Online detection method for protein content and biomass in yeast fermentation process

Abstract

The invention discloses an online detection method for protein content and biomass in a yeast fermentation process. The method comprises the following steps: in a kluyveromyces marxianus fermentation process, collecting near infrared spectrum data of yeast in a fermentation tank in situ, and measuring the protein content and biomass of the yeast at the same time; making spectral data correspond to the measured protein content and biomass, and establishing a sample database; performing preprocessing, dimensionality reduction and feature extraction on the spectral data; dividing samples into a training set and a verification set; based on a training set sample database, establishing a Kluyveromyces marxianus protein content and biomass detection model; the precision of the model is verified through data in the verification set, and the detection model is further optimized. The method is simple and efficient to operate, samples are not damaged, expensive instruments are not needed, real-time detection of multiple parameters in the high-density fermentation process of the kluyveromyces marxianus is achieved, and technical support is provided for large-scale biological manufacturing.

Description

Technical Field

[0001] The invention belongs to the field of near-infrared spectrum online monitoring, and particularly relates to an online detection method for protein content and biomass in a yeast fermentation process. Background Art

[0002] Kluyveromyces marxianus is a high-yielding protein fungus that produces proteins with the potential to replace some traditional meat and plant proteins. Because of its fast growth rate, wide carbon source utilization, high temperature resistance, and high-density fermentation capabilities, it is an important research subject in the field of alternative proteins. Through high-density fermentation technology, Kluyveromyces marxianus can efficiently convert nutrients to produce protein-rich biomass. It is a high-quality protein raw material that can be used to prepare high-fiber beverages, functional foods, pet food, and other high-protein products. Chinese invention patent ZL202410576609.7 authorizes a Kluyveromyces marxianus strain NS127 suitable for high-density culture and its application in single-cell protein, providing technical support for its industrialization. The rational development and application of Kluyveromyces marxianus strains can not only effectively alleviate my country's protein shortage problem, but also promote innovation and diversification in the food industry, showing broad market prospects.

[0003] During the high-density fermentation process of Kluyveromyces marxianus, protein content and biomass not only directly affect the fermentation efficiency, but also determine the controllability of the process and the quality of the final product. Although traditional determination methods are accurate and reliable, they have problems such as long detection cycle, dependence on chemical reagents, and easy environmental pollution, making it difficult to meet the needs of rapid and real-time monitoring. Near-infrared spectroscopy technology has been widely used in the field of traditional food testing because it does not require sample destruction, has fast detection speed, low cost, and can simultaneously determine multiple components. Chinese invention patent ZL202311679452.2 authorizes a rapid detection method for mycelial protein content. Based on near-infrared spectroscopy and chemometrics, mycelial protein content and moisture content can be non-destructively detected, but it is impossible to perform real-time detection of samples during the fermentation process. The present invention aims to develop an online detection method based on near-infrared spectroscopy technology and machine learning to achieve real-time monitoring of protein content and biomass during the fermentation process of Kluyveromyces marxianus, providing accurate and efficient technical support for fermentation process optimization and biomanufacturing. Summary of the Invention

[0004] In view of the problems that traditional methods for detecting protein content and biomass in high-density fermentation of Kluyveromyces marxianus are time-consuming, damage to samples, and environmental pollution caused by the use of chemical reagents, the present invention establishes a method for rapid, online detection of protein content and biomass based on near-infrared spectroscopy technology.

