A method for rapid detection of 10 yeast species and its application in fermented foods

Abstract

The present invention discloses a method for rapid detection of 10 types of yeast and its application in fermented foods, belonging to the field of microbial detection and ecological applications. The method of the present invention is used to perform Raman detection on 10 types of yeast, and a training model is established using characteristic bands combined with XGboost to identify unknown strains. The method is applied to Daqu to prepare a Daqu suspension sample by gradient centrifugation, homogenization or vortexing, and Raman detection is performed on single cells. The established yeast training model is used to perform machine learning on the Raman spectrum of unknown bacteria to predict the type of strain, thereby realizing real-time, rapid and label-free detection of yeast in Daqu and determining its yeast type, and the detection and analysis time is greatly shortened.

Description

Technical Field

[0001] The invention relates to a method for quickly detecting 10 kinds of yeast and application thereof in fermented foods, belonging to the field of microbial detection and ecological application. Background Art

[0002] Saccharomyces is a unicellular fungus with oval or cylindrical cells that primarily reproduce asexually. Extensive research has been conducted on the community structure and diversity within high-temperature daqu (white liquor) and fermented grains. Yeasts are the primary bacteria responsible for alcohol production and flavor development in the early stages of fermentation. During fermentation, yeasts produce alcohol-enhancing enzymes that act on glucose to produce ethanol through glycolysis and anaerobic degradation of pyruvate. The esterases they secrete, in turn, utilize substrates such as ethanol and acids in the fermented grains to synthesize various flavor compounds through esterification. Therefore, rapid and reliable identification of yeast species during the fermentation of fermented foods is essential. By identifying yeast species and changes in their content, fermentation progress and abnormalities can be monitored, thereby ensuring product quality during food production.

[0003] Currently, rapid yeast identification mainly involves standard plate colony counts, polymerase chain reaction (PCR), and enzyme-linked immunosorbent assay (ELISA). Colony counts are a standard method for bacterial detection, but they require 24 to 72 hours, are labor-intensive, and time-consuming. ELISA and PCR offer high sensitivity, shortening detection times to a few hours. However, cross-reactions between antibodies in ELISA reactions can occur, protein samples have a short retention time, and PCR is prone to false positives for bacteria. Furthermore, both require multiple pretreatment steps, precision instruments, and specialized operations. Therefore, yeast detection based on plate colony counts, ELISA, and PCR techniques is slow and difficult. To overcome these shortcomings, it is necessary to develop methods for detecting yeast and spores with shorter analysis times and higher sensitivity.

[0004] Raman spectroscopy is a type of scattering spectrum. Based on the Raman scattering effect discovered by Indian scientist C.V. Raman, Raman spectroscopy analyzes scattered light at frequencies different from the incident light to obtain information about molecular vibrations and rotations, which is then applied to molecular structure research. Generally speaking, a Raman spectrum is a unique chemical fingerprint for a specific molecule or material. In recent years, Raman spectroscopy has been found to have applications in biology, including analysis of single cells. The Raman spectrum of a single cell is a superposition of the vibrational modes of its intracellular components, composed of Raman peaks corresponding to each type of chemical bond. This provides multidimensional information about the composition and relative abundance of metabolites within a single cell. Raman spectroscopy can be used to differentiate between bacteria and fungi and identify storage compounds and cytochromes within individual microbial cells. Raman spectroscopy enables label-free, rapid, and in situ detection of microorganisms at the species level. This approach identifies strain types with high sensitivity, specificity, and accuracy. Compared to plate culture methods, it can also be used to analyze uncultured microorganisms. As recorded in the authorized patents with publication numbers CN104136908B and CN102272585B, Raman spectroscopy technology processes culture and blood samples through sealed separation devices, thereby enabling effective Raman analysis. However, for special food fermentation process samples, such as the solid raw materials of barley, wheat and other raw materials where microorganisms are attached to Daqu, and the vegetable leaf raw materials where microorganisms are attached to kimchi, it is necessary to optimize sample processing to facilitate rapid Raman detection of yeast and its spores. The advantage of traditional Raman spectroscopy machine learning methods for strain identification is that they can handle binary classification problems, but for the multi-classification problem of yeast at the species level, corresponding strategies need to be designed to increase the complexity of modeling, so as to achieve accurate identification of key microbial yeasts, and then determine the types and proportions of yeasts in the fermentation process. This is of great significance for providing food brewing quality assurance measures and reducing pollution in the brewing industry. Summary of the Invention

[0005] Since conventional yeast identification takes a long time and has poor specificity, it is not suitable for brewing companies to quickly identify the type and content of yeast during the fermentation process to guide production. It is necessary to develop a method for rapid yeast identification and proportion determination for yeast-rich fermented food samples.

[0006] The present invention aims to provide a gold nanoparticle-based Raman spectroscopy yeast detection method. This method uses Raman spectroscopy to pre-treat 10 yeasts found in high abundance during traditional brewing. A window sliding algorithm combined with the XGboost algorithm is used to select the yeast's optimal Raman bands for training on the 10 yeasts. The trained supervised algorithm can provide accurate classification criteria, shortening the time required for strain identification and achieving higher accuracy compared to other strain identification models.

[0007] Fermented food samples, such as daqu (Chinese koji), fermented mash, and kimchi, require the determination of yeast species and content to guide production. This invention utilizes specialized pretreatment for complex, solid, yeast-rich fermentation samples. Unlike conventional sample processing, this method utilizes gradient centrifugation, homogenization, and vortexing to prepare bacterial suspension samples. This treatment maximizes sample biomass, resulting in extremely low background Raman noise and a high single-cell spectral signal-to-noise ratio. This invention enables real-time, rapid, and label-free detection of yeast in fermentation samples, offering advantages such as high timeliness, high sensitivity, strong specificity, and the ability to perform large-scale online testing.

