A method and application for predicting and controlling metabolite profiles based on the structure of process microbial communities

Abstract

The present invention discloses a method and application for predicting and controlling metabolite profiles based on the structure of process microbiota, belonging to the field of biotechnology. The method inoculates microbiota at the beginning and during the process of Daqu fermentation by means of in-situ system microbiota enrichment, effectively changing the microbiota structure and succession pattern, making the fermentation process in deterministic assembly (MST < 0.5). At the same time, this enrichment method ensures the direct colonization and safety of in-situ microorganisms and effectively saves costs. Based on the artificially modified microbiota structure and the metabolites at the end of fermentation, the present invention constructs an elastic net model by training the above data, establishes the relationship between the microbiota data during the fermentation process and the metabolites at the end point through the elastic net model, and realizes good prediction of the metabolites at the end point (Spearman R > 0.3, P < 0.05) only according to the microbiota data during the fermentation process without fermentation environment parameters.

Description

Technical Field

[0001] The present invention relates to a method and application for predicting and controlling metabolite profiles based on the structure of process microbial communities, belonging to the field of biotechnology. Background Art

[0002] The metabolite profile at the end of food fermentation refers to the types and contents of different metabolites (including quantitative ratios, etc.), which determines the quality of fermented foods. Metabolites are accumulated by the metabolism of complex microbial communities at different fermentation time points during the fermentation process. Predicting and controlling the metabolite profile at the end of fermentation or the content of specific metabolites is fundamental to controlling the quality of fermented foods. However, due to the long fermentation process of microbial communities, the continuous succession of microbial communities during the fermentation process, the differences in the structure of microbial communities at different time nodes of fermentation, and the unclear relationship between the succession of microbial communities during the fermentation process and the entire metabolite profile at the end of fermentation, it is impossible to effectively predict or control the entire metabolite profile at the end of fermentation.

[0003] Daqu, as the saccharifying and fermenting agent for Baijiu, provides microorganisms, enzyme systems, nutrients, and flavor substances for the Baijiu fermentation process. Approximately 70% of Baijiu and most vinegar brewing use Qu as the fermenting agent. The type of Daqu determines the style and flavor type of Baijiu. There has always been a saying that "good Qu makes good wine". However, there are a large number of natural or environmental factors to control during the Qu-making process, which makes the quality of Daqu have strong batch and seasonal fluctuations.

[0004] In actual production, Daqu is often adjusted in a certain proportion according to its sensory quality for application in Baijiu fermentation. However, adjusting based on sensory quality cannot unify the metabolite components therein. Therefore, the final quality of Daqu is unstable. Although some functional strains (such as Bacillus spp., ester-producing yeasts, etc.) or combined bacterial agents (lactic acid bacteria agents) have been added during the Daqu fermentation process, achieving an increase in some metabolites in Daqu, the unstable situation of the final quality of Daqu still cannot be changed. Adjusting the abundance of dominant species during the Daqu fermentation process is a method for regulating Daqu metabolites. However, changing a small number of species is difficult to make the fermentation process more orderly. Therefore, it can be considered to regulate the microbial community to achieve the orderliness of the fermentation process, which is helpful for the stability of the final metabolites.

[0005] Currently, the Daqu fermentation process is still a black-box process. The final metabolite profile cannot be predicted, and the adjustment through the production process still has limitations on the control of Daqu quality. Currently, there are studies on predicting the end-point microbial community based on the initial fermentation microbial community. The research shows that predicting the metabolites at the end of fermentation based on the microbial community structure at a certain time point of Daqu helps to monitor and regulate the quality of Daqu in a timely manner. However, there are still deviations between the prediction results obtained based on the initial microbial community and the actual metabolite situation. Moreover, in the existing methods, when predicting the metabolites at the end of fermentation based on the initial microbial community, it is also necessary to obtain fermentation environment data. Summary of the Invention

[0006] In view of the complex relationship between the microbial community succession during the fermentation process and the entire metabolite profile at the end of fermentation, the present invention has developed a method and its application for predicting and controlling the metabolite profile at the end of fermentation based on the microbial community structure at key nodes of multiple fermentation processes. This method and application technology can be applied to all food fermentations to predict and control the metabolite profile at the end of food fermentation, thereby predicting and controlling the quality of fermented foods, such as solving the problem of fluctuating quality in the production of Daqu.

