Methods for predicting and identifying the presence of yeast extract in soy sauce

Abstract

This invention belongs to the fields of analytical chemistry and mathematical modeling. Specifically, it is a method for predicting and identifying whether soy sauce contains yeast extract, comprising the following steps: 1) using different types of soy sauce as modeling samples; 2) pre-treating the modeling samples by decolorization; 3) performing spectral scanning on the modeling samples using Raman spectroscopy to obtain the original Raman spectral information; 4) pre-processing the Raman spectral data obtained in step 3); 5) establishing a sparse logistic regression model through multiple sampling and recording important wavenumbers; 6) statistically analyzing the importance of wavenumbers; 7) selecting important wavenumbers; 8) establishing a prediction model using TabPFN; 9) using the established prediction model to predict and determine whether yeast extract has been added to the soy sauce sample. Compared with existing technologies, this invention can quickly and accurately identify whether light soy sauce, dark soy sauce, etc., contain yeast extract, and the identification process is simple and fast, with short analysis time, high accuracy, and high practicality.

Description

[Technical Field]

[0001] This invention belongs to the fields of analytical chemistry and mathematical modeling. Specifically, it is a method for predicting and identifying whether soy sauce contains yeast extract based on Raman spectroscopy and Monte Carlo-sparse logistic regression combined with tabular probabilistic feedforward network algorithm. [Background Technology]

[0002] Soy sauce is a nutritious condiment with a unique flavor and color, widely loved in China and an indispensable part of people's dining tables. As people's living standards gradually improve, consumer demand for soy sauce is trending towards higher quality, with greater emphasis on additive-free ingredients and taste. Green and organic, delicious flavor, and a lingering aftertaste are key indicators. Additive-free soy sauce refers to soy sauce produced without any food additives, using only raw and auxiliary materials with food ingredient properties.

[0003] Yeast extract is a natural condiment obtained from yeast using modern enzymatic hydrolysis technology. Since it does not involve artificial chemical synthesis, it can be used in additive-free soy sauce. Due to its rich meaty aroma, delicious flavor, and healthy properties, it is increasingly widely used in the condiment industry. When combined with modern brewing processes, it can produce a synergistic effect that enhances flavor and freshness. Yeast extract can increase the amino acids, peptides, organic acids, and flavor compounds in soy sauce, compensating for the shortcomings of short fermentation times, such as a bland taste and unbalanced flavors, thus improving the umami and mouthfeel of the soy sauce.

[0004] In the current market, some manufacturers, in pursuit of short-term profits, falsely add yeast extract, or omit yeast extract or add artificially synthesized additives to pass off as soy sauce with added yeast extract. This not only lowers the quality of soy sauce but also poses food safety risks, bringing a negative impact to the soy sauce market.

[0005] Raman spectroscopy is a non-destructive spectroscopic technique that analyzes the chemical composition and structure of a sample by measuring the scattering spectrum after the sample interacts with light. When a laser beam strikes a sample, most of the light is scattered at the same frequency (Raman scattering), but a small portion is scattered at different frequencies. These frequency variations are related to the vibrations and rotations of molecules in the sample and can therefore be used to identify the sample's components. However, in practical food testing, the complexity of food components and the mutual interference between Raman signals from different compounds pose challenges to the application of Raman spectroscopy in the analysis of complex samples. In recent years, the application of Raman spectroscopy combined with machine learning in the field of food testing has made some breakthroughs, and the application of Raman spectroscopy is attracting increasing attention, promising to enable Raman spectroscopy-based analytical detection techniques to be applied to complex food samples.

[0006] The rapid development of technology has led to the widespread application of data-driven algorithms, such as deep learning, across various fields. While general deep learning algorithms have high requirements for data volume and computing hardware, the sample size available for model training in food analysis is typically small, and the algorithm's response speed is critical, making general deep learning algorithms unsuitable. Tabular Probabilistic Feedforward Networks (TabPFNs) are a special deep learning method for tabular data, with its main advantages being efficiency and applicability to small datasets. The TabPFN architecture fully leverages the advantages of probabilistic modeling and feedforward neural networks. Raman spectroscopy data is often tabular, thus TabPFN has great potential in Raman spectroscopy-based data analysis. However, the number of variables (wavenumber points) in Raman spectroscopy is usually greater than the number of samples; therefore, dimensionality reduction, i.e., variable selection, is necessary before using TabPFN.

