used to acquire component profile data for the reference wines. Component detection unit 5 capable of detecting and analyzing the components that make up each reference wine can be an instrument that performs known chemical techniques such as gas chromatography / mass spectrometry (GC / MS), liquid chromatography / mass spectrometry (LC / MS), capillary electrophoresis / mass spectrometry (CE / MS), mass spectrometry using MALDI-TOF-MS, GC-TOF / MS analysis, and metabolome analysis (metabolomics analysis). Specifically, commercially available metabolomics analysis software such as Profiler TM AM + , UPLC® / Synapt for Metabonomics TM HDMS TM System (manufactured by Nihon Waters Co., Ltd.), triple quadrupole mass spectrometer LCMS TM The following instruments can be used: GCMS-8050, 8060NX, GCMSNX series, GCMS-QPTM2020, GCMS-TQTM series, GCMS-QP2010 SE (all manufactured by Shimadzu Corporation); and for solid-phase microextraction (SPME), Pegasus® HT GC-TOFMS (manufactured by LECO) and Agilent 1290 Infinity II LC system (manufactured by Agilent). The data acquisition unit 10 of the present invention can acquire profile data on the components that make up wine using the so-called external component detection and analysis system described above. Alternatively, data can be acquired directly by the data acquisition unit 10, which includes the component detection unit 5 that detects and analyzes the components that make up the reference wine, using homemade software or commercially available multivariate analysis software.

[0017] The components included in the component profile data of the reference wine are not particularly limited as long as they are components constituting each reference wine, but may include amino acids such as histidine, arginine, valine, methionine, tryptophan, phenylalanine, isoleucine, leucine, lysine, cysteine, tyrosine, proline, glycine, alanine, threonine, serine, asparagine, aspartic acid, glutamic acid, and ornithine; succinic acid, citric acid, malic acid, lactic acid, tartaric acid, gluconic acid, and glycerin. Examples of the basic flavors include organic acids such as phosphonic acid and threonic acid; nucleic acids and nucleic acid bases such as inosinic acid, adenylic acid, hypoxanthine, adenine, guanine, thymine, cytosine, uracil, purine, and pyrimidine; vitamins such as nicotinic acid, pantothenic acid, nicotinamide, riboflavin, and 4-aminobenzoic acid; pigments such as anthocyanins and tannins; sugars such as glucose and fructose; and dipeptides such as glycyl-glutamine and alanyl-glutamine. However, it is desirable that the basic flavors contain the components constituting each of the basic flavors of the present invention.

[0018] The component profile data of the reference wine can be acquired in the data acquisition unit 10 through processes such as detection, analysis, and identification, and then stored in the component profile data storage unit 71. The stored data can also be output to the quantification unit 20 to extract the characteristic values ​​of the components that make up the basic flavors.

[0019] In addition, in order to obtain wine profile data, the necessary wine pretreatment was performed for each of the multiple reference wines. 0 Specifically, for example, before applying the reference wine to the LCMS, the reference wine may be passed through an ultrafiltration filter to remove sediment, precipitates, etc., or may be diluted with water. After such treatments are performed in the pretreatment unit 4, detection, analysis, identification, and other treatments may be performed.

[0020] The quantification step (b) above is not particularly limited as long as it is a step of quantifying the content of each component constituting the component profile data of the reference wine into a feature value, and examples include a step of converting the peak height of a chromatogram specific to each component into a relative quantitative value, or a step of processing the concentration specific to each metabolite (compound) such as an amino acid as an absolute value. However, since using peak heights from a single chromatogram measurement may result in fluctuations in peak heights from measurement to measurement, a preferred example is a step of measuring a specific concentration of each metabolite as a control value and converting the peak value into a concentration. The feature value (peak value) is not particularly limited as long as it is a numerical value that represents the feature of each component, and examples include 1) identity feature value, 2) mean feature value, 3) max feature value, and 4) min feature value.

[0021] For example, the height and area of ​​a peak measured by LC-MS can be digitized using known software, and this digitized value can be used as the peak value. Alternatively, the content (concentration) of each component can be converted to a corresponding value by creating a calibration curve.

[0022] When the identity characteristic value is used, the identity characteristic value indicates a characteristic value that indicates the content itself.

[0023] The process of quantifying the content of each component constituting the reference wine component profile data acquired by the data acquisition unit into a feature value can be executed by the quantification unit 20, and the peak height of a chromatogram specific to each component can be quantified into a relative quantitative value, or the component concentration in each wine specific to each metabolite (compound) such as an amino acid can be quantified. Furthermore, the data related to the feature values ​​of the reference wine obtained as a result of this process can be stored in the feature value data storage unit 72, or can be presented by the information presentation unit 80.

[0024] The extraction step (c) above is not particularly limited as long as it is an extraction step that extracts the characteristic values ​​of the components that constitute the basic tastes from the characteristic values ​​of each component of the reference wine created in the quantification step (b).The basic tastes are generally defined as sweetness, umami, bitterness, sourness, and saltiness, which are tastes that are perceived by taste cells and can be clearly distinguished from one another.However, the basic tastes in the present invention can include umami, sweetness, sourness, bitterness, and unpleasant tastes, and can be limited to umami, sweetness, sourness, and bitterness, or can be limited to umami, sweetness, sourness, and unpleasant tastes, but must include at least umami and sweetness.

[0025] The extraction step (c) performed by the data extraction unit 30 to extract the feature values ​​of the components can be followed by a step (c0) of determining the components of each basic taste to be used in the subsequent clustering step, etc., for each of the basic tastes. One method of determining the components that make up such basic tastes is to select components known from literature, etc. to constitute each basic taste. However, for example, when there is a strong correlation (multicollinearity) between the variables of multiple components, if the correlation coefficient of peak or concentration data is greater than a predetermined value, the effect can be predicted by other components that are strongly correlated with the variables, and there is a risk that the importance of the variables inherent in a specific component may not be properly reflected. Therefore, data for one or more components can be deleted, and the information composed of the remaining data can be used as quantified data of the components that make up each basic taste, and can also be used as input data for generating a learning model.

[0026] The component determination unit 35 can determine whether or not multicollinearity exists by displaying the correlation coefficients between each component as a graph or by displaying a correlation matrix consisting of the correlation coefficients in heat map format. By examining the results, the components to be used for clustering can be determined. When determining the components with variables to be used for classification, for example, even if the content of a component in a wine is below the taste threshold, knowledge regarding the enhanced taste due to the synergistic effect of, for example, glutamic acid and inosinic acid, and / or succinic acid and inosinic acid, when consumed with food can be further taken into account. This process unifies data for components with overlapping characteristics, thereby improving the accuracy of the generated learning model. The correlation matrix and heat map can be created using publicly known software, such as the Python library, the pandas library, matplotlib, and seaborn. Graphs showing the correlation coefficients between components and heat maps showing the correlation matrix consisting of the correlation coefficients can be displayed on the display 810 of the information presentation unit 80.