[0008] The present invention provides a discriminant analysis classification model for identifying yeast strain types, wherein the model is established according to the following steps:

[0009] (1) Preparation of yeast suspension:

[0010] Saccharomyces cerevisiae, Wickerhamomyces anomalus, Saccharomycopsis fibuligera, Pichia manshurica, Pichia kudriavzevii, Zygosaccharomyces baili, Schizosaccharomyces pombe, Pichia fermentans, Rhodotorula mucilaginosa, and Candida glabrata were inoculated into a culture medium for fermentation, and the culture fluids at different culture times were collected to obtain yeast suspensions;

[0011] (2) Pretreatment of yeast suspension:

[0012] After centrifuging the bacterial suspension prepared in step (1), the precipitate was collected, sterile water was added to the precipitate, and then centrifuged. This was repeated 2 to 3 times to obtain a pretreated yeast suspension.

[0013] (3) Preparation of gold nanoparticles:

[0014] Using the trisodium citrate heating reduction method, ultrapure water and potassium chloroaurate solution are added to a flask, mixed evenly, and heated to boiling. Then, trisodium citrate solution is quickly added and the boiling is continued. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles with a diameter of 10 to 50 nm.

[0015] (4) Raman spectroscopy detection:

[0016] The yeast suspension obtained in step (2) and the gold nanoparticles obtained in step (3) were mixed and then single-cell Raman spectrum was collected, wherein the spectrum collection conditions were: using a 532 nm laser and scanning the spectrum at 500 to 3750 cm -1 , laser intensity was 1-300 mW, acquisition time was 1-20 s / time, cumulative number was 1, and 50-2000 cells were collected from yeast suspensions of different culture times;

[0017] (5) Processing of single-cell Raman spectroscopy data:

[0018] The single-cell Raman spectroscopy data obtained in step (4) are subjected to cosmic ray elimination, background noise removal, baseline correction, Savitzky-Golay smoothing, and all data are normalized;

[0019] (6) Model building:

[0020] Use machine learning to establish a classification model for yeast strains: divide the spectral window into intervals, every 20cm -1 Extract a characteristic Raman band for an interval. After the window is updated, the updated characteristic Raman band is used to extract a new data set. The extracted characteristic Raman band is used as a training data set using the XGboost algorithm. The training data set and the detection data set are set. Among them, the training data set is 70% of the data set, and the detection data set is 30% of the data set. The model training accuracy of the characteristic Raman band built based on the updated window sliding characteristic Raman band is the highest, and the optimal characteristic Raman band and the optimal model parameters are obtained.

[0021] The best characteristic Raman band from the extraction window is: 540-560cm -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm-1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 ;

[0022] The optimal characteristic Raman band extraction process is to generate a closed data set by sliding the window step. The calculation formula for characteristic Raman band extraction is as follows:

[0023]

[0024] Where SG is the characteristic band, t is the number of sliding windows, n is the total number of times the sliding window moves forward, c is the number of sliding windows from a correctly classified species, P k (xgb) is the probability of accurate classification based on the k-th original prediction of the XGboost classification model, Pijk(r) represents the probability of accurate classification based on the k-th sliding window covering the i-th occlusion, and Wi is the weight of the i-th sliding window to offset the effect caused by the size of the sliding window.

[0025] The specific settings of the XGboost algorithm are as follows: the number of leaf nodes is 31, the objective function is "multiclass", the number of categories is 10, the learning rate is 0.05, the feature selection ratio for tree building is 0.9, the sample sampling ratio for tree building is 0.9, and the bagging is performed every k iterations is 2.

[0026] In one embodiment of the present invention, step (3) gold nanoparticles are prepared by using a trisodium citrate heating reduction method, adding ultrapure water and a 10 mg / mL potassium chloroaurate solution into a flask, mixing them evenly and heating them to boiling, then quickly adding 0.1% trisodium citrate solution and continuing to boil. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles, and the obtained gold nanoparticles have a diameter of 10 to 50 nm.

[0027] In one embodiment of the present invention, the yeast is: Saccharomyces cerevisiae CICC32165, Wickerhamomyces anomalus CICC 1312, Saccharomycopsis fibuligera CICC2161, Pichia manshurica CICC 31428, Pichia kudriavzevii CICC 33179, Zygosaccharomyces baili CICC 31217, Schizosaccharomyces pombe CICC 1056, Pichiafermentans CICC 33123, Rhodotorula mucilaginosa CICC 33124, Candida glabrata CICC 2.697.

[0028] In one embodiment of the present invention, the different culture times in step (1) are culture solutions obtained by culturing for 12 hours, 24 hours, and 48 hours.

[0029] In one embodiment of the present invention, in the Raman spectroscopy detection in step (3), the laser intensity is 3 mW, the acquisition time is 2 s / time, the cumulative number is 1 time, and 100 cells are collected from the yeast suspensions of different culture times.

[0030] In one embodiment of the present invention, the bacterial suspension in step (4) is mixed with gold nanoparticles and placed on a Raman chip, and then dried for Raman spectrum detection; the volume ratio of the gold nanoparticles to the bacterial suspension is (1:1) to (1:1×10 5 ).

[0031] In one embodiment of the present invention, the volume ratio of the gold nanoparticles to the bacterial suspension is 1:1.

[0032] In one embodiment of the present invention, the Raman chip is an aluminum-plated Raman chip.

[0033] The present invention also provides a method for rapidly identifying and detecting yeast in fermented foods, the method comprising the following steps:

[0034] (1) Collect samples;

[0035] (2) Sample pretreatment:

[0036] Add the solution to the sample obtained in step (1), stir, shake with a homogenizer or a vortex instrument, let it stand, centrifuge and take the supernatant, centrifuge the obtained supernatant and take the precipitate;

[0037] The precipitate was added with sterile water and centrifuged, and this was repeated 2 to 3 times to obtain the pretreated sample;

[0038] (3) Preparation of gold nanoparticles:

[0039] Using the trisodium citrate heating reduction method, ultrapure water and potassium chloroaurate solution are added to a flask, mixed evenly, and heated to boiling. Then, trisodium citrate solution is quickly added and the boiling is continued. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles with a diameter of 10 to 50 nm.