[0005] The present invention provides a method for establishing a rapid detection model for protein content and biomass during a fermentation process, which comprises the following steps: S1: Inoculate yeast seed liquid into fermentation medium and perform fermentation in a fermenter using a fed-batch culture method; use a spectrometer to collect raw near-infrared spectral data of the yeast in the fermenter at different culture time points, and collect fermentation liquid samples at the sampling port, and number and store the samples; wherein the spectral scanning range of the near-infrared spectrometer is 11111~5882cm -1 The method of using a spectrometer to collect raw near-infrared spectral data of yeast in a fermentation tank in situ comprises connecting the spectrometer to a light source, an optical fiber connection assembly, and a computer, wherein the other end of the optical fiber connection assembly is connected to a fiber optic probe, and the fiber optic probe is connected to the fermentation tank in situ so as to be located inside the fermentation tank and below the level of the fermentation liquid, and collecting spectral data in real time to the computer; S2 determines the protein content and biomass of yeast in the sample according to the national standard method; then the sample protein content, biomass and the original near-infrared spectral data collected in step S1 are matched one by one to establish a database; S3 uses one or more methods including standard normal transformation, multivariate scattering correction, trend correction, SG smoothing, wavelet transformation, normalization, derivative correction, and sliding smoothing filtering to preprocess the collected raw near-infrared spectral data; and uses one or more methods including competitive adaptive reweighted sampling, non-information variable elimination, genetic algorithm, principal component analysis, continuous projection algorithm, and decision tree algorithm to reduce data dimension and extract characteristic bands; S4 randomly divides the samples into a training set and a validation set: uses the random forest ensemble regression algorithm to fit the characteristic spectral data of the training set samples after principal component analysis as the independent variable and the protein content as the dependent variable, and establishes a model A for detecting the protein content in yeast fermentation samples. The spectral data and protein content data of the validation set samples are then used to verify the accuracy and stability of the model. If the correlation coefficient R>0.9 and the root mean square error RMSE<0.1, the model meets the requirements; S5 uses the random forest ensemble regression algorithm to fit the characteristic spectral data of the training set samples extracted by the uninformative variable elimination method as the independent variable and the biomass as the dependent variable to establish model B for detecting biomass in yeast fermentation sample liquid; the spectral data and biomass data of the validation set samples are then used to verify the accuracy and stability of the model. If the correlation coefficient R>0.9 and the root mean square error RMSE<1g / 100mL, the model meets the usage requirements.

[0006] Specifically, in step S1, in order to reduce the interference of the stirring paddle on the in-situ collection during the fermentation process, the spectral data of each sample were collected three times, and the average value was taken as the original near-infrared spectral data of the sample; The spectrometer is a fiber optic near-infrared spectrometer, the input end of which is connected to a computer via a USB or serial port, the light source is a halogen lamp light source, and the light source receiving end of the spectrometer is connected to the halogen lamp light source via a multimode Y-shaped optical fiber; The fiber optic probe is sealed and installed at the preset interface on the top of the fermentation tank through a standard interface component, so that the probe end can be directly inserted into the fermentation liquid inside the tank; The near-infrared light signal emitted by the light source is transmitted via a 6-core optical fiber to the near-infrared gold-plated reflective lens below the optical fiber probe. After total reflection, it penetrates the sample. The light signal carrying the sample component information is then transmitted back to the spectrometer via a 1-core optical fiber, achieving the purpose of using the spectrometer to collect the original near-infrared spectral data of the yeast in the fermentation tank in situ.

[0007] In step S3, the original near-infrared spectral data of the sample are preprocessed using SG smoothing and wavelet transform.

[0008] In the extraction of characteristic bands of protein content spectral data, the principal component analysis method was used. When the cumulative variance contribution rate reached 0.9999, the principal component at the front was selected as the characteristic wavelength related to protein content. In biomass, the uninformative variable elimination method was used, and the noise matrix size was set to the number of samples and the number of features, the noise mean was 1, and the standard deviation was 0.2 to screen characteristic bands as characteristic wavelengths related to biomass.

[0009] Specifically, in step S4, the SPXY algorithm is used to randomly divide the samples into a training set and a validation set.

[0010] More specifically, the Kluyveromyces marxianus is a high-protein-producing strain isolated and purified from traditional fermented dairy products, with a preservation number of CGMCC No. 30034.

[0011] The present invention further provides a model for detecting protein content and biomass during the high-density fermentation of Kluyveromyces marxianus obtained by the establishment method.

[0012] The present invention further provides a rapid detection method during yeast fermentation, which comprises the following steps: S1 In situ collection of raw near-infrared spectral data within the fermentation tank during the fermentation process; in situ collection of raw near-infrared spectral data of yeast within the fermentation tank using a spectrometer involves connecting the spectrometer to a light source, an optical fiber connection assembly, and a computer, with the other end of the optical fiber connection assembly connected to a fiber optic probe, which is in situ connected to the fermentation tank so as to be located within the fermentation tank and below the fermentation liquid level, and real-time spectral data is collected and transmitted to the computer; S2 inputs the collected original near-infrared spectral data into the yeast protein content detection model and biomass detection model, outputs the results, and obtains the bacterial protein content and biomass during the fermentation process.