[0027] In a method for determining the components that make up each taste using the heat map, for example, if the correlation coefficient of each component on the heat map is 0.7 or less, 0.5 or less, preferably 0.4 or less, and more preferably 0.3 or less, the component can be picked up as a component of each taste. Furthermore, even if the correlation between two specific components is 0.7 or more, 0.5 or more, 0.4 or more, or 0.3 or more, the component can also be picked up as a component of each taste if the correlation with multiple other components on the heat map satisfies, for example, 0.3 or less.

[0028] The heat map display and the graph of the correlation coefficients can also be displayed on the display of the information presentation unit 80. For example, the display 810 may display the correlation coefficients in a graph, or may display a correlation matrix consisting of the correlation coefficients in a heat map format. The memory 700 of the data storage unit can also store the correlation coefficients and knowledge about the thresholds at which humans sense the components that make up each taste, and the operator can determine the components that make up each taste by referring to such data.

[0029] As described above, when overlapping data information is deleted from the data of one peak and the remaining peak data is used to generate input data from which unnecessary peak (feature) data has been deleted, the accuracy of the generated learning model is improved by using data that associates peak information and concentration information indicating the results of mass spectrometry for each wine sample with characteristic information indicating specific characteristics of the sample, such as its taste.

[0030] Umami is one of the basic tastes that was confirmed to be detectable in 2002 when receptors that respond to glutamate were identified in the sensory cells in the taste buds of the tongue. Examples of components that make up umami include one or more selected from the group consisting of glutamic acid, inosinic acid, aspartic acid, adenylic acid, succinic acid, glutathione, histidine, carnosine, and lactic acid, but it is preferable to select at least glutamic acid and succinic acid.

[0031] Examples of components that contribute to the sweetness include proline, glycine, alanine, threonine, serine, ornithine, glycerol, etc. The yeast in grape juice consumes the sugars contained in the grapes, such as glucose and fructose, through fermentation, and when the alcohol content reaches about 15%, the yeast dies once the sugars are consumed. Therefore, although sweet wines and very sweet wines may contain residual sugars such as glucose and fructose, proline, glycine, alanine, threonine, and serine can be selected as components that impart a sweetness common to all wines, including dry wines, and sweetness can be evaluated based on their content and content ratio.Furthermore, sweetness containing one or more of the above-mentioned proline, glycine, alanine, threonine, and serine can be defined as amino acid sweetness, and in addition to the above-mentioned (amino acid) sweetness, sweetness containing sugars such as glucose and fructose as constituent components can be further defined as sugar sweetness as a basic taste in this invention, and the wine classification method of this invention can be implemented by adding such sugar sweetness to the basic tastes, but sugar sweetness can be included or excluded, and it is preferable to define alanine and proline as the components that make up sweetness and implement the wine classification method of this invention.

[0032] In the present invention, examples of components that constitute sourness include acids derived from grapes, such as tartaric acid, malic acid, and citric acid; acids produced by yeast fermentation during the brewing process, such as succinic acid, lactic acid, and acetic acid; and dipeptides with flavor, such as glycyl glutamine, seryl glutamine, and glycyl aspartic acid. Although sourness can be evaluated based on the content and content ratio of tartaric acid, citric acid, malic acid, lactic acid, gluconic acid, and glycyl glutamine, which are components that commonly impart sourness to wines in general, it is preferable to set malic acid, lactic acid, citric acid, and succinic acid as the components that constitute sourness and carry out the wine classification method of the present invention.

[0033] Of the components that make up the acidity, malic acid and lactic acid can also be used as indicators for determining whether or not a wine has undergone malolactic fermentation (MLF). A cluster with a high malic acid content and a low lactic acid content can be determined to be a cluster of wines that have not undergone malolactic fermentation, while a cluster with a high lactic acid content and a low malic acid content can be determined to be a cluster of wines that have undergone malolactic fermentation. The presence or absence of malolactic fermentation can be included in the comments created by the comment creation unit.

[0034] Malolactic fermentation is a process in which lactic acid bacteria are used to convert malic acid in wine into lactic acid after primary fermentation (alcoholic fermentation) by yeast in the winemaking process, with the aim of reducing the acidity of the wine and giving it a mellow flavor. The lactic acid bacteria that perform malolactic fermentation may be those that are present on grapes or that have been artificially added.

[0035] Examples of the components that contribute to astringency include histidine, arginine, valine, methionine, tryptophan, phenylalanine, isoleucine, leucine, lysine, cysteine, tyrosine, nicotinic acid, hypoxanthine, pantothenic acid, nicotinamide, riboflavin, 4-aminobenzoic acid, anthocyanins, catechins (e.g., tannins), magnesium salts, and calcium salts. While literature suggests that tannins contribute to the astringency of wine, the present invention is characterized by selecting these components as components that contribute to the astringency common to wines in general, rather than the tannins that impart the characteristic astringency of red wines and some white wines, and evaluating astringency based on their content and proportion. Furthermore, it is preferable to implement the wine classification method of the present invention by defining proline, histidine, tryptophan, arginine, valine, and phenylalanine as components that contribute to astringency. Alternatively, the presence or absence of bitterness can be determined simply by the amount of arginine.

[0036] The basic flavors mentioned above can also contain unpleasant components, such as bioamines (such as histamine, putrescine, putrescine, cadaverine, beta-phenylethylamine, tryptamine, spermine, and spermidine), as well as sulfur compounds (such as 2-aminoethanol, guaiacol, ethylphenol, vinylphenol, ethyl acetate, acetaldehyde, acetone, sulfur dioxide, hydrogen sulfide, mercaptan, and methionol), and trichloroanisole (TCA), which are known to cause off-flavors in wine. Drinking wines containing higher levels of bioamines than standard wines can cause headaches, asthma, and allergic symptoms. The above-mentioned off-flavor components, excluding TCA, can sometimes be considered to add depth to the flavor in small amounts.

[0037] In the data extraction step of step (c) above, a process is carried out to extract the characteristic values ​​of the components that make up the basic tastes from the characteristic values ​​of each component of the reference wine created in the quantification step of step (b). This process can be executed in the data extraction unit 30, which extracts the quantified data of the components that make up the basic tastes from the quantified profile data. Furthermore, the data regarding the characteristic values ​​of each component that makes up the component profile data of the reference wine obtained as a result of this process can be stored in the characteristic value data storage unit 72, or can be presented as information in the information presentation unit 80.