[0007] A model for predicting the metabolite profile at the end of fermentation based on the microbial community during the fermentation process provided by the present invention realizes accurate prediction of the metabolite profile at the end of fermentation, and at the same time provides an inoculation strategy for changing the microbial community structure during the fermentation process, making the microbial community during the fermentation process more controllable.

[0008] A method for predicting and controlling the metabolite profile based on the microbial community structure during the process, the method comprising:

[0009] Step 1, set several groups of Daqu fermentation experiments, determine the time points with significant differences between different groups, obtain the microbial community structure of each group at the differential time points, and measure the metabolite profile at the end of fermentation; use the microbial community structure at the differential time points as the independent variable and the corresponding metabolite profile at the end of fermentation as the dependent variable to form a training sample;

[0010] Step 2, use the training sample to train an elastic net model, and use the trained elastic net model as a prediction model;

[0011] Step 3, take the Daqu fermentation experimental group for which the metabolite profile at the end of fermentation is to be predicted, determine the microbial community structure at its differential time node, input the microbial community structure at the differential time node into the trained prediction model to obtain the predicted value of the metabolite profile at the end of fermentation, and supplement specific strains and adjust the abundance of the entire microbial community according to the predicted value so that the metabolite at the end of fermentation reaches the expected target.

[0012] Optionally, the obtaining of the microbial community structure of each group at the differential time points and the measurement of the metabolite profile at the end of fermentation include:

[0013] Collect samples of each group of experiments at the initial fermentation moment, the koji-turning moment, and the end of fermentation moment;

[0014] Perform amplicon sequencing and untargeted metabolome analysis on the collected samples to obtain their microbial community structure and metabolite structure respectively;

[0015] Perform differential analysis on the microbial community structure at each moment;

[0016] Determine the differential time points according to the results of the differential analysis, and obtain the microbial community structure at the differential time points.

[0017] Optionally, the training process of the elastic net model in step 2 includes:

[0018] Input the microbial community structure at different time points and the corresponding metabolic profiles;

[0019] Use the elastic net model to fit the model;

[0020] Maximize the cross-validation accuracy of all metabolites;

[0021] Calculate the correlation coefficient between the measured values and the predicted values, and metabolites with a Spearman correlation greater than 0.3 are marked as well-predicted metabolites;

[0022] Obtain all abundance data for prediction and the predicted metabolite profiles.

[0023] Optionally, the elastic net model is an adaptive gene interaction regularization elastic net model.

[0024] Optionally, the ANOSIM analysis method is used to determine the time when significant differences occur between different groups.

[0025] Optionally, the microbial community structure at the different time points consists of bacteria, yeast, and filamentous fungi.

[0026] Optionally, for bacteria, the ANOSIM analysis parameters are R = 0.040 and P = 0.046; for yeast and filamentous fungi, the ANOSIM analysis parameters are R = 0.088 and P = 0.032;

[0027] Among them, the parameter R is used to determine whether there are differences between different groups, and the parameter P is used to determine whether there are significant differences between different groups.

[0028] Optionally, the microbial community structure at the different time points includes: Bacillus, Kroppenstedtia, Pediococcus, Thermoactinomyces, Saccharopolyspora, Weissella, Saccharomyces cerevisiae, Aspergillus, Thermomyces, and Thermoascus.

[0029] Optionally, the different time point is the 7th day after fermentation.

[0030] Optionally, the method inoculates the microbial community at the beginning and during the process of Daqu fermentation by means of in-situ microbial community enrichment.

[0031] Optionally, the fermentation process is a natural fermentation process, and there are no other artificial control factors except for turning the koji.

[0032] Optionally, the Daqu can be replaced by other multi-strain fermented foods or koji produced in the same indoor semi-open production method.