[0013] Preferably, in step S2, the spectrometer is set to a mode width of 8.2, an average scan number of 8 times, and spectral data are collected three times in situ at 22° C. The average value is taken as the original spectral data of the sample; The spectrometer is a fiber optic near-infrared spectrometer, the input end of which is connected to a computer via a USB or serial port, the light source is a halogen lamp light source, and the light source receiving end of the spectrometer is connected to the halogen lamp light source via a multimode Y-shaped optical fiber; The fiber optic probe is sealed and installed at the preset interface on the top of the fermentation tank through a standard interface component, so that the probe end can be directly inserted into the fermentation liquid inside the tank; The near-infrared light signal emitted by the light source is transmitted via a 6-core optical fiber to the near-infrared gold-plated reflective lens below the optical fiber probe. After total reflection, it penetrates the sample. The light signal carrying the sample component information is then transmitted back to the spectrometer via a 1-core optical fiber, achieving the purpose of using the spectrometer to collect the original near-infrared spectral data of the yeast in the fermentation tank in situ.

[0014] The present invention also provides a method for screening, regulating and / or optimizing yeast fermentation process parameters, which comprises the following steps: S1 protein content and biomass obtained using the rapid detection method; S2 screens the optimal yeast fermentation process parameters based on the protein content and biomass data obtained in step S1, or regulates the yeast fermentation process parameters, or optimizes the fermentation process parameters.

[0015] Specifically, the screening, regulating and / or optimizing yeast fermentation process parameters is to dynamically adjust key parameters of the fermentation tank using a PLC control system, wherein the key parameters are selected from temperature, pH, and feed rate.

[0016] The present invention establishes a method for rapid, nondestructive detection of key parameters during yeast fermentation: protein content and biomass. An online monitoring system is also developed based on near-infrared spectroscopy. Using a near-infrared spectrometer and an optical fiber probe, yeast spectral data from different fermentation batches are collected in situ and samples are collected. Protein content and biomass are then measured. The collected raw spectral data are preprocessed using one or more of the following methods: standard normal transformation, multivariate scattering correction, trend correction, SG smoothing, wavelet transform, normalization, derivative correction, and sliding smoothing filtering. Principal component analysis and non-informative variable elimination are then used to reduce the dimensionality or extract characteristic bands from the spectral data for protein content and biomass, respectively. A random forest ensemble regression algorithm is then used to fit the characteristic spectral data to protein content and biomass, establishing a method for detecting protein content and biomass during high-density yeast fermentation. The model is embedded in a computer, combined with a spectrometer, a light source, and an optical fiber. The probe is then inserted into the fermentation tank to achieve in-situ continuous detection, forming a complete online monitoring system.

[0017] To address the challenges of protein content and biomass measurement methods during high-density fermentation, such as cumbersome sampling, long detection cycles, delayed results, high costs, and reliance on chemical reagents, this paper proposes an in-situ immersive real-time detection technology based on near-infrared spectroscopy and an embedded predictive model. This technology offers significant technological innovation and application advantages. This method utilizes in-situ spectral multi-band absorbance immersion acquisition within the tank, coupled with machine learning-based feature extraction and prediction models. This method enables one-click rapid analysis of multi-band absorbance data during fermentation, resulting in real-time or near-real-time output of protein content and biomass data. Complex sample pretreatment and chemical analysis are not required during the detection process, significantly shortening detection time and providing real-time dynamic feedback for process parameter control during yeast fermentation. Compared to existing technologies, this method significantly improves detection efficiency and intelligence. It can detect real-time data and promptly adjust parameters (such as feed rate, temperature, pH, and stirring speed) to optimize the process and enhance fermentation efficiency. It reduces reliance on expensive detection equipment and chemical reagents, as well as the lag associated with these measurements, significantly reducing operating costs. This method is suitable for online monitoring and control of large-scale, high-density fermentation production processes and has promising prospects for industrialization and expansion. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flow chart of the present invention.

[0019] Figure 2 This is a diagram of an online monitoring system, where: 1-near-infrared spectrometer, 2-halogen lamp light source, 3-optical fiber, 4-probe, 5-computer, 6-sample fermentation liquid, 7-fermentation tank, 8-PLC control cabinet.

[0020] Figure 3 It is the original spectrum data graph.

[0021] Figure 4 It is a scatter plot of the protein content calibration set and the validation set.