[0038] The clustering step in step (d) is not particularly limited as long as it is a step of creating n clusters for each basic taste by performing clustering by unsupervised learning for each feature value of the components constituting each basic taste extracted in the extraction step in step (c), and clustering is performed by unsupervised learning based on the feature values ​​for each basic taste such as umami, sweetness, sourness, and / or astringency, unpleasant taste, etc.

[0039] Unsupervised learning is a technique that allows a learning system to learn the distribution of input data by providing it with a large amount of input data, and then compress, classify, and reshape the input data without providing it with the training data that corresponds to the input data. Unsupervised learning is used in many fields because it does not require a correct answer, but instead of having a correct answer, the results can be difficult to interpret, and even slight changes in the initial parameters can lead to completely different results, so it does not necessarily reach the optimal solution, and it may reach an intermediate solution (local solution) and end up learning there, so it is also possible to run multiple trials with different parameters.

[0040] The clustering mentioned above is a method of classifying a group of data, which belongs to the unsupervised learning mentioned above, by collecting data with similar properties and grouping them into one group (cluster), for which the correct answer is not known in advance. Specifically, a method using the k-means method, which is a technique that uses a non-hierarchical clustering algorithm to find the center of each cluster and classify each piece of data into the cluster with the closest cluster center, can be given as an example.

[0041] The k-means method is named after the fact that it is an algorithm for creating any specified k number of clusters. For example, it is an algorithm that first divides the data into appropriate clusters, and then adjusts the data so that it is divided more appropriately by iteratively minimizing the sum of squared errors (SSE) within the clusters.

[0042] While the above clustering is an exploratory data analysis method, and the division can be said to be based on some subjectivity or perspective, the elbow method or silhouette method can be used to automatically estimate the number of clusters (n). The elbow method, for example, tries methods such as the K-means method with various numbers of clusters, searches for the point where increasing the number of clusters does not significantly improve accuracy, and identifies the number of clusters obtained as a result of the search as the optimal number of clusters. This method plots the sum of squared errors (SSE) for each cluster, and considers the number of clusters at the point where the graph showing the change in SSE bends like an elbow to be the optimal number of clusters.

[0043] The results obtained by the clustering can be used to determine whether each cluster formed by the clustering is appropriate, if necessary. One method for determining whether each cluster is appropriate is to set model wines and select a classification in which clustering has been performed so that each model wine belongs to a different cluster. The selected clustering prevents or suppresses the possibility of falling into a local solution, and can be determined to be appropriate. Such a determination can also be made by the operator.

[0044] The model wines may be wines with established ratings that have a distinctive taste and / or that are different from one another. For example, two or more, preferably three or more, more preferably four or more, and even more preferably n or more wines as described below that have a distinctive taste and are identified as being of different types by a plurality of people who are knowledgeable about wine taste. Examples of people who are knowledgeable about wine taste include sommeliers, wine producers, wine traders, wine lovers, etc. Cut.

[0045] The number n is not particularly limited, but can be determined depending on the number of reference wines. For example, n = 3 or 4 if the number of reference wines is 50 to 500; n = 3, 4, or 5 if the number of reference wines is 100 to 1,000; n = 4 or 5 if the number of reference wines is 1,000 to 5,000; 4, 5, or 6 if the number of reference wines is 3,000 to 15,000; 5 or 6 if the number of reference wines is 5,000 to 30,000; and 6, 7, or 8 if the number is 10,000 to 100,000. However, the above numbers may vary depending on the bias in the origins of the reference wines and the diversity of wine types.

[0046] The data of each basic taste cluster generated by the above clustering can be stored in the cluster data storage unit 73, and the configuration can further include creating a graph such as a bar graph or a pie chart and displaying the graph on the display 810 so that the composition of the components in each cluster can be visually clarified.

[0047] The process of forming n clusters for each basic taste by clustering the characteristic values ​​of the components that make up each basic taste through unsupervised learning can be executed in the clustering unit 40. Furthermore, information indicating the characteristics of each cluster, obtained as a result of this process and interpreted based on the characteristics of the amount and blending ratio of each component in the cluster, can be stored in the cluster data storage unit 73. However, an information presentation device that is integrated with the cluster data storage unit 73 or separate from it can also present profile data information indicating the characteristics of the cluster in the form of a graph or the like for the n clusters for each basic taste.

[0048] The information stored in the data storage unit 72 for the above-mentioned characteristic values ​​includes the center point of the components and blend amounts that make up each cluster, the range of the components and blends that make up the cluster that are covered in the cluster, and explanatory comments about the basic tastes of the cluster to which each wine belongs.

[0049] Furthermore, by displaying the clustering results in a graph, the proportion of each component can be visually recognized immediately. For example, a bar graph is created by stacking the characteristic values ​​of the components in each cluster for each basic taste, and this graph can be used to determine whether the clustering is appropriate. This graph can also be displayed on the display 810 of the information presentation device.

[0050] Step (e) is not particularly limited as long as it is a step of classifying the reference wines based on the clustering generated in step (d), and an example of the classification method is a method of classifying the reference wines by combining the labeling numbers assigned to each cluster. For example, the labeling method can be n, n-1, . . . 2, 1 in descending order of the sum of the characteristic values ​​of the components constituting each basic taste. For example, if the number of clusters is 3, the labels can be 3, 2, 1. The labeling numbers can be determined by the operator of the classification system, or, for example, the number G of the cluster group can be used.

[0051] When the number of reference wines is small, for example, 50 or less, 100 or less, 150 or less, or 200 or less, multiple clusters can be grouped together for each basic taste.

[0052] In the present invention, the wine group classification step (e) may or may not be performed after the clustering step (d). This step involves classifying wines into groups using labeling for each basic taste obtained through the clustering. It is not necessary to classify wines into groups based on the labeling of clusters using all of umami, sweetness, sourness, bitterness, and unpleasant tastes. It is preferable to classify wines into groups based on a combination of labeling values ​​based on at least umami and sweetness. It is more preferable to classify wines into groups based on umami, sweetness, and sourness, and even more preferable to classify wines into groups based on umami, sweetness, sourness, and bitterness. It is also possible to classify wines into groups based on umami, sweetness, sourness, bitterness, and unpleasant tastes.

[0053] A specific method for the above-mentioned group classification is to integrate labeling data for each cluster of umami and sweetness; labeling data for each cluster of umami, sweetness, and sourness; labeling data for each cluster of umami, sweetness, sourness, and bitterness; or data for umami, sweetness, sourness, bitterness, and unpleasant components, sort the data by value in Excel, and arrange and classify them in order. Such group classification can be performed in the group classification unit 50, and the group classification information can also be stored in the group classification data storage unit 74.