[0033] Optionally, the fermented foods include: Chinese liquor, yellow rice wine, soy sauce, beer, wine, vinegar, fermented tea, traditional fermented vegetables, fermented beverages, alcoholic beverages, yogurt, cheese, fruit vinegar, fermented glutinous rice, fermented soybeans, fermented bean curd, fermented rice and flour foods.

[0034] Optionally, the metabolites that the prediction model can predict include: alcohols, aldehydes and ketones, amino acids and derivatives, esters, nitrogen-containing compounds, nucleotides, organic acids, oxygen-containing heterocyclic compounds, phenols, sugars and sugar alcohols, terpenes, and vitamins.

[0035] Optionally, the method further includes: using the ANOSIM analysis method to analyze the differences in metabolites between different groups.

[0036] Optionally, the method can also establish the corresponding relationship between the microbial community structure during the fermentation process and the metabolite profile at the end of fermentation at the differential time node.

[0037] The present invention also provides a method for controlling the metabolite profile at the end of multi-strain fermented foods by changing the microbial community structure during the fermentation process. The method includes using the method of predicting the metabolite profile at the end of fermentation based on the microbial community structure during the fermentation process of the present invention; according to the corresponding relationship between the microbial community structure during the fermentation process and the metabolite profile at the end of fermentation at the differential time node, substituting the desired metabolite profile at the end of fermentation into the corresponding relationship to obtain the theoretical microbial community structure at the differential time node, comparing the theoretical microbial community structure with the actual microbial community structure at the differential time node, and using inoculation or other means that can control the microbial community structure to adjust the actual microbial community structure to be consistent with the theoretical microbial community structure, so as to achieve the purpose of controlling the metabolite profile at the end of multi-strain fermented foods.

[0038] The beneficial effects of the present invention are:

[0039] The present invention constructs an elastic net model by training the above data, and establishes the relationship between the microbial community data during the fermentation process and the metabolites at the end of fermentation through the elastic net model, realizing good prediction of the metabolites at the end of fermentation (Spearman R>0.3, P<0.05) only based on the microbial community data during the fermentation process without the need for fermentation environment parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0041] Figure 1 : Microbial structure diagram.

[0042] Figure 2:Metabolite quantity diagram.

[0043] Figure 3 :Simulation diagram of the driving effect of microorganisms on metabolites.

[0044] Figure 4 :Model schematic diagram.

[0045] Figure 5 :Model evaluation simulation diagram. Detailed implementation manners

[0046] To make the objectives, technical solutions and advantages of the present invention clearer, the following will further describe the embodiments of the present invention in detail with reference to the accompanying drawings.

[0047] Embodiment 1

[0048] This embodiment provides a method for predicting and controlling metabolite profiles based on the structure of process microbial communities, including:

[0049] Step 1, set several groups of Daqu fermentation experiments, determine the time points with significant differences between different groups, obtain the microbial community structures of each group at the differential time points, and measure the metabolites at the fermentation end point; use the microbial community structures at the differential time points as independent variables and the corresponding metabolites at the fermentation end point as dependent variables to form training samples;

[0050] Among them, obtaining the microbial community structures of each group at the differential time points includes:

[0051] Collect samples of each group of experiments at the initial fermentation moment, koji turning moment, and fermentation end moment;

[0052] Perform amplicon sequencing and untargeted metabolome analysis on the collected samples to obtain their microbial community structures and metabolite profiles respectively;

[0053] Perform differential analysis on the microbial community structures at each moment;

[0054] Determine the differential time points according to the results of the differential analysis, and obtain the microbial community structures at the differential time points.

[0055] Step 2, train an elastic net model using the training samples, and use the trained elastic net model as a prediction model; specifically, it includes:

[0056] Input the microbial community structures at the differential time points and the corresponding metabolome abundances;

[0057] Use the elastic net model to fit the model;

[0058] Maximize the cross-validation accuracy of all metabolites;

[0059] Calculate the correlation coefficient between the measured values and the predicted values, and metabolites with a Spearman correlation greater than 0.3 are labeled as good predictive metabolites;

[0060] Obtain all the abundance data for prediction and the predicted metabolite profiles.