[0022] Figure 5 is a scatter plot of the biomass calibration set and the validation set.

[0023] Figure 6 It is a scatter plot of the protein content external new data test set.

[0024] Figure 7 is a scatter plot of the test set of new external biomass data. DETAILED DESCRIPTION

[0025] In order to make the purpose and technical steps of the present invention clearer, the present invention will be further described below with reference to the accompanying drawings and specific implementation methods. The examples are only for explaining the present invention but are not limited to the present invention. Unless otherwise specified, the methods in the examples are all conventional methods.

[0026] like Figure 1 As shown, the present application provides a method for rapid detection of protein content and biomass during high-density fermentation of Kluyveromyces marxianus, and simultaneously provides an online monitoring system, including: S1: Prepare seed culture medium, initial culture medium, and feed culture medium. After inoculating the seed culture medium, ferment on a shaker for 15 hours to prepare a seed broth. Transfer the broth to a 5 L fermentor containing the initial culture medium for high-density fermentation. At different fermentation time points, collect raw near-infrared spectral data of the fermentation broth in situ using an inserted fiber optic probe. Collect and number samples for future use.

[0027] S2: Measure the protein content and biomass in the sample according to the national standard method, and establish a database by corresponding them one by one with the near-infrared spectral data.

[0028] S3: The original spectra of the samples were subjected to different preprocessing methods, and then the characteristic bands were extracted using principal component analysis or non-informative variable elimination.

[0029] S4 used the SPXY algorithm to randomly divide the samples into training and validation sets; used the random forest ensemble regression algorithm to fit the protein content data of the training set and the corresponding characteristic band spectral data to establish model A for detecting the protein content of the Kluyveromyces marxianus fermentation process; used the random forest ensemble regression algorithm to fit the biomass data of the training set and the corresponding characteristic band spectral data to establish model B for detecting the biomass of the Kluyveromyces marxianus fermentation process.

[0030] S5 used the spectral data and protein content data of the validation set samples to verify the accuracy and stability of protein content detection model A in the fermentation process of Kluyveromyces marxianus; and used the spectral data and biomass data of the validation set samples to verify the accuracy and stability of biomass detection model B in the fermentation process of Kluyveromyces marxianus.

[0031] The S6 connects the spectrometer to a light source, optical fiber, and computer, and connects the fiber optic probe to the fermenter in situ through contact. The computer also embeds models for protein content and biomass detection, forming an online monitoring system. This system enables non-destructive, rapid, and continuous online measurement of protein content and biomass during high-density fermentation of Kluyveromyces marxianus. Based on the test results, the PLC control system promptly adjusts key fermenter process parameters, improving production efficiency, product quality, and cost savings.

[0032] Example 1 This example provides a scheme for constructing an online monitoring system for high-density fermentation of Kluyveromyces marxianus. Figure 2As shown, the system mainly includes a near-infrared spectrometer 1, a halogen lamp light source 2, an optical fiber 3 and a probe 4, a fermentation system (including fermentation liquid 6, a fermentation tank 7 and a PLC control cabinet 8), an optical fiber connection component connecting the optical fiber and the probe, and a computer 5 connected to control operations.

[0033] The near-infrared spectrometer 1 uses a Pynect fiber-optic near-infrared sensor with a high signal-to-noise ratio and a wide spectral response range. Its input is connected to a computer 5 via a USB or serial port for real-time spectral data acquisition and processing. The receiving end of the near-infrared spectrometer 1's halogen lamp 2 is connected to the halogen lamp 2 via a multimode Y-shaped optical fiber 3. Due to the dark color of the culture medium, a 40W halogen lamp was selected to provide stable light output, ensuring the strength and stability of the measurement signal. The probe 4 end of the optical fiber 3 is sealed and mounted to a pre-set port on the top of the fermenter 7 using a standard interface assembly (PG13.5 threaded connector), allowing the probe 4 end to be directly inserted into the fermentation broth 6 within the tank. The near-infrared light signal emitted by the halogen lamp 2 is transmitted via a six-core optical fiber 4 to a near-infrared gold-plated reflective lens located 5 mm below the probe 4 end. After total internal reflection, the light penetrates the fermentation broth 6 and undergoes absorption and scattering. The light signal, carrying information about the fermentation broth's components, is then transmitted back to the near-infrared spectrometer 1 via a single-core optical fiber 4, enabling in-situ, invasive, and non-destructive near-infrared spectral data acquisition during the fermentation process. The entire system boasts excellent sealing, strong anti-interference capabilities, and fast real-time response, making it suitable for monitoring key parameters in continuous, high-density fermentation processes.