[0054] The comment creation step (f) is not particularly limited as long as it is a step of creating a comment by associating the characteristics of each cluster group obtained by clustering and the taste information of each group obtained by group classification with each taste of the wine for which a comment is to be created, and the comment can be created by the comment creation unit 60, and the created comment can be stored in the comment data storage unit 75. Price information for each wine can also be added.

[0055] Furthermore, for each cluster that has undergone the above labeling, a factor analysis can be performed to determine in what respects reference samples classified into the same cluster are similar. Such factor analysis is not particularly limited as long as it is a known factor analysis method, but for example, it can be performed by analyzing the commonalities of samples belonging to each cluster using human intuition or a statistical analysis method based on human intuition for the classified clusters, and then generating evaluation comments using, for example, AI.

[0056] The evaluation comments created in the comment creation step can also be stored in the comment data storage unit 75, and an explanation of the overall evaluation of each wine can be presented via the comment creation unit.

[0057] In the wine classification method of the present invention, a prediction model is constructed using random forests, a support vector machine (SVM), which is a pattern recognition model using supervised learning, and / or a neural network, using data related to the clustering of each of the basic tastes and / or data related to the grouping of reference wines based on the cluster groups of each of the basic tastes as training data, and component profile data of a new wine presented by a user is used as input data. Based on the component data of the new wine, clustering is performed in accordance with the wine classification method described above, and the group or group to which the wine belongs is predicted based on the results, thereby enabling evaluation.

[0058] In addition, the training data can also be constructed as a regression-type machine learning model that allows feedback (recursive input). For example, the predicted component profile data and classification results of the new wine can be recorded as new reference wine data along with comments, thereby creating new reference wine data in the present invention. As a result, by generating machine learning data that improves classification accuracy, more advanced and comprehensive wine data can be collected.

[0059] The present invention will be explained in more detail below with reference to examples, but the technical scope of the present invention is not limited to these examples. [Example]

[0060] [Reference example 1] Analysis of 53 reference wines using a triple quadrupole mass spectrometer (LCMS) TM Cell culture profiling was performed using the -8050 (Shimadzu Corporation) and LC / MS / MS method package, and quantification was performed to create profile data for each wine. All analyzed components were then classified using unsupervised machine learning. Hierarchical cluster analysis was performed, but the classification was complex and no clear characteristics could be identified (Figure 1). Due to the large number of samples and component items, the non-hierarchical clustering method, k-means, was used, with instructions to classify the samples into six clusters. The results are shown in the graph in Figure 2.

[0061] The 53 reference wines were prepared by removing coagulated material from the samples using an ultrafiltration filter and then diluting them 20 times with water. TM Analysis was performed using the -8050. When the above reference wine was stored in a sealed container at 4°C, no significant differences in detection results were observed depending on the time of measurement. However, when wine that had been left in the bottle and stored at room temperature was analyzed, differences in measurement values ​​were observed depending on the time of measurement. Therefore, wine that had been filled to the brim in a 15 mL polypropylene tube, sealed with parafilm, and stored at 4°C was used.

[0062] As is clear from Figure 2, the average analytical values ​​of each component for the 53 reference wines were calculated and visualized in a stacked graph, but it was not possible to obtain a classification that would allow us to grasp the characteristics of the wines. The vertical axis of the graph represents the natural logarithm (ln) of the numerical peak values ​​of each component analyzed by LCMS.

[0063] [Example 1] Next, from the profile data of the reference wines, the components that make up the umami, sweetness, acidity, and bitterness of each wine were selected, and clustering was performed using unsupervised learning for each of the umami, sweetness, acidity, and bitterness components.

[0064] Numerical values ​​of the components constituting the basic tastes of umami, sweetness, astringency, and sourness were extracted from the components constituting the profile data for each of the 53 reference wines created in Reference Example 1 above.

[0065] The components that make up the umami taste were glutamic acid, aspartic acid, and succinic acid. The components that make up the sweet taste were threonine, alanine, proline, serine, ornithine, glycine, and hexose. The components that make up the sour taste were lactic acid, gluconic acid, malic acid, citric acid, and tartaric acid. The components that make up the bitter taste were arginine, phenylalanine, tyrosine, histidine, lysine, leucine, isoleucine, methionine, tryptophan, valine, nicotinic acid, nicotinamide, pantothenic acid, hypoxanthine, 4-aminobenzoic acid, and riboflavin.

[0066] Figure 3 is a graph showing the results of clustering performed on umami components with the number of clusters n set to 3. Cluster 3 is a group with a high total content of umami components, high contents of aspartic acid and glutamic acid, and low content of succinic acid. Cluster 2 is a group with an intermediate content of umami components and a good balance of aspartic acid, glutamic acid, and succinic acid. Cluster 1 is a group with a high content of succinic acid, about twice the content of each of aspartic acid and glutamic acid.

[0067] Figure 4 is a graph showing the results of clustering sweet components when the number of clusters, n, was set to 3. Cluster 3 contains a high total amount of sweet components, with equal amounts of threonine, alanine, and proline, followed by serine, and small amounts of ornithine and glycine. Cluster 2 contains the most proline, followed by threonine, alanine, serine, ornithine, and glycine. Cluster 1 contains equal amounts of proline and threonine, followed by alanine, serine, ornithine, and glycine. In addition, in both groups, the amount of hexose was so small that it could be ignored.

[0068] Figure 5 is a graph showing the results of clustering for acidic components when the cluster number n was set to 3. Cluster 3 contained a high total amount of acidic components, with a very high content of lactic acid, followed by gluconic acid, malic acid, and citric acid. Cluster 2 contained a lower total amount of acidic components than Cluster 3, but the proportion of acidic components was roughly the same as Cluster 3. It was determined that Clusters 2 and 3 were likely subjected to MLF and could be classified as the same group. However, the distinction between Clusters 2 and 3 may become clearer if the number of reference wines is increased, so they are listed separately. On the other hand, Cluster 1 contained a similar amount of acidic components to Cluster 2, but malic acid dominated the majority of the components.

[0069] Cluster 3 of the sour components has a mellow, creamy taste, Cluster 2 has a slightly high citric acid content and a slightly sour taste, and Cluster 1 has a high citric acid content and a sharp sour taste.

[0070] Figure 6 is a graph showing the results of clustering performed on bitter components with the cluster number n set to 3. Cluster 3 contained a high content of bitter components, and by component, it contained a very high content of arginine, followed by phenylalanine, tyrosine, and histidine, with small amounts of lysine, leucine, isoleucine, methionine, tryptophan, nicotinamide, and valine. Cluster 2 contained a high content of phenylalanine, followed by tyrosine and histidine, and was characterized by containing arginine, lysine, and leucine, with small amounts of isoleucine, nicotinamide, pantothenic acid, hypoxanthine, 4-aminobenzoic acid, valine, etc. Cluster 1 contained a low content of bitter components, but was characterized by a high content of phenylalanine, followed by tyrosine and histidine, with small amounts of arginine, isoleucine, nicotinamide, pantothenic acid, hypoxanthine, 4-aminobenzoic acid, valine, etc.