[0007] Sparse Logistic Regression is a machine learning method that combines logistic regression with sparsity constraints. It is particularly effective when dealing with high-dimensional data because it can automatically select the most important features, thereby reducing model complexity and improving generalization ability. [Summary of the Invention]

[0008] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for predicting and identifying whether soy sauce contains yeast extract. This method can quickly and accurately identify whether light soy sauce, dark soy sauce, and other soy sauces contain yeast extract. The identification process is simple and fast, the analysis time is short, the accuracy is high, and the practicality is high.

[0009] To achieve the above objectives, a method for predicting and identifying the presence of yeast extract in soy sauce is designed, comprising the following steps:

[0010] 1) Different types of soy sauce were used as modeling samples;

[0011] 2) Perform decolorization pretreatment on the modeling samples;

[0012] 3) The modeled sample obtained in step 2) is subjected to spectral scanning using Raman spectroscopy to obtain the original Raman spectral information;

[0013] 4) Preprocess the Raman spectral data obtained in step 3);

[0014] 5) Establish a sparse logistic regression model through multiple samplings and record important wavenumbers;

[0015] 6) The importance of statistical wavenumbers;

[0016] 7) Select important wavenumbers;

[0017] 8) Use TabPFN to build a predictive model;

[0018] 9) Use the prediction model established in step 8) to predict and determine whether yeast extract has been added to the soy sauce sample.

[0019] Further, in step 1), the sample is a soy sauce sample containing yeast extract or not containing yeast extract, and the soy sauce sample includes light soy sauce and dark soy sauce.

[0020] Further, in step 2), the decolorization pretreatment of the modeling sample is as follows: transfer 2.5 mL of soy sauce sample into a centrifuge tube, add 7.5 mL of water and 4 g of activated carbon, vortex mix, sonicate at 40 degrees Celsius for 30 minutes, centrifuge at 9000 rpm for 15 minutes, and filter the supernatant using a 0.45 μm polytetrafluoroethylene filter membrane.

[0021] Furthermore, in step 3), the instrument used is an Agilent Resolve handheld Raman spectrometer with an incident depth set to 1.5.

[0022] Further, in step 4), the preprocessing steps for the Raman spectral data are as follows: let the number of samples be n, the number of wavenumber points be p, and each spectrum correspond to a label; if yeast extract is added, the spectral label value is 1, otherwise it is 0; set the total number of Monte Carlo samples N and the sampling probability α.

[0023] Further, in step 5), in a Monte Carlo sampling, α samples are selected from all Raman spectral data to establish a sparse logistic regression model; the training data are calculated using the cyclic gradient descent method, and the regression coefficients of the sparse logistic regression model are output. The dimension of the regression coefficients is equal to p, which is consistent with the number of wavenumber points in the Raman spectral data. The Raman spectral wavenumbers represented by the coefficients that are not zero are recorded.

[0024] Further, in step 6), step 5) is repeated until the total number of samples reaches N. The ratio of the number of times each Raman spectral wavenumber is not zero in the regression coefficient of sparse logistic regression after N samplings to N is calculated. The larger the ratio, the greater the importance of the wavenumber.

[0025] Furthermore, in step 7), the wavenumbers are sorted in descending order of importance, and n wavenumbers are selected from largest to smallest, consistent with the number of samples.

[0026] Further, in step 8), the sample dimension after processing in steps 1 to 5 is n rows and n columns. At this time, the data is used as the independent variable and whether yeast extract is added is used as the dependent variable to establish a binary classification prediction model using TabPFN.

[0027] Further, in step 9), for a new soy sauce sample, its Raman spectrum is collected and the same spectral variable as in step 7) is selected. Then, it is substituted into the TabPFN prediction model in step 8) to obtain the prediction result of whether yeast extract has been added to the sample.

[0028] Compared with existing technologies, this invention provides a method for predicting and identifying the presence of yeast extract in soy sauce samples after decolorization, acquisition of Raman spectral characteristic data, and the use of a model established by Monte Carlo-sparse logistic regression combined with a tabular probabilistic feedforward network algorithm. This method is applicable to various types of soy sauce, including light soy sauce, dark soy sauce, brewed soy sauce, and blended soy sauce. The identification process is simple and fast, and the results are accurate, filling a technological gap in the market and meeting market testing needs. In summary, this invention can quickly and accurately identify whether soy sauces such as light and dark soy sauce contain yeast extract. The identification results are consistent with the actual situation, with short analysis time, high accuracy, and high practicality, making it worthy of application and promotion. [Image Description]