[0040] (4) Raman spectroscopy detection:

[0041] The sample obtained in step (2) and the gold nanoparticles obtained in step (3) were mixed and then single cell Raman spectrum was collected. The spectrum collection conditions were as follows: using a 532 nm laser and scanning the spectrum at 500 to 3750 cm -1 , laser intensity was 1-300 mW, acquisition time was 1-20 s / time, cumulative number was 1, and 50-2000 cells were collected from yeast suspensions of different culture times;

[0042] (5) Processing of single-cell Raman spectroscopy data:

[0043] The single-cell Raman spectroscopy data obtained in step (4) are subjected to cosmic ray elimination, background noise removal, baseline correction, Savitzky-Golay smoothing, and all data are normalized;

[0044] (6) Yeast identification

[0045] The above discriminant analysis classification model was used to identify the unknown strains, and the characteristic band of the single-cell Raman spectroscopy data of the sample obtained in step (5) was 540-560 cm -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm-1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 By inputting the results into the XGboost classification model, the 10 yeast species in the sample can be identified. According to the model strain classification criteria, if the score is greater than or equal to 0.9, the strain with the highest score is identified as that strain; if the score is less than 0.9, it is identified as not belonging to yeast but to other types of strains.

[0046] In one embodiment of the present invention, the yeast is: Saccharomyces cerevisiae, Wickerhamomyces anomalus, Saccharomycopsis fibuligera, Pichia manshurica, Pichiakudriavzevii, Zygosaccharomyces baili, Schizosaccharomyces pombe, Pichiafermentans, Rhodotorula mucilaginosa, Candida glabrata.

[0047] In one embodiment of the present invention, step (3) gold nanoparticles are prepared by using a trisodium citrate heating reduction method, adding ultrapure water and a 10 mg / mL potassium chloroaurate solution into a flask, mixing them evenly and heating them to boiling, then quickly adding 0.1% trisodium citrate solution and continuing to boil. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles, and the obtained gold nanoparticles have a diameter of 10 to 50 nm.

[0048] In one embodiment of the present invention, step (2) is as follows: take 5 g of the sample obtained in step (1), add 20 mL of the solution, mix evenly with a homogenizer or vortexer, let it stand, centrifuge at a low speed of 500 rpm, and take the supernatant;

[0049] The supernatant was centrifuged at 7000 rpm to collect the precipitate; the precipitate was repeatedly rinsed with sterile water and centrifuged three times.

[0050] In one embodiment of the present invention, the sample in step (1) refers to a solid or liquid raw material mixture containing multiple microorganisms, the block or powdered fermentation sample is crushed into powder using a mortar or grinder, and the liquid fermentation sample is filtered using gauze.

[0051] In one embodiment of the present invention, the sample obtained in step (1) is added to the solution, stirred, and rotated using a homogenizer for 10 to 60 minutes at a speed of 3 to 12 times / second, or vortexed using a vortexer for 15 to 90 minutes, then allowed to stand for 2 to 5 minutes, the solution is aspirated, and the solution is centrifuged at 500 rpm for 2 minutes, and the supernatant is retained;

[0052] Remove the supernatant and centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Rinse with sterile water and centrifuge three times.

[0053] In one embodiment of the present invention, the samples include but are not limited to brewed and fermented foods such as Daqu, fermented mash, vinegar mash, and kimchi.

[0054] The present invention also provides application of the above method in detecting the relative content of yeast in fermentation sample microorganisms.

[0055] In one embodiment of the present invention, the yeast is: Saccharomyces cerevisiae, Wickerhamomyces anomalus, Saccharomycopsis fibuligera, Pichia manshurica, Pichiakudriavzevii, Zygosaccharomyces baili, Schizosaccharomyces pombe, Pichiafermentans, Rhodotorula mucilaginosa, Candida glabrata.

[0056] In one embodiment of the present invention, the application is:

[0057] According to the above method, 300 to 10,000 single-cell Raman spectra were collected for microorganisms in the fermentation sample community;

[0058] The single-cell Raman spectrum selected characteristic bands were input into the yeast XGboost classification model to output the unknown single-cell types in the fermentation sample. The total number of single-cell categories was counted from the output 300 to 10,000 single-cell prediction results. The ratio of each yeast number to the total number of single cells was calculated to obtain the relative content of yeast in the fermentation sample microorganisms.

[0059] In one embodiment of the present invention, the percentage of yeast in the fermentation sample microorganisms is calculated by counting the total number of output single cell types and calculating the ratio of the number of each yeast type to the total number of single cells to obtain the relative content of yeast in the fermentation sample microorganisms.

[0060] The calculation formula is:

[0061]

[0062] Where C(P) represents the relative content of yeast in the fermentation sample microorganisms, n(P) represents the number of yeast obtained by using the model to predict unknown single cells, and n represents the total number of collected single cells.

[0063] In one embodiment of the present invention, the percentage of yeast in the fermentation sample microbial community can be calculated by collecting single-cell Raman spectra from 300 to 10,000 single-cell Raman spectra. A larger number of single-cell Raman spectra can better represent the microbial community in the fermentation sample. In this embodiment, the number of single-cell Raman spectra collected is 800 to 1,000.

[0064] Beneficial effects

[0065] (1) The present invention uses gold nanoparticles to establish a yeast Raman spectrum standard library of different growth cycles. Yeast in different fermentation cycles are included in the yeast Raman spectrum standard library, which has strong adaptability. The yeast prediction model can be reused. After it is established once, as long as the same Raman spectrum measurement conditions are maintained, the single cell to be identified can be directly input into the model for identification.

[0066] (2) Complex substances in fermentation samples are prone to produce fluorescence Raman interference. The direct separation of cereals, vegetable substances and microorganisms can be achieved by combining sample homogenization and gradient centrifugation. The present invention is easier to use in actual production in fermentation food factories, has low cost and is simple to operate.

[0067] (3) The present invention regionalizes complex sample spectra, which can not only highlight the influence weight of the identified spectrum in the entire correlation analysis, but also realize the identification of the entire spectrum in a sliding form. The established yeast Raman spectral characteristic band strain prediction model can shorten the sample detection time to about 1 hour and the accuracy rate can reach more than 90% compared with other strain identification models. The strain identification time is shorter and the accuracy rate is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 : Raman spectra of 10 yeast species in Example 1 of the present invention.

[0069] Figure 2 : Schematic diagram of the sliding window in the model in Example 1 of the present invention.

[0070] Figure 3 : The model confusion matrix obtained by training in Example 1 of the present invention. DETAILED DESCRIPTION

[0071] The present invention will be further described below with reference to specific embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the present invention and are not intended to limit the present invention.