[0071] (group classification) Based on the clustering results for each basic taste thus far, the data was integrated to classify the wines into groups. One method for classifying the wines into groups is to sort them by value in Excel, arrange them in order, and then classify them by the total value. As the number of reference wines in this study was not very large at 53, the number of groups was conveniently divided into 11 groups (categorized as Groups in Tables 1 and 2), but the number of groups can be increased if the number of reference wines increases. An example of the sorted table is shown in Table 1 below. Table 2 also shows a list of the prices of each wine and their average prices.

[0072] [Table 1]

[0073] [Table 2]

[0074] As can be seen from Table 2, there was not much difference in average price among Groups 1 to 11, but Groups 1 and 2 contained some expensive wines. For example, 2021013 and 2021014 also received high ratings in taste evaluations, which is reflected in their high prices. Therefore, wines belonging to Groups 1 and 2 that received high scores for each basic taste are likely to be highly rated in taste evaluations, and it can be determined that these ratings should also be reflected in their prices.

[0075] Next, using the group classification of this reference wine as a training value, we constructed group prediction models using three types of methods: random forest, SVM, and neural network, and predicted the group of the test sample wine. The test sample was also analyzed for its components using a triple quadrupole mass spectrometer (LCMS). TM The data was analyzed using a Shimadzu LC-8050 and the LC / MS / MS method package Cell Culture Profiling, and quantified to create profile data for each wine. While predictions were possible using all of the above prediction models, the classification results were obtained using the SVM prediction model, as shown in Table 3.

[0076] [Table 3]

[0077] (result) The above G2021-0031 to 0036 wines are commercially available, while the others are not. By classifying the taste into groups, it was possible to confirm which commercially available wines the wines tasted similar to. It was also confirmed that the price can be compared with wines with similar characteristics and used as a reference value. For example, wine G2021-0027 is not yet on the market, but it can be suggested to traders that it may be priced slightly higher.

[0078] [Example 2] As described above, we have evaluated the taste based on the components of wine. However, since it is also important to evaluate the taste of wine in the context of the so-called "marriage" with the taste of the food itself in a meal, we attempted to further improve the accuracy of the generated learning model by incorporating additional indicators, taking into account not only the threshold values ​​of each component but also the synergistic effect of taste when combined with a meal.

[0079] [taste] The reference wine stored at 4°C was analyzed using the LCMS™-8050 described in the Reference Example above. In Example 1, glutamic acid, aspartic acid, and succinic acid were selected and analyzed as components that contribute to umami. This time, however, a correlation matrix was created to determine whether strong correlations (multicollinearity) existed between component variables. Correlations were calculated using pandas, a library for data manipulation and analysis. The results were displayed in heatmap format using matplotlib and seaborn (Figure 9(a)). The correlation between glutamic acid and aspartic acid for each cluster was also plotted (Figure 9(b)).

[0080] As is clear from Figures 9(a) and (b), there is a strong correlation between glutamate and aspartate, with glutamate concentrations tending to increase as the aspartate concentration increases. While the taste threshold for glutamate is said to be 30 mg / 100 mL, the glutamate content in wine has been reported to be 0 mg to 16.8 mg / 100 mL (Food Chem. 2021 Oct 30;360:128971). Therefore, it is unlikely that the umami flavor of glutamate is perceived in wine alone. However, when combined with inosinic acid, its umami flavor is reported to be 8 to 100 times stronger due to a synergistic effect (Advances in Biochemistry and Physiology (eds. Filer, LJ, Jr. et al.) 35-54).

[0081] As mentioned above, there is a considerable correlation between glutamic acid and aspartic acid. However, it has been reported that the threshold value for aspartic acid is 160 mg / 100 mL (Nihon Nogeikagaku Kaishi 65 (2), 163-169, 1991), and that the umami flavor is only 8% of that of glutamic acid (Advances in Biochemistry and Physiology (eds. Filer, LJ, Jr. et al.) 35-54). From the perspective of food pairing, aspartic acid was excluded and glutamic acid was selected as the umami component in the present invention.

[0082] On the other hand, as is clear from Figure 9(a), it was confirmed that there was little correlation between glutamic acid and succinic acid. Based on the above, we decided to analyze the components that make up the umami flavor in wine based on the peak concentrations of glutamic acid and succinic acid.

[0083] (Determining the number of umami clusters) When performing cluster analysis, the elbow method (Figure 10(a)) and silhouette method (Figure 10(b)) were used to predict appropriate clusters, and seven clusters were predicted to be appropriate. Therefore, we decided to classify umami into seven clusters.

[0084] (Characteristics of each umami cluster) The K-means method was used to divide the clusters into seven groups. The scatter plot in Figure 11(a) shows that the plots for each cluster are color-coded and do not intersect. The silhouette score, a quantitative evaluation index, was calculated based on the silhouette method graph in Figure 10(b) to determine how well each data point is separated from its own cluster and other clusters. The average silhouette coefficient was calculated to be 0.403. A mean value of 0.4 or higher is considered to be "good." Calculating the feature values ​​for each cluster reveals that each cluster is well characterized, as shown in Figures 12(a) and 12(b). The values ​​on the X and Y axes are the peak values ​​quantified in the component analysis of the reference wine conducted in Reference Example 1. This was also confirmed in the case of red wine alone (Figure 11(b)) and white wine and sparkling wine (Figure 11(c)). It was shown that there is an inverse correlation between glutamic acid and succinic acid in red wine. It was confirmed that wines with a high level of glutamic acid had less succinic acid, and wines with a high level of succinic acid had less glutamic acid.

[0085] In the above cluster classification, cluster numbers were automatically assigned, so the amounts of glutamic acid and succinic acid in each cluster were confirmed and each cluster was assigned to a new cluster group (G) using the peak value data quantified in the component analysis of the reference wine performed in Reference Example 1. Graphs showing the peak values ​​for each cluster group (G) for glutamic acid ( FIG. 12(a) ) and succinic acid ( FIG. 12(b) ) were created. Characteristics of each cluster G7: Cluster 3: Very high in glutamic acid, moderate in succinic acid G6: Cluster 1: High in glutamic acid, low in succinic acid G5: Cluster 5: High in glutamic acid, medium in succinic acid G4: Cluster 0: Medium glutamic acid, high succinic acid G3: Cluster 4: Very low in glutamic acid, very high in succinic acid G2: Cluster 2: Low in glutamic acid, medium in succinic acid G1: Cluster 6: Low in glutamic acid and succinic acid Figure 11(d) shows a redrawn graph of the scatter plot in Figure 11(c) after assigning each cluster group (G). This grouping confirmed that clusters G4 and G3 do not exist in white wine or sparkling wine. For the subsequent scatter plots of sweetness, sourness, and bitterness, the scatter plots with automatically assigned cluster numbers are modified to display the scatter plots with assigned cluster groups (G).