[0034] Example 2 This example provides the preparation and method for a high-density fermentation experiment with Kluyveromyces marxianus and the national standard method for detecting protein content and biomass of Kluyveromyces marxianus samples. The specific experimental method is as follows: (1) Prepare culture media with different component contents. Seed culture medium: glucose 20 g / L, yeast extract 10 g / L, peptone 20 g / L. Initial culture medium: molasses 25 g / L, corn steep liquor 10-15 g / L, sulfuric acid 3-5 g / L, KH2PO4 1-2 g / L, MgSO4·7H2O 1 g / L, EDTA-2Na 0.015 g / L, CaCl2·2H2O 0.05 g / L, trace element stock solution 100 μl / L, vitamin stock solution 1 mL / L, 0.1% by volume defoamer. The trace element stock solution formula is: ZnSO4·7H2O 45g / L, MnCl2·7H2O 1g / L, CuSO4·5H2O 3g / L; Na2MoO4·2H2O 4g / L, FeSO4·7H2O 30g / L, KI 1g / L. The vitamin stock solution formula is: Biotin 0.1g / L, Calcium Pantothenate 1g / L, Niacin 1g / L. Feed medium: molasses 300 g / L, ammonium sulfate 100-120 g / L, KH2PO4 12-15 g / L, MgSO4·7H2O 3.90 g / L, EDTA-2Na 0.17 g / L, trace element mother solution CaCl2·2H2O 0.59 g / L, ZnSO4·7H2O mother solution 0.613 ml / L, FeSO4·7H2O 5.747 ml / L, 1 ml / L of other trace element mother solutions, 10 mL / L of vitamin mother solution.

[0035] (2) Seed solution preparation: Activated Kluyveromyces marxianus cells were inoculated into the sterilized seed culture medium, and then shaker fermentation culture was carried out. The culture temperature was set at 30 °C and the shaker speed was set at 200 rpm. After culturing for 15 h, the seed solution was obtained.

[0036] (3) High-density fermentation in a fermenter: In a 5L fermenter system, the initial culture medium and the feed medium were mixed in a ratio of 4:3. First, the seed liquid was inoculated into the fermenter containing the initial culture medium. Subsequently, the fermenter, feed medium, defoamer, PLC control system, condensation system, vacuum pump and other equipment were connected according to the specifications. Before the start of fermentation, the dissolved oxygen electrode, pH electrode and feed pump rate were calibrated. Through the PLC control system, the culture temperature was set to 30 °C, the stirring speed was controlled at 400-800 rpm, the tank pressure was maintained at 0.08 MPa, and the initial ventilation volume was set to 1 L / (L·min). After the fermentation was started, the feed rate and feeding time were controlled in stages using a sequential control table, and the linkage control of dissolved oxygen and stirring speed was set to adjust the feed rate to maintain the dissolved oxygen level between 10% and 30%. When the feed medium was exhausted, the dissolved oxygen level rose rapidly to 80%. At this time, the fermentation time was about 36-40 h, and the fermentation was terminated.

[0037] (4) Determination of protein content and biomass: The protein content was determined according to the combustion method in the national standard GB 5009.5-2016. The specific steps are as follows: the sample was placed in a drying oven and then crushed. 50-150 mg of the standard and sample were weighed and wrapped with tin foil. The weight was recorded and placed on the sample tray. The weight information was entered into the computer software. After the sample was fully burned, the computer software displayed the initial protein content of the sample. The correction coefficient was calculated using the standard to correct the final protein content of the sample. The biomass was dried according to the national standard GB 5009.3-2016 and weighed and converted. The specific steps are as follows: the sample was transferred to a centrifuge tube using a 5 ml pipette and placed in a high-speed centrifuge at 7500 rpm for 5 min. The supernatant after centrifugation was discarded and 5 ml of ultrapure water was added for washing and centrifugation again. After centrifugation, the supernatant was discarded and placed in a drying oven until dry. The sample was placed on a balance for weighing and conversion to obtain the biomass.