[0061] After performing the fitting calculation of the predicted values and the actual values, the P-value is required to be less than 0.05.

[0062] The elastic net model is an adaptive gene interaction regularization elastic net model, and the construction process of this adaptive gene interaction regularization elastic net model can refer to the introduction in http: / / huttenhower.sph.harvard.edu / melonnpan.

[0063] Step 3, set up a new experimental group for Daqu fermentation, and determine the microbial community structure at its differential time nodes. Input the microbial community structure at the differential time nodes into the trained prediction model to obtain the predicted values of the metabolites at the fermentation end point, and supplement specific strains and adjust the abundance of the entire microbial community according to the predicted values so that the metabolites at the fermentation end point reach the expected target.

[0064] Since there is a corresponding relationship between the inoculated microbial community and the final metabolites, specific strains can be supplemented and the abundance of the entire microbial community can be adjusted according to the difference between the final predicted value and the expected target, so that the metabolites at the fermentation end point reach the expected target, thereby predicting and controlling the quality of fermented foods and solving the problem of fluctuating quality in the production of Daqu.

[0065] The method of the present application can be used for predicting the metabolites at the fermentation end point of all Qu produced by indoor semi-open production, including all Daqu for Baijiu (medium-temperature Daqu, medium-high temperature Daqu, and high-temperature Daqu), soy sauce Qu, and vinegar Qu.

[0066] The method of the present application can also be used for predicting the end metabolites of the fermentation agents made of complex bacteria used in fermented foods, and such fermented foods include: beer, wine, fermented tea, traditional fermented vegetables, alcoholic beverages, yogurt, cheese, fruit vinegar, fermented glutinous rice, fermented soybeans, fermented bean curd, fermented rice and flour foods, etc.

[0067] Example 2

[0068] This example provides a method for predicting and controlling metabolite profiles based on the microbial community structure during the process, including:

[0069] Step 1: Collect training data

[0070] Collect Daqu samples from the high-temperature Daqu fermentation process. Set up 21 koji rooms for sample collection, and collect Daqu samples at 0, 7, 15 days of fermentation and at the end point respectively. Perform amplicon sequencing and untargeted metabolome analysis on the collected samples to obtain their microbial community structures and metabolite structures (metabolite profiles) respectively, asFigure 1 and Figure 2 As shown in Figure 2 , Bacillus, Kroppenstedtia, Pantoea, Weissella, Thermoascus, Thermomyces, and Aspergillus are the main microorganisms during the fermentation process. A total of 546 metabolites were identified, belonging to 12 categories, namely Alchols, Aldehydes and ketones, Aminoacids and derivatives, Esters, Nitrogen-containing compounds, Nucleotides, Organic acids, Oxygen-containing heterocyclic compounds, Phenols, Sugar and sugar alcohol, Terpenes, and Vitamins.

[0071] If it is used for predicting the end metabolites of other koji fermentations, the corresponding data collection process shall be determined according to the actual fermentation process in this step.

[0072] Step 2: Receipt screening

[0073] (1) Determination of microbial data

[0074] Differential analysis of the microbial structures of different schemes was carried out, and it was found that there were significant differences in the flora of different groups (bacteria, ANOSIM: R = 0.040, P = 0.046; fungi, ANOSIM: R = 0.088, P = 0.032). At the same time, differential analysis of the flora at four time points was carried out. The flora structures of the 7 experimental groups were similar at 0 d of fermentation, but starting from the 7th day, the flora structures of different experimental groups were different, which was considered the starting point for the formation of differences in fermentation quality. Therefore, the flora on the 7th day was regarded as the time node for the differences generated by different fermentation processes.