[0086] [sweetness] In Example 1 above, glutamic acid, aspartic acid, and succinic acid were selected and examined as components that make up sweetness. However, this time, whether or not there is a strong correlation (multicollinearity) between component variables was first confirmed by calculating the correlation using pandas to create a correlation matrix, which is used to confirm the strength of the correlation between variables, and displaying it in heat map format using matplotlib and seaborn (Figure 13).

[0087] Psychophysical methods have shown that adding inosinic acid (IMP) to sweet amino acids such as L-alanine and L-serine enhances taste perception (Chemical Senses, Volume 27, Issue 8, October 2002, Pages 739-745). Based on the peak concentrations of sweet components in wine, a correlation matrix was plotted using a heat map for proline, alanine, glycine, threonine, and serine. As shown in Figure 13, proline had no correlation with the other components. The correlation coefficients for alanine, glycine, threonine, and serine were high, at 0.81. Therefore, from the group consisting of alanine, glycine, threonine, and serine, alanine, which is known to give wine a strong sweet taste (Physiol Behav. 1981 Jul;27(1):51-9), was selected, and together with proline, which has a low correlation with the above group, a cluster analysis of sweetness was performed based on the peak values ​​of the concentrations of sweet components in wine.

[0088] (Determining the number of sweetness clusters) The elbow method predicted that the appropriate number of clusters was 3 (data not shown). Note that, since the cluster numbers were automatically assigned in the above cluster classification, the peak value data quantified in the component analysis of the reference wine performed in Reference Example 1 was used to reassign each cluster group (G).

[0089] (Characteristics of each sweetness cluster) The K-means method was used to divide the clusters into three, and to confirm whether the clusters were properly separated, a pair plot of proline and alanine was checked by changing the X and Y axes. The scatter plots of the pair plots in Figures 14(a-1), (a-2), and (b) show that the plots for each cluster are color-coded without overlapping. Calculating the silhouette score yielded an average silhouette coefficient of 0.503. A mean value of 0.4 or higher is considered "good" for the clustering results. To confirm the distribution of the various wines, each group was labeled, and as is clear from Figure 14(b), which shows the correlation between proline and alanine, each cluster was found to have a distinctive distribution.

[0090] We also checked the amount of proline (Figure 15(a)) and alanine (Figure 15(b)) in each cluster. The characteristics of each clustered group of the components that make up umami are shown below. Characteristics of each cluster G3: High in proline and low in alanine G2: Slightly more proline, more alanine G1: Low in proline and alanine

[0091] [acidity] As is clear from the heat map in Figure 16, the correlations between malic acid and lactic acid, and between malic acid and citric acid, were high, and the correlation between lactic acid and citric acid was also somewhat high. However, malic acid, lactic acid, and citric acid are the main acids in wine acidity. Malic acid and lactic acid, in particular, increase or decrease due to the influence of malolactic fermentation, and the results are very important. Therefore, acids were included as analytical parameters. Succinic acid, which is also an umami component, is also perceived as sour, so it was added as an analytical parameter for sourness. Based on the peak concentrations of the acidity components in wine, malic acid, lactic acid, citric acid, and succinic acid, these four components were used for the study. Note that glucuronic acid is primarily found in the juice of noble rot grapes infected with Botrytis cinerea, and therefore was not included in this study because it was deemed inappropriate for evaluating the overall wine as a sugar-sweetness component.

[0092] (Determining the number of sourness clusters) When the elbow method was used to predict the number of clusters, it was found that a cluster count of 5 was appropriate. On the other hand, the silhouette method indicated that a cluster count of 6 was appropriate. To verify this further, the silhouette score was calculated for each cluster count. When the number of clusters was 5, the average silhouette score was 0.39. While close to 0.4, some data points had negative silhouette scores. This suggests that the separation between clusters was insufficient, leading to inappropriate classification. Furthermore, when the number of clusters was changed to 4, 5, and 6, the average silhouette score was 0.41, and all data points had silhouette scores above 0. This confirmed improved separation between clusters and more appropriate classification. For these reasons, we determined that a cluster count of 6 was appropriate.

[0093] (Characteristics of each acidity cluster) When the clusters were divided into six using the K-means method and checked using the scatter plot of the pair plots in Figure 17, some combinations of clusters were seen to be mixed together. However, when the box plots of the features were checked, differences in the distribution of the features were seen for each cluster, and it was confirmed that a certain degree of characterization could be made between the clusters. For this reason, it was determined that the cluster division had been performed appropriately.

[0094] As is clear from Figure 18, cluster G1 (the group on the top left), which has low lactic acid and high malic acid, does not contain any red wines among the reference wines measured in this study, and it is thought that the wines in cluster G1 have not undergone lactic acid fermentation.

[0095] In the above cluster classification, cluster numbers were automatically assigned, so the amounts were confirmed and each cluster group (G) was assigned again using the peak value data quantified in the component analysis of the reference wine performed in Reference Example 1, as shown in Figures 19(a) to (d). The characteristics of each cluster group (G) for the components that make up the sourness are shown below. Characteristics of each cluster G6: Low malic acid, high lactic acid, low citric acid, medium succinic acid, G5: Low malic acid, medium lactic acid, medium citric acid, medium succinic acid G4: Low malic acid, medium lactic acid, low citric acid, medium succinic acid G3: Low malic acid, medium to high lactic acid, medium citric acid, low succinic acid G2: Low malic acid, medium lactic acid, low citric acid, medium to high succinic acid G1: High malic acid, low lactic acid, medium to high citric acid, low to medium succinic acid

[0096] [bitter taste] Regarding bitterness, a correlation matrix used to confirm the strength of correlation between variables was created using 10 amino acids: proline, histidine, arginine, tryptophan, arginine, valine, methionine, tryptophan, phenylalanine, isoleucine, leucine, and lysine, and was plotted in Figure 20 as a heat map.