[0072] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a" does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0073] The microbial community structure described in the following examples refers to the collection of different single cells in a fermentation sample using single-cell Raman spectroscopy within a certain space.

[0074] The Daqu, fermented grains, vinegar fermented grains, and pickles involved in the following examples were sampled from the fermentation processes of Maotai Town Distillery, Shanxi Laochen Vinegar Factory, and Sichuan Pickle Factory, respectively.

[0075] Saccharomyces cerevisiae CICC 32165, Wickerhamomyces anomalus CICC 1312, Saccharomycopsis fibuligera CICC 2161, Pichia manshurica CICC 31428, Pichia kudriavzevii CICC 33179, Zygosaccharomyces baili CICC 31217, Schizosaccharomyces pombe CICC 1056, Pichia fermentans CICC33123, and Rhodotorula mucilaginosa CICC 33124 involved in the following examples were purchased from China Industrial Microorganism Culture Collection Center.

[0076] Candida glabrata CGMCC 2.697 involved in the following examples was purchased from China General Microorganism Collection Center.

[0077] Saccharomyces cerevisiae, Wickerhamomyces anomalus, Saccharomycopsis fibuligera, Pichia manshurica, Pichia kudriavzevii, Zygosaccharomyces baili, Schizosaccharomyces pombe, Pichia fermentans, Rhodotorula mucilaginosa, and Candida glabrata involved in the following examples are numbered SC, WA, SF, PM, PK, ZB, SP, PF, RM, and CG, respectively.

[0078] The culture medium involved in the following examples is as follows:

[0079] YPD medium: yeast extract 10.0 g / L, peptone 20.0 g / L, D-glucose 20.0 g / L, pH 5.6, sterilized at 115°C for 30 min.

[0080] The calculation of the percentage of yeast in the fermentation sample microorganisms involved in the following examples is:

[0081] The total number of output single cell categories is counted, and the ratio of the number of each yeast to the total number of single cells is calculated to obtain the relative content of yeast in the fermentation sample microorganisms. The calculation formula is:

[0082]

[0083] Where C(P) represents the relative content of yeast in the fermentation sample microorganisms, n(P) represents the number of yeast obtained by using the model to predict unknown single cells, and n represents the total number of collected single cells.

[0084] Example 1: Construction of classification model

[0085] (1) Preparation of yeast suspension

[0086] Saccharomyces cerevisiae CICC 32165, Wickerhamomyces anomalus CICC 1312, Saccharomycopsis fibuligera CICC 2161, Pichia manshurica CICC 31428, Pichia kudriavzevii CICC 33179, Zygosaccharomyces baili CICC 31217, Schizosaccharomyces pombe CICC 1056, Pichia fermentans CICC 33123, Rhodotorulamucilaginosa CICC 33124, and Candida glabrata CGMCC 2.697 were inoculated into YPD medium and cultured at 37°C for 24 h. The culture fluids of the yeasts cultured for 12 h, 24 h, and 48 h were collected to prepare yeast suspensions.

[0087] (2) Pretreatment of yeast suspension:

[0088] 1 mL of culture solution cultured for 12 h, 24 h, and 48 h was respectively aspirated into 1.5 mL centrifuge tubes, centrifuged at 7000 rpm for 2 min, the supernatant was discarded, 500 μL of sterile water was added to the precipitate, and the mixture was mixed by pipetting. The supernatant was discarded, 500 μL of sterile water was added to the precipitate, and the mixture was mixed by pipetting. The above steps were repeated 3 times to prepare yeast suspensions cultured for 12 h, 24 h, and 48 h, respectively.

[0089] (3) Preparation of gold nanoparticles:

[0090] Using the trisodium citrate heating reduction method, add 47mL of ultrapure water and 3mL of 10mg / mL potassium chloroaurate solution into a flask, mix well and heat to boiling, then quickly add 2mL of 0.1% trisodium citrate solution and continue boiling. When the solution turns purple-red, stop heating. The resulting gold nanoparticles have a diameter of 10-50nm. After cooling to room temperature, store at 4℃ in the dark.

[0091] (4) Raman spectroscopy detection

[0092] 2.5 μL of the bacterial suspension obtained in step (2) after culturing for 12 h, 24 h, and 48 h was taken and mixed with gold nanoparticles in a volume ratio of 1:1. 2.5 μL was then placed on an aluminum-coated Raman chip and allowed to air-dry for 10 min. Single-cell Raman spectra were collected using a confocal Raman spectrometer with an excitation wavelength of 532 nm and a scanning spectral range of 500–3750 cm -1 , the power under the microscope is 3mW, and the scanning time is 5s.

[0093] 100 samples of each yeast were collected from the bacterial suspension cultured for 12h, 24h, and 48h. The Raman spectra were subjected to cosmic ray removal, baseline correction (iterative adaptive weighted penalized least squares method, airPLS), Savitzky-Golay smoothing, and normalization to obtain a Raman spectral standard database ( Figure 1 ).

[0094] (5) Use machine learning to establish a discriminant analysis classification model for yeast strains: Use machine learning to establish a classification model for yeast strains: divide the spectral window into intervals, every 20cm -1 Extract a characteristic Raman band for an interval. After the window is updated, the updated characteristic Raman band is used to extract a new data set. The extracted characteristic Raman band is used as a training data set using the XGboost algorithm. The training data set and the detection data set are set. Among them, the training data set is 70% of the data set, and the detection data set is 30% of the data set. The model training accuracy of the characteristic Raman band built based on the updated window sliding characteristic Raman band is the highest, and the optimal characteristic Raman band and the optimal model parameters are obtained.

[0095] The best characteristic Raman band from the extraction window is: 540-560cm -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 .

[0096] The optimal characteristic Raman band extraction process is: through the sliding window ( Figure 2 ) steps to produce a closed data set, and the calculation formula of the characteristic Raman band is as follows:

[0097]

[0098] Where SG is the characteristic band, t is the number of sliding windows, n is the total number of times the sliding window moves forward, c is the number of sliding windows from a correctly classified species, P k (xgb) is the probability of accurate classification based on the k-th original prediction of the XGboost classification model, Pijk(r) represents the probability of accurate classification based on the k-th sliding window covering the i-th occlusion, and Wi is the weight of the i-th sliding window to offset the effect caused by the size of the sliding window.