[0038] Example 3 This embodiment provides a method for collecting and preprocessing near-infrared spectral data during high-density fermentation of Kluyveromyces marxianus. Relying on the online monitoring system hardware system described in Example 1, the specific process for collecting in-situ near-infrared spectral data during the fermentation process is as follows: Before sterilizing the fermenter, turn on the light source for 15 minutes to preheat the light source to ensure the stability of the light intensity and avoid spectral shift during the data collection process. Start the spectrometer and set the collection range to 11111~5882cm -1 , the exposure time was 0.635ms, and the average number of acquisitions was 10. After preheating was completed, the air was collected as a white reference spectrum and saved. During the fermentation process, after setting the feeding rate to the sequential control table and entering a stable operating state, three spectra were collected continuously every 30 minutes, and the average value was taken as the original spectral data of the sample to reduce the interference of the stirring paddle movement on the spectral stability. Finally, a total of 210 groups of samples were collected under different fermentation batches, and each group of samples contained 228 bands of original spectral information ( Figure 3 To improve the modeling accuracy, after comparing the modeling accuracy of various pre-processing methods, the collected spectral data were pre-processed using SG smoothing combined with wavelet transform to effectively remove noise and baseline drift.

[0039] Example 4

[0040] This example provides a method for extracting characteristic wavelengths for protein content and biomass during high-density fermentation of Kluyveromyces marxianus and establishing a detection model. The spectral data and sample measurements used were derived from the 210 high-density fermentation samples described in Example 3. The preprocessed near-infrared spectral data, along with the corresponding protein content and biomass physicochemical indicators, constitute the model input and output datasets.

[0041] First, the sample data set was divided into a calibration set and a validation set according to the ratio of 3:1 using the SPXY algorithm, with 158 samples in the calibration set and 52 samples in the validation set. In the dimensionality reduction of the characteristic bands of the protein content spectral data, the principal component analysis dimensionality reduction method was used. When the cumulative variance contribution rate reached 0.9999, the principal component at the front was selected as the characteristic wavelength related to the protein content. There were 50 principal components in total, namely pc1~pc50. The principal components with a variance contribution rate higher than 0.1% were selected to reduce the interference of irrelevant variables and reduce the complexity, namely pc1~pc13. The variance contribution rates of the first 15 principal components are shown in Table 1. The random forest ensemble regression algorithm was used to extract 13 principal components of the characteristic wavelengths of the training set samples through principal component analysis as independent variables, and protein content as the dependent variable. A model A for detecting protein content in the high-density fermentation process of Kluyveromyces marxianus was established. The spectral data and protein content data of the validation set samples were used to verify the accuracy and stability of the model. If the correlation coefficient R>0.9 and the root mean square error RMSE<0.1, the model meets the requirements. In the specific experiment, such as Figure 4 As shown, the correlation coefficient between the true value of protein content in the calibration set and the model predicted value is 0.986, and the root mean square error is 0.011. The correlation coefficient between the true value of protein content in the validation set and the model predicted value is 0.940, and the root mean square error is 0.017.

[0042] Table 1. Variance contribution of the top 15 principal components

[0043] In the biomass, the non-information variable elimination method was used to introduce noise and evaluate the stability of each feature through leave-one-out cross-validation. The features with less than the maximum stability value in the noise data were gradually eliminated. The size of the noise matrix was set to the number of samples and the number of features, with a noise mean of 1 and a standard deviation of 0.2. The stability value was calculated and the characteristic bands were screened. Among them, the bands with stability lower than 16.962 were eliminated, and a total of 87 characteristic bands were screened, with a total band spectrum range of 11111~5882cm -1 The distribution range of the 87 characteristic bands is 11025~10535cm -1 7687~7349cm -1 7093~6887cm -1 、6630~6116cm -1 and 5988~5913cm -1 The random forest ensemble regression algorithm was used to take the 87 characteristic wavelength spectral data extracted from the training set samples by the non-information variable elimination method as the independent variable and the biomass as the dependent variable to establish the biomass model B for detecting the fermentation process of Kluyveromyces marxianus. The spectral data and biomass data of the validation set samples were then used to verify the accuracy and stability of the model. Figure 5As shown, the correlation coefficient between the true value of the biomass in the calibration set and the model predicted value is 0.983, the root mean square error is 0.510 g / 100 mL, and the correlation coefficient between the true value of the biomass in the validation set and the model predicted value is 0.984, the root mean square error is 0.485 g / 100 mL.