[0075] (2) Determination of metabolite data

[0076] By constructing a metabolite network, significant differences in final metabolism were found among different fermentation groups (ANOSIM: R = 0.328, P = 0.002). Among them, Alchols had the highest content in the F3 group, Aldehydes and ketones had the highest content in the F3 group, Amino acids and derivatives had the highest content in the F2 group, Esters had the highest content in the F4 group, Nitrogen-containing compounds had the highest content in the F4 group, Nucleotides had the highest content in the F3 group, Organic acids had the highest content in the F3 group, Oxygen-containing heterocyclic compounds had the highest content in the F2 group, Phenols had the highest content in the F4 group, Sugar and sugar alcohol had the highest content in the F5 group, Terpenes had the highest content in the F1 group, and Vitamins had the highest content in the F4 group.

[0077] Step 3: Calculation of the driving effect of the microbial community structure on metabolites

[0078] By constructing a structural equation model, the driving effect of different microbial community structures on metabolites was studied. After calculation, it was found that, referring to Figure 3 , the different microbial community structures of the 7 experimental groups had a significant driving effect on 121 metabolites (Spearman correlation > 0.3), including 2 kinds of Alchols, 9 kinds of Aldehydes and ketones, 13 kinds of Amino acids and derivatives, 5 kinds of Esters, 15 kinds of Nitrogen-containing compounds, 13 kinds of Nucleotides, 36 kinds of Organic acids, 3 kinds of Oxygen-containing heterocyclic compounds, 11 kinds of Phenols, 9 kinds of Sugar and sugar alcohol, 1 kind of Terpenes, and 4 kinds of Vitamins. Based on the above results, a prediction model was constructed for 121 metabolites at the end of fermentation using the dominant microbial genera on the 7th day (a total of 27 genera).

[0079] Step 4: Model construction and application

[0080] Please refer to Figure 4 , the model training process includes:

[0081] (1) Model training

[0082] The selected data were used for model training, and the training steps are as follows:

[0083] ① Input: Microbiota structure and corresponding metabolome abundances.

[0084] ② Model construction: Fit the model using the elastic net regression algorithm.

[0085] ③ Model selection: Maximize the cross-validation accuracy of all metabolites.

[0086] ④ Marking well-predicted metabolites: Calculate the correlation coefficient between the measured values and the predicted values, and mark the metabolites with a Spearman correlation greater than 0.3 as well-predicted metabolites.

[0087] ⑤ Results: Obtain all abundance data for prediction and the predicted metabolite profiles

[0088] The elastic net used in this application is an adaptive gene interaction regularization elastic net model, which is an existing publicly available model. The construction process can refer to the introduction in http: / / huttenhower.sph.harvard.edu / melonnpan, mainly including determining the weights of each metabolite according to the training data collected in step 1 of this application, and introducing the determined metabolite weights into the least squares loss function to construct an adaptive elastic net model. After inputting the microbiota structure data described in step 1 of this application, the input data is subjected to arcsine square root conversion to form a microbial importance quantification module, and finally a microbial interaction network matrix for quantifying the importance of each measured microorganism is formed. Combining the adaptive elastic net model with the microbial interaction network matrix, an adaptive gene interaction regularization elastic net model is constructed.

[0089] (2) Model evaluation

[0090] Based on the training model described in (1), verify the data of the samples. As Figure 5 shown, the predicted values and the actual values of all data show significant correlations (R 2 = 0.999). Based on the PLS analysis calculation, the predicted metabolite profiles of different groups are also significantly correlated with the actual values.

[0091] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk, etc.

[0092] Example 3

[0093] This invention refers to the previous research (CN114292772A, a method for improving the quality of Daqu based on process intensification and its application). According to the prediction results provided in this application, in this application, the inoculation of microbial communities is carried out at the beginning and during the process of Daqu fermentation by means of in-situ system microbial community enrichment, which effectively changes the microbial community structure and succession pattern, making the fermentation process in deterministic assembly (MST < 0.5). At the same time, this enrichment method ensures the direct colonization and safety of in-situ microorganisms and effectively saves costs. Based on the artificially modified microbial community structure, the microbial communities at the key nodes (the 7th day and the 15th day) of the fermentation process are effectively adjusted, and significant differences and predictability in the metabolites at the end of fermentation are achieved.