[0029] Figure 1 This is a flowchart illustrating the present invention;

[0030] Figure 2 This is a schematic diagram of variable selection in the TabPFN model of the present invention. [Detailed Implementation]

[0031] This invention belongs to the fields of analytical chemistry and mathematical modeling. Specifically, it is a method for identifying yeast extract in soy sauce based on Raman spectroscopy and Monte Carlo-sparse logistic regression combined with tabular probabilistic feedforward network algorithm. The method acquires spectra based on Raman spectroscopy, combines sparse logistic regression and Monte Carlo sampling for variable selection, and selects a number of variables smaller than the number of samples used for subsequent analysis. Then, TabPFN is used to further model and analyze the data.

[0032] The basic principle of this invention is as follows: Soy sauce samples are decolorized, Raman spectral characteristic data are collected, and a Monte Carlo-sparse logistic regression combined with a tabular probabilistic feedforward network algorithm is used to predict and identify whether yeast extract is present in the soy sauce. The specific steps are as follows:

[0033] 1) Different types of soy sauce were used as modeling samples;

[0034] 2) Perform decolorization pretreatment on the modeling samples;

[0035] 3) Raman spectroscopy was used to perform spectral scanning on the modeled sample to obtain the original Raman spectral information;

[0036] 4) Preprocess the Raman spectral data obtained in step 3);

[0037] 5) Establish a sparse logistic regression model through multiple samplings and record important wavenumbers;

[0038] 6) The importance of statistical wavenumbers;

[0039] 7) Select important wavenumbers;

[0040] 8) Use TabPFN to build a predictive model;

[0041] 9) Use the prediction model established in step 8) to predict and determine whether yeast extract has been added to the unknown soy sauce sample.

[0042] In step 1), the sample is a soy sauce sample containing yeast extract or not containing yeast extract. It should be understood that the soy sauce sample includes light soy sauce and dark soy sauce.

[0043] In step 2), the decolorization pretreatment of the modeling sample is as follows: accurately transfer 2.5 mL of soy sauce sample into a centrifuge tube, add 7.5 mL of water and 4 g of activated carbon, vortex mix, sonicate at 40 degrees Celsius for 30 minutes, centrifuge at 9000 rpm for 15 minutes, and filter the supernatant using a 0.45 μm polytetrafluoroethylene filter membrane.

[0044] In step 3), the instrument used is an Agilent Resolve handheld Raman spectrometer with an incident depth set to 1.5.

[0045] In step 4), the preprocessing steps for the Raman spectral data are as follows: Let the number of samples be n and the number of wavenumber points be p; each spectrum corresponds to a label; if yeast extract is added, the spectral label value is 1, otherwise it is 0; set the total number of Monte Carlo samples N and the sampling probability α.

[0046] In step 5), in a Monte Carlo sampling, α samples are selected from all Raman spectral data to establish a sparse logistic regression model; the training data are calculated using the cyclic gradient descent method, and the regression coefficients of the sparse logistic regression model are output. The dimension of the regression coefficients is equal to p, which is consistent with the number of wavenumber points in the Raman spectral data. The Raman spectral wavenumbers represented by the non-zero regression coefficients are recorded.

[0047] In step 6), repeat step 5) until the total number of samples reaches N; calculate the ratio of the number of times each Raman spectral wavenumber is not zero in the regression coefficient of sparse logistic regression after N samplings to N. The larger the ratio, the greater the importance of the wavenumber.

[0048] In step 7), the wavenumbers are sorted in descending order of importance, and n wavenumbers are selected from largest to smallest, consistent with the number of samples.

[0049] In step 8), the sample dimension after processing in steps 1 to 5 is n rows and n columns. At this time, the data is used as the independent variable and whether yeast extract is added is used as the dependent variable. A binary classification prediction model is established using TabPFN.

[0050] In step 9), for a new soy sauce sample, its Raman spectrum is collected and the same spectral variable as in step 7) is selected. Then, it is substituted into the TabPFN prediction model in step 8) to obtain the prediction result of whether yeast extract has been added to the sample.

[0051] The present invention will be further described below with reference to specific embodiments:

[0052] Example 1: Identification of yeast extract in 10 samples of a certain brand of light soy sauce

[0053] (1) Decolorization pretreatment was performed on 10 light soy sauce samples. 2.5 mL of light soy sauce sample was accurately transferred into 10 centrifuge tubes, 7.5 mL of water and 4 g of activated carbon were added, and the mixture was vortexed and then sonicated at 40 degrees Celsius for 30 minutes. The mixture was centrifuged at 9000 rpm for 15 minutes, and the supernatant was filtered through a 0.45 μm polytetrafluoroethylene filter membrane.