[0044] Example 5

[0045] This example is used to verify the accuracy and stability of the protein content and biomass detection model and online monitoring system during the high-density fermentation of Kluyveromyces marxianus in Examples 1 and 4.

[0046] Spectrometer is connected to light source, optical fiber and computer, and optical fiber probe is contact-connected to fermentation tank in situ, detection model A and B in embodiment 4 are embedded in computer, and this system can drive spectrometer to collect spectral data in Kluyveromyces marxianus fermentation process and detect its protein content and biomass by one key.In different batches of fermentation processes, the growth of yeast is all different, therefore according to the high-density fermentation method of embodiment 2, the cultivation of Kluyveromyces marxianus is carried out again, the collection of spectral data of different time nodes in the fermentation process is carried out using online monitoring system and the protein content and biomass of its corresponding moment are detected, synchronously sample and preserve sample, measure its protein content and biomass according to the national standard method in embodiment 2. Collected 34 samples in total.

[0047] like Figure 6 Shown is a scatter plot of the protein content predicted by the model and the protein content detected by the national standard, with a correlation coefficient of 0.929 and a root mean square error of 0.018.

[0048] like Figure 7 Shown is a scatter plot of the biomass predicted by the model and the biomass measured by the national standard. The correlation coefficient is 0.979, and the root mean square error is 0.481 g / 100 mL. The above method can timely adjust fermentation parameters, providing a reference and solution for process screening, control, and optimization during the fermentation process.

[0049] The above embodiments represent preferred embodiments of the present application and are not intended to be limiting. Any modifications or alternatives within the technical scope disclosed in this application that can be readily conceived by a person skilled in the art should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope specified in the claims.

Claims

1. A method for establishing a rapid detection model for protein content and biomass during fermentation, characterized in that: The following steps are involved: S1: Inoculate yeast seed liquid into fermentation medium and perform fermentation in a fermenter using a fed-batch culture method; use a spectrometer to collect raw near-infrared spectral data of the yeast in the fermenter at different culture time points, and collect fermentation liquid samples at the sampling port, and number and store the samples; wherein the spectral scanning range of the spectrometer is 11111~5882cm -1 The method of using a spectrometer to collect raw near-infrared spectral data of yeast in a fermentation tank in situ comprises connecting the spectrometer to a light source, an optical fiber connection assembly, and a computer, wherein the other end of the optical fiber connection assembly is connected to a fiber optic probe, and the fiber optic probe is connected to the fermentation tank in situ so as to be located inside the fermentation tank and below the level of the fermentation liquid, and collecting spectral data in real time to the computer; S2 determines the protein content and biomass of yeast in the sample according to the national standard method; then the sample protein content, biomass and the original near-infrared spectral data collected in step S1 are matched one by one to establish a database; S3 uses one or more methods including standard normal transformation, multivariate scattering correction, trend correction, SG smoothing, wavelet transformation, normalization, derivative correction, and sliding smoothing filtering to preprocess the collected raw near-infrared spectral data; and uses one or more methods including competitive adaptive reweighted sampling, non-information variable elimination, genetic algorithm, principal component analysis, continuous projection algorithm, and decision tree algorithm to reduce data dimension and extract characteristic bands; S4 randomly divides the samples into a training set and a validation set: uses the random forest ensemble regression algorithm to fit the characteristic spectral data of the training set samples after principal component analysis as the independent variable and the protein content as the dependent variable, and establishes a model A for detecting the protein content in yeast fermentation samples. The spectral data and protein content data of the validation set samples are then used to verify the accuracy and stability of the model. If the correlation coefficient R>0.9 and the root mean square error RMSE<0.1, the model meets the requirements; S5 uses the random forest ensemble regression algorithm to fit the characteristic spectral data of the training set samples extracted by the uninformative variable elimination method as the independent variable and the biomass as the dependent variable to establish model B for detecting biomass in yeast fermentation sample liquid; the spectral data and biomass data of the validation set samples are then used to verify the accuracy and stability of the model. If the correlation coefficient R>0.9 and the root mean square error RMSE<1g / 100mL, the model meets the usage requirements.