[0099] The specific settings of the XGboost algorithm are as follows: the number of leaf nodes is 31, the objective function is "multiclass", the number of categories is 10, the learning rate is 0.05, the feature selection ratio for tree building is 0.9, the sample sampling ratio for tree building is 0.9, and the bagging is performed every k iterations is 2.

[0100] Obtain model evaluation parameters and confusion matrix of 10 bacteria ( Figure 3 ), the obtained model has the best effect.

[0101] Example 2: Identification of 10 yeast species in Daqu

[0102] Daqu, the fermentation base for baijiu, is mostly made from wheat. This wheat is ground with water, mixed with the mother koji, and pressed into briquettes. This briquettes, accumulated in a storage tank, form the daqu, which contains various bacteria and enzymes. Daqu plays a crucial role in baijiu brewing, including inoculating microorganisms, feeding grain, saccharifying and fermenting, and producing aroma. Researching yeast species can help better control baijiu flavor compounds, improve traditional craftsmanship, and produce higher-quality baijiu, ultimately enhancing the state-of-the-art in baijiu brewing.

[0103] (1) Sample processing

[0104] Three kinds of Daqu samples were collected from different qu rooms, and the qu blocks were crushed in a mortar to a size of 1-10 mm. After mixing the powders, 5 g was weighed, added with 20 mL of sterile water, and vibrated in a homogenizer for 10 minutes at a speed of 10 times / second, working for 5 seconds, and resting for 5 seconds to ensure that the microorganisms are fully dispersed in the sterile water solvent.

[0105] Then, the solution was allowed to stand for 5 minutes, 6 mL of the solution was aspirated, and the solution was centrifuged at 500 rpm for 2 minutes, and the supernatant was retained;

[0106] Remove the supernatant and centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Rinse with sterile water and centrifuge three times.

[0107] Pretreated Daqu samples 1 to 3 were prepared respectively.

[0108] Preparation of gold nanoparticles:

[0109] Using the trisodium citrate heating reduction method, add 47mL of ultrapure water and 3mL of 10mg / mL potassium chloroaurate solution into a flask, mix well and heat to boiling, then quickly add 2mL of 0.1% trisodium citrate solution and continue boiling. When the solution turns purple-red, stop heating. The resulting gold nanoparticles have a diameter of 10-50nm. After cooling to room temperature, store at 4℃ in the dark.

[0110] (2) The pretreated Daqu samples 1 to 3 were mixed with gold nanoparticles in a volume ratio of 1:1, and 2.5 μL was taken onto an aluminum-coated Raman chip. The mixture was allowed to air-dry for 10 min, and single-cell Raman spectra were collected using a confocal Raman spectrometer. The excitation wavelength was 532 nm, and the scanning spectral range was 500–3750 cm -1; The power under the microscope was 3 mw, and the scanning time was 5 s; the number of accumulations was: 1, followed by cosmic ray removal, baseline correction (iterative adaptive weighted penalized least squares method, airPLS), Savitzky-Golay smoothing, and normalized Raman data processing to obtain the Raman spectrum of Daqu microorganisms. 1,000 single-cell Raman spectra were collected for each type of Daqu.

[0111] (3) The above-mentioned discriminant analysis classification model is used to identify the unknown strains, and the single-cell Raman spectrum data of the sample obtained in step (2) is selected to select the characteristic Raman band (540-560 cm -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1) is input into the XGboost classification model obtained in Example 1, and the 10 yeast types in the sample can be identified; according to the model strain classification standard, if the score is greater than or equal to 0.9, the strain with the highest score is identified as this strain; if the score is less than 0.9, it is identified as not belonging to yeast, but to other types of strains.

[0112] Based on the model strain identification and classification criteria, the discriminant model was used to identify 3,000 yeast cells collected. The results are shown in Table 1.

[0113] Table 1: Proportions of yeast in different types of Daqu in Example 2

[0114]

[0115] The results showed that by counting the identified species of 1,000 single cells in each Daqu, the proportion of yeast in the Daqu microbial community could be obtained.

[0116] Furthermore, using the method of the present invention, the detection and analysis time for a single cell is 2 s, while the detection and analysis time for 1000 single cells in each Daqu is 2000 s, i.e., 0.33 h.

[0117] Example 3: Identification of 10 yeast species in fermented grains

[0118] Fermented mash is the primary vehicle for microbial fermentation and the direct source of baijiu's aromatic compounds. Functional microorganisms are enriched during the mash stacking process, and fungal populations vary significantly between rounds. Identifying the types and abundance of yeast microorganisms during this process allows for better monitoring of the dynamic microbial changes during stacked fermentation.

[0119] (1) Sample processing

[0120] Two samples of mash from the first and second rounds of Maotai-flavor liquor were collected respectively. The mash was crushed in a mortar to a size of 1-10 mm. After mixing the powders, 5 g was weighed and added with 20 mL of sterile water. The mixture was shaken in a homogenizer for 10 minutes at a speed of 10 times / second, working for 5 seconds and resting for 5 seconds to ensure that the microorganisms were fully dispersed in the sterile water solvent.

[0121] Then, the solution was allowed to stand for 5 minutes, 6 mL of the solution was aspirated, and the solution was centrifuged at 500 rpm for 2 minutes, and the supernatant was retained;

[0122] Remove the supernatant and centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Rinse with sterile water and centrifuge three times.

[0123] Pretreated mash samples 1 to 2 were prepared respectively.

[0124] Preparation of gold nanoparticles:

[0125] Using the trisodium citrate heating reduction method, add 47mL of ultrapure water and 3mL of 10mg / mL potassium chloroaurate solution into a flask, mix well and heat to boiling, then quickly add 2mL of 0.1% trisodium citrate solution and continue boiling. When the solution turns purple-red, stop heating. The resulting gold nanoparticles have a diameter of 10-50nm. After cooling to room temperature, store at 4℃ in the dark.