[0097] As is clear from Figure 20, proline, histidine, tryptophan, and arginine have low correlations with each other, but a correlation was observed in the group consisting of valine, methionine, phenylalanine, isoleucine, leucine, and lysine. On the other hand, arginine, phenylalanine, valine, and leucine have been reported to be primarily associated with bitterness at thresholds in the range of 40 to 90 mg / 100 mL (Food Sci Biotechnol. 2022 May 19;31(7):767-785). Therefore, in addition to proline, histidine, tryptophan, and arginine, we also selected valine and phenylalanine, and performed cluster analysis of the concentrations of bitter components in wine using the peak values ​​of the six amino acid components.

[0098] The elbow method predicted a value of 3 (data not shown).

[0099] (Characteristics of each bitterness cluster) In the above cluster classification, cluster numbers were automatically assigned, and the amount of each component in each cluster was confirmed using the quantified peak value data from the component analysis of the reference wine performed in Reference Example 1. Since there were no major characteristics other than arginine in clusters 0 and 1, these clusters were grouped together and reassigned to each cluster group (G). The characteristics of each clustered group of bitter components are shown below. Characteristics of each cluster G2: Cluster 2: High arginine G1: Cluster 0: Low arginine G1: Cluster 1: Low arginine

[0100] (comprehensive evaluation) Based on the above evaluation, the group classification results were summarized as follows. The results were sorted in descending order of total points. There were only two groups for bitterness, and few wines corresponded to Group 2. Wine contains bitter compounds such as histidine, arginine, valine, methionine, phenylalanine, leucine, and lysine, but all of these were below the threshold of 40-90 mg / 100 mL, and since these were confirmed to be below the threshold, they were not included in the overall evaluation.

[0101] The group points for umami, sweetness, and sourness were summed up and divided into 12 groups of 3 points each from 16 points, as shown in Table 4 below.

[0102] Wines with an umami score of 5 points or more have high glutamic acid, and wines with high sweetness and acidity are seen with a total score of 13 points or more. The price range is 2,800 to 16,400 yen, but the average price is 8,750 yen, so they tend to be rated as high priced. For those with a umami score of 5 or more, the price range was 1,345 yen to 16,400 yen, with an average of 6,523 yen. Wines with umami scores of 4 and 3 points were high in succinic acid, and 8 out of 15 were Japanese wines. Of the 52 wines, 13 were Japanese, and 8 out of 13 were high in succinic acid. Succinic acid pairs well with seafood, confirming that Japanese wines are a good match for Japanese cuisine. On the other hand, there were few Japanese wines with high glutamic acid. Of course, there may not have been enough reference wines and some bias may have been present, but it is highly likely that these ingredients are the reason why Japanese wines are often said to go well with Japanese food.

[0103] [Table 4]

[0104] When examining the relationship between total points and market price (Table 5), there were no wines with low total points but high prices. On the other hand, there were wines with high total points but low market prices. Since wine prices are determined not only by taste but also by brand, there is often no correlation with taste. However, by using the evaluation method of the present invention, it is possible to set a high price based on taste. To confirm this, the points were predicted for wines without market prices using the above grouping as a model. The results are shown in Table 6.

[0105] [Table 5]

[0106] Table 6 includes wines that are not currently on the market, and after analyzing the ingredients, we confirmed that some wines have high total points. This can be used as a useful resource for predicting market prices.

[0107] [Table 6]

[0108] Table 7 below lists the taste characteristics of each group as assessed by a panel of wine experts. This assessment was conducted according to strict rules, such as rinsing the mouth with water to refresh before moving on to another sample, as is done in alcoholic beverage tastings, and involved not only a rigorous evaluation of the aroma, mouthfeel, and throat feel of the wine itself, but also by tasting meat and fish dishes while tasting the wines to assess the effect of food pairing on each taste bud. The results are shown in Table 7 below.

[0109] [Table 7]

[0110] (result) As is clear from Table 7, although the glutamic acid content in wine is below the threshold, when it is tasted with dishes containing inosinic acid such as meat or cheese, the umami flavor is enhanced through a synergistic effect.In addition, it was confirmed that the umami flavor of Groups 3 and 2, which contain a lot of succinic acid, is enhanced when tasted with seafood dishes. With the exception of sweet wines and champagne, the sugar in wine is consumed during alcoholic fermentation, so the sweetness of the sugar should not be perceived, but it is known that some wines have a sweet taste. Measurements confirmed that wines with high levels of proline and alanine are perceived as sweeter and have a richer flavor. Furthermore, proline also has the effect of enhancing the umami flavor of glutamate, and it was confirmed that the umami flavor is enhanced when enjoyed with kelp stock, seafood, or meat. It is well known that malolactic fermentation reduces malic acid, increases lactic acid, and other amino acids, resulting in an overall mellow flavor. In fact, when measured, it was clear whether there was a lot of lactic acid or not. However, although there was little malic acid, there was variation in the concentration of lactic acid. When proline, glutamic acid, or succinic acid is high in a wine, it feels rich, but when the acidity is low, it is unbalanced and does not go well with light dishes. However, there were many wines that could be enjoyed on their own. Wines with a high malic acid content in the measurements had a sharper taste. The bitter components correspond to amino acids produced during the fermentation process, but as shown in Additional Example 2, the measured values ​​for each component were below the threshold, making it difficult to evaluate each component individually.

[0111] [Example 3] [Analysis by component concentration] Until now, measurements have been made based on peak values, but evaluation based on peak values ​​is a relative evaluation, and tends to lack reproducibility when the number of samples increases. Therefore, calculations were made based on the concentration of each component in wine using different samples from those used in the past.

[0112] [taste] As is clear from Figures 22(a) and (b), the succinic acid concentration in both red wine and white wine was 200 mg / L or more, which was above the threshold value.

[0113] As mentioned above, the taste threshold for glutamate is said to be 30 mg / 100 mL, while the glutamate content in wine is reported to range from 0 mg to 16.8 mg / 100 mL. However, when glutamate and inosinic acid are combined, the umami effect of glutamate is enhanced by at least 8 times through a synergistic effect. Therefore, a glutamate concentration of at least 37.5 mg / L (30 mg / 100 mL ÷ 8) is required for the umami effect of glutamate to be expected. Therefore, a line is drawn at the glutamate concentration of 37.5 mg / L in the graphs of Figures 22(a) and (b). It has been confirmed that red wines, particularly those belonging to the G7 and G6 groups, exhibit a synergistic umami effect when consumed with dishes rich in inosinic acid.

[0114] For example, wine No. 306 in the graph shows that both glutamic acid and succinic acid are relatively high. Meanwhile, wine No. 305 also shows a high amount of succinic acid. Both No. 306 and No. 305 are produced by the same winery. This indicates that wines from this winery have a high umami flavor. On the other hand, white wines and sparkling wines generally have glutamic acid levels below 37.5 mg / L, which does not have much of a synergistic effect on umami, and can be expected to have a refreshing taste even when paired with dishes containing inosinic acid.