2. The establishment method according to claim 1, characterized in that In step S1, spectral data of each sample was collected three times, and the average value was taken as the original near-infrared spectral data of the sample; The spectrometer is a fiber optic near-infrared spectrometer, the input end of which is connected to a computer via a USB or serial port, the light source is a halogen lamp light source, and the light source receiving end of the spectrometer is connected to the halogen lamp light source via a multimode Y-shaped optical fiber; The fiber optic probe is sealed and installed at the preset interface on the top of the fermentation tank through a standard interface component, so that the probe end can be directly inserted into the fermentation liquid inside the tank; The near-infrared light signal emitted by the light source is transmitted via a 6-core optical fiber to the near-infrared gold-plated reflective lens below the optical fiber probe. After total reflection, it penetrates the sample. The light signal carrying the sample component information is then transmitted back to the spectrometer via a 1-core optical fiber, achieving the purpose of using the spectrometer to collect the original near-infrared spectral data of the yeast in the fermentation tank in situ.

3. The establishment method according to claim 1, characterized in that In step S3, the original near-infrared spectral data of the sample are preprocessed using SG smoothing and wavelet transform; In the extraction of characteristic bands of protein content spectral data, the principal component analysis method was used. When the cumulative variance contribution rate reached 0.9999, the principal component at the front was selected as the characteristic wavelength related to protein content. In biomass, the uninformative variable elimination method was used, and the noise matrix size was set to the number of samples and the number of features, the noise mean was 1, and the standard deviation was 0.2 to screen characteristic bands as characteristic wavelengths related to biomass.

4. The establishment method according to claim 1, characterized in that In step S4, the SPXY algorithm is used to randomly divide the samples into training set and validation set.

5. The establishment method according to any one of claims 1 to 4, characterized in that: The yeast is Kluyveromyces marxianus, which is a high-protein-yielding strain isolated and purified from traditional fermented dairy products, and has a preservation number of CGMCC No. 30034.

6. A model for detecting protein content and biomass during high-density fermentation of Kluyveromyces marxianus obtained by the method according to any one of claims 1 to 5.

7. A method for rapid detection of yeast fermentation using the model according to claim 6, characterized in that: The steps include: S1 In situ collection of raw near-infrared spectral data within the fermentation tank during the fermentation process; in situ collection of raw near-infrared spectral data of yeast within the fermentation tank using a spectrometer involves connecting the spectrometer to a light source, an optical fiber connection assembly, and a computer, with the other end of the optical fiber connection assembly connected to a fiber optic probe, which is in situ connected to the fermentation tank so as to be located within the fermentation tank and below the fermentation liquid level, and real-time spectral data is collected and transmitted to the computer; S2 inputs the collected original near-infrared spectral data into the yeast protein content detection model and biomass model, outputs the results, and obtains the bacterial protein content and biomass during the fermentation process.

8. The rapid detection method according to claim 7, wherein In step S2, the spectrometer was set to a mode width of 8.2, an average scan number of 8 times, and spectral data were collected three times in situ at 22°C. The average value was taken as the original spectral data of the sample; The spectrometer is a fiber optic near-infrared spectrometer, the input end of which is connected to a computer via a USB or serial port, the light source is a halogen lamp light source, and the light source receiving end of the spectrometer is connected to the halogen lamp light source via a multimode Y-shaped optical fiber; The fiber optic probe is sealed and installed at the preset interface on the top of the fermentation tank through a standard interface component, so that the probe end can be directly inserted into the fermentation liquid inside the tank; The near-infrared light signal emitted by the light source is transmitted via a 6-core optical fiber to the near-infrared gold-plated reflective lens below the optical fiber probe. After total reflection, it penetrates the sample. The light signal carrying the sample component information is then transmitted back to the spectrometer via a 1-core optical fiber, achieving the purpose of using the spectrometer to collect the original near-infrared spectral data of the yeast in the fermentation tank in situ.

9. A method for screening, regulating and / or optimizing yeast fermentation process parameters, characterized in that: S1 protein content and biomass obtained by the rapid detection method according to claim 7 or 8; S2 screens the optimal yeast fermentation process parameters based on the protein content and biomass data obtained in step S1, or regulates the yeast fermentation process parameters, or optimizes the fermentation process parameters.

10. The method according to claim 9, wherein The screening, regulation and / or optimization of yeast fermentation process parameters is to dynamically adjust key parameters of the fermentation tank using a PLC control system, wherein the key parameters are selected from temperature, pH, and feed rate.

Images