[0054] (2) The instrument used was an Agilent Resolve handheld Raman spectrometer. Raman spectra of 10 light soy sauce sample solutions were collected at an incident depth of 1.5.

[0055] (3) The Raman spectra of 10 light soy sauce samples were processed using spectral variables consistent with the TabPFN prediction model established in this invention, and then substituted into the TabPFN prediction model to obtain the sample prediction results. "0" indicates no yeast extract, and "1" indicates yeast extract. After verification, the prediction results are consistent with the actual results, as shown in the table below.

[0056]

[0057]

[0058] Contents not described in detail in this specification are existing technologies known to those skilled in the art and will not be elaborated upon here. This invention is not limited to the above-described embodiments; any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of this invention should be considered equivalent substitutions and are included within the scope of protection of this invention.

Claims

1. A method for predicting and identifying whether soy sauce contains yeast extract, characterized in that, Includes the following steps: 1) Different types of soy sauce were used as modeling samples; 2) Perform decolorization pretreatment on the modeling samples; 3) The modeled sample obtained in step 2) is subjected to spectral scanning using Raman spectroscopy to obtain the original Raman spectral information; 4) Preprocess the Raman spectral data obtained in step 3); 5) Establish a sparse logistic regression model through multiple samplings and record important wavenumbers; 6) The importance of statistical wavenumbers; 7) Select important wavenumbers; 8) Use TabPFN to build a predictive model; 9) Use the prediction model established in step 8) to predict and determine whether yeast extract has been added to the soy sauce sample.

2. The method as described in claim 1, characterized in that: In step 1), the sample is a soy sauce sample containing yeast extract or not containing yeast extract, and the soy sauce sample includes light soy sauce and dark soy sauce.

3. The method as described in claim 1, characterized in that, In step 2), the decolorization pretreatment of the modeling sample is as follows: transfer 2.5 mL of soy sauce sample into a centrifuge tube, add 7.5 mL of water and 4 g of activated carbon, vortex mix, sonicate at 40 degrees Celsius for 30 minutes, centrifuge at 9000 rpm for 15 minutes, and filter the supernatant using a 0.45 μm polytetrafluoroethylene filter membrane.

4. The method as described in claim 1, characterized in that: In step 3), the instrument used is an Agilent Resolve handheld Raman spectrometer with an incident depth set to 1.

5.

5. The method as described in claim 1, characterized in that, In step 4), the preprocessing steps for the Raman spectral data are as follows: Let the number of samples be n, the number of wavenumber points be p, and each spectrum correspond to a label; if yeast extract is added, the spectral label value is 1, otherwise it is 0; set the total number of Monte Carlo samples N and the sampling probability α.

6. The method as described in claim 5, characterized in that: In step 5), in a single Monte Carlo sampling, α samples are selected from all Raman spectral data to establish a sparse logistic regression model. The training data is calculated using the cyclic gradient descent method, and the regression coefficients of the sparse logistic regression model are output. The dimension of the regression coefficients is equal to p, which is consistent with the number of wavenumber points in the Raman spectral data. The Raman spectral wavenumbers represented by the non-zero regression coefficients are recorded.

7. The method as described in claim 6, characterized in that: In step 6), repeat step 5) until the total number of samples reaches N. Calculate the ratio of the number of times each Raman spectral wavenumber is not zero in the regression coefficient of sparse logistic regression after N samplings to N. The larger the ratio, the greater the importance of that wavenumber.

8. The method as described in claim 7, characterized in that: In step 7), the wavenumbers are sorted in descending order of importance, and n wavenumbers are selected from largest to smallest, consistent with the number of samples.

9. The method as described in claim 8, characterized in that: In step 8), the sample dimension after processing in steps 1 to 5 is n rows and n columns. At this time, the data is used as the independent variable and whether yeast extract is added is used as the dependent variable. A binary classification prediction model is established using TabPFN.

10. The method as described in claim 9, characterized in that: In step 9), for a new soy sauce sample, its Raman spectrum is collected and the same spectral variable as in step 7) is selected. Then, it is substituted into the TabPFN prediction model in step 8) to obtain the prediction result of whether yeast extract has been added to the sample.

Images