[0115] Neither the WSET, the world's largest wine education organization, nor the Japan Sommelier Association includes the umami of wine in their evaluation criteria, but the marriage of alcoholic beverages and food is an important factor and is thought to be a particularly useful indicator when evaluating the potential for umami.

[0116] (sweetness) Cluster analysis of proline and alanine was performed as described above. The number of clusters was set to 3, and the results are shown in the scatter plots in Figures 23(a) and (b). There are various reports on the sweetness threshold of proline, but we have set it to 690 mg / L, which is indicated by a red line on the graph.

[0117] The abundance of proline (FIG. 24(a)) and alanine (FIG. 24(b)) in each cluster was confirmed, and each cluster was assigned to a group (G). G1: Cluster 0: Alanine and proline were scarce, and proline was below the threshold. G2: Cluster 2: Proline was very abundant, above the threshold. G3: Cluster 1: Proline was above the threshold, and alanine had a moderate tendency. This result was almost the same as that obtained by analyzing the peak value.

[0118] (summary) - By calculating the concentration instead of the peak value, it was possible to compare with the threshold value. Also, it became clear that measuring by concentration is preferable because it can be saved as an absolute value in the database. -It was found that it is possible to classify the taste of wine by measuring the components, particularly glutamic acid, succinic acid, proline, alanine, malic acid, lactic acid, and citric acid. In particular, there were previously no indicators for evaluating the umami of wine or the sweetness of non-sweet wines, but in fact, umami was present and sweet components that were not sugars could be detected. The glutamate content in wine is below the threshold, so it cannot be evaluated by normal tasting, but can be detected by measuring it. When considering enjoying wine with food, measuring umami is thought to be very useful. -Of course, glycerin and tannins also affect the taste as it passes through the throat, and if aromas are also included, many other factors need to be taken into account. However, complex factors are difficult for the actual end user to understand, so a simple classification is useful. [Industrial Applicability]

[0119] The present invention is based on a wine classification method that places importance on taste, and can suggest to wine traders and consumers the taste and pricing of individual wines, including their compatibility with food. [Explanation of symbols]

[0120] 1. Classification System 4 Pretreatment section 5 Component detection section 10 Data Acquisition Section 20. Digitization Department 30 Data Extraction Section 35 Component determination unit 40 Clustering Department 50 Group classification section 60 Comment Creation Department 70 Data storage unit 71 Profile data storage section 72 Feature value data storage section 73 Cluster Data Storage 74 Group classification data storage section 75 Comment data storage section 80 Information presentation section 810 Display

Claims

1. A method for classifying wines comprising the following (a) to (f) in order: (a) a data acquisition step of acquiring component profile data of a reference wine; (b) a digitization step of converting the content of each component constituting the component profile data of the reference wine acquired in the data acquisition step of step (a) into a characteristic value; (c) a data extraction step of extracting characteristic values ​​of components constituting basic tastes from the characteristic values ​​of each component of the reference wine created in the digitization step of step (b); (d) a clustering step of forming n cluster groups for each basic taste by performing clustering by unsupervised learning for each feature value of the components constituting each basic taste extracted in the extraction step of step (c); (e) a group classification step of classifying the reference wines into groups based on the cluster groups for each basic taste formed in step (d); (f) a comment creation step of associating taste information and creating a comment;

2. 2. The classification method according to claim 1, wherein the basic tastes consist of umami and sweetness.

3. 2. The classification method according to claim 1, wherein the basic tastes consist of umami, sweetness, and sourness.

4. 2. The classification method according to claim 1, wherein the basic tastes consist of umami, sweetness, sourness, and bitterness.

5. 5. The classification method according to claim 2, wherein the components constituting umami include one or more selected from the group consisting of glutamic acid, inosinic acid, aspartic acid, adenylic acid and succinic acid.

6. 5. The classification method according to claim 2, wherein the components constituting umami are glutamic acid and succinic acid.

7. 5. The classification method according to claim 2, wherein the components constituting sweetness include one or more selected from the group consisting of proline, glycine, alanine, threonine, serine, ornithine, and asparagine.

8. The classification method according to any one of claims 2 to 4, wherein the components constituting sweetness are proline and alanine.

9. 5. The classification method according to claim 3, wherein the components constituting the sour taste include one or more selected from the group consisting of tartaric acid, malic acid, citric acid, succinic acid, lactic acid, glycyl glutamine acetate, seryl glutamine, and glycyl asparagine.

10. 5. The classification method according to claim 3, wherein the components constituting the sour taste are malic acid, citric acid, succinic acid, and lactic acid.

11. A wine classification system that outputs comments about wine, including classification results using wine component profile data, comprising the following (A) to (F). (A) a data acquisition unit for acquiring component profile data of a reference wine; (B) a digitization unit that digitizes the content of each component constituting the component profile data of the reference wine acquired by the data acquisition unit into a feature value; (C) a data extraction unit that extracts the quantified data of components constituting the basic tastes from the quantified profile data; (D) a clustering unit that performs clustering for each of the quantified data of the components constituting each basic taste by unsupervised learning to form n clusters for each basic taste; (E) a group classification unit that performs group classification of reference wines based on n clusters formed for each basic taste; (F) a comment creating unit that associates taste information and creates a comment;

12. 12. The classification system according to claim 11, wherein the basic tastes consist of umami and sweetness.

13. 12. The classification system of claim 11, wherein the basic tastes consist of umami, sweetness, and sourness.

14. 12. The classification system of claim 11, wherein the basic tastes consist of umami, sweetness, sourness, and bitterness.

15. The classification system according to any one of claims 12 to 14, characterized in that the components constituting umami include one or more selected from glutamic acid, inosinic acid, aspartic acid, adenylic acid and succinic acid.

16. The classification system according to any one of claims 12 to 14, wherein the components that constitute umami are glutamic acid and succinic acid.

17. The classification system according to any one of claims 12 to 14, characterized in that the components constituting sweetness include one or more selected from proline, glycine, alanine, threonine, serine, ornithine, and asparagine.

18. The classification system according to any one of claims 12 to 14, wherein the components constituting sweetness are proline and alanine.

19. The classification system according to claim 13 or 14, characterized in that the components constituting the sour taste include one or more selected from the group consisting of tartaric acid, malic acid, citric acid, succinic acid, lactic acid, glycyl glutamine acetate, seryl glutamine, and glycyl asparagine.

20. 15. The classification system according to claim 13 or 14, wherein the components that constitute sourness are malic acid, citric acid, succinic acid, and lactic acid.

21. A method for collecting data for wine classification, which uses wine component profile data presented by a user as input data, classifies wines according to the wine classification method described in any one of claims 1 to 4, and records the results as reference data.

Images