[0094] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting and controlling metabolite profiles based on the structure of process microbial communities, characterized in that, The method includes: Step 1, set several groups of Daqu fermentation experiments, determine the time points with significant differences between different groups, obtain the microbial community structures of each group at the differential time points, and measure the metabolite profiles at the fermentation end points; using the microbial community structures at the differential time points as independent variables and the corresponding metabolite profiles at the fermentation end points as dependent variables to form training samples; Step 2, train an elastic net model using the training samples, and use the trained elastic net model as a prediction model; Step 3, set a new Daqu fermentation experimental group, determine the microbial community structure at its differential time node, input the microbial community structure at the differential time node into the trained prediction model to obtain the predicted values of the metabolites at the fermentation end points, and supplement specific strains according to the predicted values and adjust the abundance of the entire microbial community so that the metabolites at the fermentation end points reach the expected goals; The method inoculates the microbial community at the start and during the Daqu fermentation by means of in-situ system microbial community enrichment; The fermentation process is a natural fermentation process, and there are no other artificial control factors except for turning the Daqu.

2. The method according to claim 1, wherein The obtaining of the microbial community structures of each group at the differential time points and the measurement of the metabolite profiles at the fermentation end points include: Collect samples of each group of experiments at the initial fermentation moment, the Daqu-turning moment, and the fermentation end point moment; Perform amplicon sequencing and non-targeted metabolome analysis on the collected samples to obtain their microbial community structures and metabolite profiles respectively; Perform differential analysis on the microbial community structures at each moment; Determine the differential time points according to the results of the differential analysis, and obtain the microbial community structures at the differential time points.

3. The method according to claim 2, characterized in that, The training process of the elastic net model in Step 2 includes: Input the microbial community structures at the differential time points and the corresponding metabolome abundances; Use the elastic net model to fit the model; Maximize the cross-validation accuracy of all metabolites; Calculate the correlation coefficient between the measured values and the predicted values, and mark the metabolites with a Spearman correlation greater than 0.3 as well-predicted metabolites; Obtain all the abundance data for prediction and the predicted metabolite profiles.

4. The method according to claim 3, wherein The elastic net model is an adaptive gene interaction regularization elastic net model.

5. The method according to claim 4, wherein The ANOSIM analysis method is used to determine the time points with significant differences between different groups.

6. The method according to claim 5, characterized in that The microbial community structure at the differential time points consists of bacteria, yeasts, and filamentous fungi.

7. The method according to claim 6, characterized in that, For bacteria, the ANOSIM analysis parameters are R = 0.040 and P = 0.046; for yeasts and filamentous fungi, the ANOSIM analysis parameters are R = 0.088 and P = 0.032; Among them, the parameter R is used to determine whether there are differences between different groups, and the parameter P is used to determine whether there are significant differences between different groups.

8. The method according to claim 7, wherein The microbial community structure at the differential time points includes: Bacillus, Kroppenstedtia, Pediococcus, Thermoactinomyces, Saccharopolyspora, Weissella, Saccharomyces, Aspergillus, Thermomyces, and Thermoascus.

9. The method according to claim 8, wherein The differential time points are the 7th day, 15th day after fermentation, and the end point.

10. The method according to claim 9, wherein The Daqu can be replaced with other multi-microbial fermentation foods or koji produced in the same indoor semi-open production method.

11. The method according to claim 10, characterized in that, The fermentation foods include: soy sauce, vinegar, fermented tea, traditional fermented vegetables, fermented beverages, alcoholic beverages, yogurt, cheese, douchi, fermented bean curd, fermented rice and flour foods.

12. The method according to claim 11, wherein The metabolites that the prediction model can predict include: alcohols, aldehydes and ketones, amino acids and derivatives, esters, nitrogen-containing compounds, nucleotides, organic acids, oxygen-containing heterocyclic compounds, phenols, sugars and sugar alcohols, terpenes, and vitamins.

13. The method according to claim 12, wherein The method further includes: analyzing the differences in metabolites between different groups by using the ANOSIM analysis method.

Images