[0126] (2) 2.5 μL of pretreated mash samples 1-2 were mixed with gold nanoparticles in a volume ratio of 1:1, and 2.5 μL was placed on an aluminum-coated Raman chip. The sample was air-dried for 10 min, and single-cell Raman spectra were collected using a confocal Raman spectrometer. The excitation wavelength was 532 nm, and the scanning spectral range was 500-3750 cm -1 ; The power under the microscope was 3 mw, and the scanning time was 5 s; the number of accumulations was: 1, followed by cosmic ray removal, baseline correction (iterative adaptive weighted penalized least squares method, airPLS), Savitzky-Golay smoothing, and normalized Raman data processing to obtain the Raman spectrum of Daqu microorganisms. 1000 single-cell Raman spectra were collected for each mash.

[0127] (3) Use the above-mentioned discriminant analysis classification model to identify unknown strains, and use the single cell characteristic Raman band of the sample obtained in step (2): 540-560 cm -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -11360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 The 10 yeast species in the sample can be identified by inputting them into the XGboost classification model obtained in Example 1. According to the model strain classification standard, if the score is greater than or equal to 0.9, the strain with the highest score is identified as this strain; if the score is less than 0.9, it is identified as not belonging to yeast, but to other types of strains.

[0128] Based on the model strain identification and classification criteria, the discriminant model was used to identify 2,000 yeast cells collected. The results are shown in Table 2.

[0129] Table 2: Proportions of yeast in different types of fermented grains in Example 3

[0130]

[0131] The results showed that the proportion of yeast in the microbial community of each mash could be obtained by counting the identified species of 1,000 single cells in each mash.

[0132] Moreover, using the method of the present invention, the detection and analysis time of a single cell is 2 s, while the detection and analysis time of 1000 single cells in each mash is 2000 s, that is, 0.33 h.

[0133] Example 4: Identification of 10 yeast species in vinegar mash

[0134] Traditional vinegar brewing mostly involves an open, multi-strain mixed fermentation process. The complexity of the raw materials and the open environment contribute to the complexity and diversity of the vinegar-making microbial community. Vinegar production can be divided into the winemaking stage and the vinegarization stage. The winemaking stage primarily involves the production of alcohol through fermentation of the raw materials. Yeasts play a key role in this stage. Yeasts not only play a crucial role in the winemaking process but also influence the aroma of the fermentation product, thereby influencing the flavor of the vinegar. After the winemaking stage, the vinegarization stage begins, where yeasts also play a vital role. Therefore, rapid yeast detection is crucial in the vinegar brewing process.

[0135] (1) Sample processing

[0136] Four types of vinegar mash samples were collected, and vinegar mash samples from the 1st, 5th, 10th and 15th days of acetic acid fermentation were taken respectively. 5 g of sample was weighed, 20 mL of sterile water was added and the mixture was shaken in a homogenizer for 10 minutes at a speed of 10 times / second, working for 5 seconds and resting for 5 seconds to fully disperse the microorganisms in the sterile water solvent.

[0137] Then, the solution was allowed to stand for 5 minutes, 6 mL of the solution was aspirated, and the solution was centrifuged at 500 rpm for 2 minutes, and the supernatant was retained;

[0138] Remove the supernatant and centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Rinse with sterile water and centrifuge three times.

[0139] Pretreated vinegar mash samples 1 to 4 were prepared respectively.

[0140] Preparation of gold nanoparticles:

[0141] Using the trisodium citrate heating reduction method, add 47mL of ultrapure water and 3mL of 10mg / mL potassium chloroaurate solution into a flask, mix well and heat to boiling, then quickly add 2mL of 0.1% trisodium citrate solution and continue boiling. When the solution turns purple-red, stop heating. The resulting gold nanoparticles have a diameter of 10-50nm. After cooling to room temperature, store at 4℃ in the dark.

[0142] (2) 2.5 μL of pretreated vinegar mash samples 1 to 4 were mixed with gold nanoparticles in a volume ratio of 1:1, and 2.5 μL was placed on an aluminum-coated Raman chip. The mixture was air-dried for 10 min and single-cell Raman spectra were collected using confocal Raman spectroscopy. The excitation wavelength was 532 nm and the scanning spectral range was 500–3750 cm -1 ; The power under the microscope was 3 mw, the scanning time was 5 s; the number of accumulations was: 1, and then cosmic ray removal, baseline correction (iterative adaptive weighted penalized least squares method, airPLS), Savitzky-Golay smoothing, and normalized Raman data processing were performed to obtain the Raman spectrum of Daqu microorganisms. 800 single-cell Raman spectra were collected for each vinegar mash.

[0143] (3) Use the above-mentioned discriminant analysis classification model to identify unknown strains, and use the single cell characteristic Raman band (540-560 cm) of the sample obtained in step (2) to identify the unknown strains. -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 ) is input into the XGboost classification model obtained in Example 1, and the 10 yeast types in the sample can be identified; according to the model strain classification standard, if the score is greater than or equal to 0.9, the strain with the highest score is identified as this strain; if the score is less than 0.9, it is identified as not belonging to yeast, but to other types of strains.

[0144] Based on the model strain identification and classification criteria, the discriminant model was used to identify 3200 yeast cells collected. The results are shown in Table 3.

[0145] Table 3: Proportions of yeast in different types of fermented mash in Example 4

[0146]

[0147] The results showed that the proportion of yeast in the vinegar mash microbial community could be obtained by counting the identified species of 800 single cells in each vinegar mash.

[0148] Moreover, using the method of the present invention, the detection and analysis time of a single cell is 2 seconds, while the detection and analysis time of 800 single cells of each shrimp paste is 1600 seconds, that is, 0.26 hours.

[0149] Example 5: Identification of 10 yeast species in kimchi

[0150] Kimchi is made from fresh fruits and vegetables, supplemented with spices, with or without the addition of leavening agents, and is made by anaerobic fermentation through salt water immersion. Yeast is a functional microorganism among kimchi bacteria.

[0151] (1) Sample processing

[0152] Three types of kimchi samples were collected, namely pickled peppers, pickled radishes, and pickled cucumbers. The pickled vegetables were crushed in a mortar to a size of 1-10 mm. After mixing the powders, 5 g was weighed and added with 20 mL of sterile water. The mixture was shaken in a homogenizer for 10 minutes at a speed of 10 times / second, working for 5 seconds and resting for 5 seconds to ensure that the microorganisms were fully dispersed in the sterile water solvent.

[0153] Then, the solution was allowed to stand for 5 minutes, 6 mL of the solution was aspirated, and the solution was centrifuged at 500 rpm for 2 minutes, and the supernatant was retained;

[0154] Remove the supernatant and centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Centrifuge at 7000 rpm for 2 minutes. Discard the supernatant and add 6 mL of sterile water to the pellet. Mix thoroughly by pipetting. Rinse with sterile water and centrifuge three times.

[0155] Pretreated kimchi samples 1 to 3 were prepared respectively.

[0156] Preparation of gold nanoparticles:

[0157] Using the trisodium citrate heating reduction method, add 47mL of ultrapure water and 3mL of 10mg / mL potassium chloroaurate solution into a flask, mix well and heat to boiling, then quickly add 2mL of 0.1% trisodium citrate solution and continue boiling. When the solution turns purple-red, stop heating. The resulting gold nanoparticles have a diameter of 10-50nm. After cooling to room temperature, store at 4℃ in the dark.

[0158] (2) The pretreated kimchi samples 1 to 3 were mixed with gold nanoparticles in a volume ratio of 1:1, and 2.5 μL was placed on an aluminum-coated Raman chip. The mixture was air-dried for 10 min, and single-cell Raman spectra were collected using a confocal Raman spectrometer. The excitation wavelength was 532 nm, and the scanning spectral range was 500–3750 cm -1; The power under the microscope was 3 mw, the scanning time was 5 s; the number of accumulations was: 1, and then cosmic ray removal, baseline correction (iterative adaptive weighted penalized least squares method, airPLS), Savitzky-Golay smoothing, and normalized Raman data processing were performed to obtain the Raman spectrum of Daqu microorganisms. 1000 single-cell Raman spectra were collected for each kimchi.

[0159] (3) Use the above-mentioned discriminant analysis classification model to identify unknown strains, and use the single cell characteristic Raman band (540-560 cm) of the sample obtained in step (2) to identify the unknown strains. -1 , 620~640cm -1 , 660~680cm -1 , 680~700cm -1 720~740cm -1 740~760cm -1 , 780~800cm -1 , 820~840cm -1 , 860~880cm -1 940~960cm -1 1000~1020cm -1 1020~1040cm -1 1080~1100cm -1 1120~1140cm -1 1160~1180cm -1 1200~1220cm -1 1220~1240cm -1 1300~1320cm -1 1320~1340cm -1 1360~1380cm -1 1440~1460cm -1 1560~1580cm -1 1600~1620cm -1 1660~1680cm -1 1720~1740cm -1 2920~2940cm -1 2960~2980cm -1 2880~2900cm -1 3320~3340cm -1 ) is input into the XGboost classification model obtained in Example 1, and the 10 yeast types in the sample can be identified; according to the model strain classification standard, if the score is greater than or equal to 0.9, the strain with the highest score is identified as this strain; if the score is less than 0.9, it is identified as not belonging to yeast, but to other types of strains.

[0160] Based on the model strain identification and classification criteria, the discriminant model was used to identify the 4,000 yeast cells collected. The results are shown in Table 4.

[0161] Table 4: Proportions of yeast in different types of kimchi in Example 5

[0162]

[0163]

[0164] The results showed that the proportion of yeast in the kimchi microbial community could be obtained by counting the identified species of 1,000 single cells in each kimchi.

[0165] Moreover, using the method of the present invention, the detection and analysis time of a single cell is 2 s, while the detection and analysis time of 1000 single cells in each kimchi is 2000 s, that is, 0.33 h.

[0166] Although the present invention has been disclosed above in terms of preferred embodiments, it is not intended to limit the present invention. Anyone familiar with this technology can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the definition of the claims.

Claims

1. A classification model for identifying yeast strain types, characterized in that: The model was established according to the following steps: (1) Preparation of yeast suspension: Saccharomyces cerevisiae ( Saccharomyces cerevisiae ), abnormal Wickham yeast ( Wickerhamomyces anomalus )、Cellulose capsule compound yeast( Saccharomycopsis fibuligera )、Pichia pastoris( Pichia manshurica ), Pichia kudriavzevii ( Pichia kudriavzevii ), Zygosaccharomyces beijerinii ( Zygosaccharomyces baili ), Schizosaccharomyces pombe ( Schizosaccharomyces pombe ), fermentation of Pichia pastoris ( Pichia fermentans )、Red yeast rice( Rhodotorula mucilaginosa ) and Candida glabrata ( Candida glabrata ) The strain was inoculated into the culture medium for fermentation, and the culture fluids after culturing for 12 h, 24 h, and 48 h were collected to form bacterial suspensions; (2) Pretreatment of yeast suspension: After centrifuging the bacterial suspension prepared in step (1), the precipitate was collected, sterile water was added to the precipitate, and the precipitate was centrifuged. This was repeated 2 to 3 times to obtain a pretreated yeast suspension. (3) Preparation of gold nanoparticles: Using the trisodium citrate heating reduction method, ultrapure water and potassium chloroaurate solution are added to a flask, mixed evenly, and heated to boiling. Trisodium citrate solution is then quickly added and the boiling is continued. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles with a diameter of 10-50 nm. (4) Raman spectroscopy detection: The yeast suspension obtained in step (2) and cultured for 12 h, 24 h, and 48 h was mixed with the gold nanoparticles obtained in step (3) at a volume ratio of 1:1, and then single-cell Raman spectra were collected. The spectrum collection conditions were as follows: using a 532 nm laser and scanning the spectrum range from 500 to 3750 cm -1 , laser intensity was 1–300 mW, acquisition time was 1–20 s / time, cumulative number was 1, and 50–2000 cells were collected from yeast suspensions at different culture times; (5) Processing of single-cell Raman spectroscopy data: The single-cell Raman spectroscopy data obtained in step (4) are subjected to cosmic ray elimination, background noise removal, baseline correction, Savitzky-Golay smoothing, and all data are normalized; (6) Model building: Use machine learning to establish a classification model for yeast strains: divide the spectral window into intervals, every 20 cm -1 Extract a characteristic Raman band for an interval. After the window is updated, the updated characteristic Raman band is used to extract a new data set. The extracted characteristic Raman band is used as a training data set using the XGboost algorithm. The training data set and the detection data set are set. Among them, the training data set is 70% of the data set, and the detection data set is 30% of the data set. The model training accuracy of the characteristic Raman band built based on the updated window sliding characteristic Raman band is the highest, and the optimal characteristic Raman band and the optimal model parameters are obtained. The extracted characteristic Raman bands are: 540~560 cm -1 , 620~640 cm -1 , 660~680 cm -1 , 680~700cm -1 , 720~740 cm -1 , 740~760 cm -1 , 780~800 cm -1 , 820~840 cm -1 , 860~880 cm -1 , 940~960 cm -1 , 1000~1020 cm -1 , 1020~1040 cm -1 , 1080~1100 cm -1 、1120~1140 cm -1 , 1160~1180 cm -1 , 1200~1220 cm -1 、1220~1240 cm -1 、1300~1320 cm -1 、1320~1340 cm -1 、1360~1380 cm -1 、1440~1460 cm -1 、1560~1580 cm -1 、1600~1620 cm -1 、1660~1680 cm -1 、1720~1740 cm -1 2920~2940cm -1 、2960~2980 cm -1 , 2880~2900 cm -1 、3320~3340 cm -1 ; The characteristic Raman band extraction process is as follows: a closed data set is generated by sliding the window step. The calculation formula for characteristic Raman band extraction is as follows: ； Where SG is the characteristic Raman band, t is the number of sliding windows, n is the total number of times the sliding window moves forward, c is the number of sliding windows from a correctly classified species, and P k (xgb) is the probability of accurate classification of the kth original based on the prediction of the XGboost classification model, Pijk(r) represents the probability of accurate classification based on the kth sliding window covering the i-th occlusion, and Wi is the weight of the i-th sliding window to offset the effect caused by the size of the sliding window; The specific settings of the XGboost algorithm are as follows: the number of leaf nodes is 31, the objective function is "multiclass", the number of classes is 10, the learning rate is 0.05, the feature selection ratio for tree building is 0.9, the sample sampling ratio for tree building is 0.9, and the number of bagging performed every k iterations is 2; The yeast is: Saccharomyces cerevisiae CICC 32165, Wickerhamomyces anomalus CICC 1312 , Saccharomycopsis fibuligera CICC 2161, Pichia manshurica CICC 31428 、 Pichia kudriavzevii CICC 33179 、Zygosaccharomyces baili CICC 31217 、 Schizosaccharomyces pombe CICC 1056 、Pichia fermentans CICC 33123 、Rhodotorula mucilaginosa CICC 33124, Candida glabrata CGMCC 2.697; The bacterial suspension in step (4) is mixed with gold nanoparticles and placed on a Raman chip, and then blown dry for Raman spectrum detection.

2. A method for rapid identification and detection of yeast in fermented foods, characterized in that: The method comprises the following steps: (1) Collect samples; (2) Sample pretreatment: Take 5 g of the sample obtained in step (1) and add 20 mL of solution, mix evenly with a homogenizer or vortexer, let it stand, centrifuge at 500 rpm, and take the supernatant; The supernatant was centrifuged at 7000 rpm to collect the precipitate; the precipitate was repeatedly washed with sterile water and centrifuged three times; (3) Preparation of gold nanoparticles: Using the trisodium citrate heating reduction method, ultrapure water and potassium chloroaurate solution are added to a flask, mixed evenly, and heated to boiling. Trisodium citrate solution is then quickly added and the boiling is continued. When the solution turns purple-red, heating is stopped to obtain gold nanoparticles with a diameter of 10-50 nm. (4) Raman spectroscopy detection: The sample obtained in step (2) and the gold nanoparticles obtained in step (3) were mixed and then single-cell Raman spectra were collected. The spectrum collection conditions were as follows: using a 532 nm laser and scanning the spectrum range from 500 to 3750 cm -1 , laser intensity was 1-300 mW, acquisition time was 1-20 s / time, cumulative number was 1, and 50-2000 cells were collected from yeast suspensions of different culture times; (5) Processing of single-cell Raman spectroscopy data: The single-cell Raman spectroscopy data obtained in step (4) are subjected to cosmic ray elimination, background noise removal, baseline correction, Savitzky-Golay smoothing, and all data are normalized; (6) Identification of yeast: The unknown strain is identified using the classification model for identifying yeast strain types according to claim 1, and the characteristic band of the Raman spectral data obtained in step (5): 540~560 cm -1 , 620~640 cm -1 , 660~680 cm -1 , 680~700cm -1 , 720~740 cm -1 , 740~760 cm -1 , 780~800 cm -1 , 820~840 cm -1 , 860~880 cm -1 , 940~960 cm -1 , 1000~1020 cm -1 , 1020~1040 cm -1 , 1080~1100 cm -1 、1120~1140 cm -1 , 1160~1180 cm -1 , 1200~1220 cm -1 、1220~1240 cm -1 、1300~1320 cm -1 、1320~1340 cm -1 、1360~1380 cm -1 、1440~1460 cm -1 、1560~1580 cm -1 、1600~1620 cm -1 、1660~1680 cm -1 、1720~1740 cm -1 2920~2940cm -1 、2960~2980 cm -1 , 2880~2900 cm -1 、3320~3340 cm -1 Input into the XGboost classification model, the 10 yeast species in the sample can be identified. According to the model strain classification criteria, if the score is greater than or equal to 0.9, the strain with the highest score is identified as that strain; if the score is less than 0.9, it is identified as not belonging to yeast, but to other types of strains. The yeast is: Saccharomyces cerevisiae CICC 32165, Wickerhamomyces anomalus CICC 1312 , Saccharomycopsis fibuligera CICC 2161, Pichia manshurica CICC 31428 、 Pichia kudriavzevii CICC 33179 、Zygosaccharomyces baili CICC 31217 、 Schizosaccharomyces pombe CICC 1056 、Pichia fermentans CICC 33123 、Rhodotorula mucilaginosa CICC 33124, Candida glabrata CGMCC 2.697; The samples include but are not limited to Daqu, fermented grains, vinegar fermented grains, and kimchi.

3. Use of the method according to claim 2 in detecting the relative content of yeast in a fermentation sample microorganism.

Images