Microbial fermentation strain breeding control system based on neural network model

Abstract

The invention discloses a microbial fermentation strain breeding control system based on a neural network model, and relates to the technical field of digital fermentation, and the system comprises a thallus data acquisition module, a biological feature recognition module, a prediction classification module, a breeding decision module, an optimization training module, an early warning monitoring module and a sorting regulation module. The bacterial data acquisition module preprocesses bacterial data; the biological feature recognition module constructs a bacterial feature library; the prediction classification module predicts the adaptability of bacterial strains and classifies the bacterial strains; the selection decision module selects the bacterial strains and generates culture parameters; the sorting regulation and control module generates a three-dimensional scatter diagram and a hereditary character report. According to the method, the precision and efficiency of breeding of the microbial fermentation strain are remarkably improved, intelligence and automation of the breeding process are achieved, the labor cost is reduced, the stability and controllability of the fermentation process are enhanced, and powerful technical support is provided for innovative development of the microbial fermentation industry.

Description

Technical Field

[0001] The present invention relates to the technical field of digital fermentation, and particularly relates to a control system for the selection and breeding of microbial fermentation strains based on a neural network model. Background Art

[0002] The selection and breeding of microbial fermentation strains refers to the process of screening and optimizing microbial strains suitable for specific fermentation processes through a series of scientific methods and means. This process aims to improve the production efficiency of microorganisms, the quality and stability of products, and the adaptation to various conditions of industrial production. In the selection and breeding process, first, a strain library with different genetic characteristics needs to be obtained through means such as natural isolation or artificial mutagenesis. Then, screening techniques, such as indicators like growth rate, metabolite yield, and tolerance, are used to screen out excellent strains that meet production requirements. In addition, gene engineering techniques can be used to genetically modify the strains to further improve their performance. The selected excellent strains can produce target products more efficiently during the fermentation process, while reducing the generation of by-products and improving the overall production efficiency. This is of great significance for enhancing the competitiveness of the microbial fermentation industry and promoting the development of related industries. Therefore, the selection and breeding of microbial fermentation strains is a key link in microbial fermentation technology and an important driving force for promoting the progress of biotechnology.

[0003] In order to solve the problems of strong dependence on manual experience and low screening efficiency currently existing in the selection and breeding of microbial fermentation strains, the existing technology mainly relies on traditional screening methods and manual judgment for processing. However, this method highly depends on the experience and intuition of researchers, which is not only time-consuming and laborious, but also prone to unstable screening results and poor consistency. At the same time, due to the low screening efficiency, it is difficult to meet the needs of large-scale industrial production, which further leads to an increase in production costs and restricts the wide application of microbial fermentation technology. In view of this, a control system for the selection and breeding of microbial fermentation strains based on a neural network model is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a control system for the selection and breeding of microbial fermentation strains based on a neural network model to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: A control system for the selection and breeding of microbial fermentation strains based on a neural network model, including a bacterial cell data acquisition module, a biological feature recognition module, a prediction and classification module, a selection and breeding decision module, an optimization and training module, an early warning and monitoring module, and a sorting and control module;

[0006] The bacterial cell data acquisition module collects and preprocesses microbial fermentation strain data;

[0007] The biometric recognition module performs biometric recognition on the preprocessed microbial fermentation strain data to construct a strain feature library;

[0008] The prediction and classification module is used to construct a fitness prediction model and a microbial community metabolic network topology map in combination with the strain feature library, predict the strain fitness, and classify according to the strain yield;

[0009] The strain selection decision module combines the strain fitness and the strain classification results through weighted combination, constructs a strain selection decision model, and outputs the optimal strain number and culture parameters;

[0010] The optimization and training module optimizes the hyperparameters of the fitness prediction model, the microbial community metabolic network topology map, and the strain selection decision model;

[0011] The early warning and monitoring module sets a lower limit for the warning of the microbial community metabolic rate and a threshold for abnormal fluctuations in the cell density, thereby triggering a strain elimination warning;

[0012] The sorting and regulation module visualizes the distribution of the main components of the strain and the genetic characteristic report through a three-dimensional scatter plot, triggers the automatic sorter to perform strain optimization, and synchronously regulates the culture parameters.

[0013] A further improvement of the technical solution of the present invention lies in that: in the cell data acquisition module, the process of collecting and preprocessing the microbial fermentation strain data includes:

[0014] Obtain microbial fermentation strain data through a voltammetric biosensor probe and a near-infrared spectroscopy analyzer. Among them, the microbial fermentation strain data includes metabolite concentration data and cell density data in the microbial fermentation broth;

[0015] Deploy voltammetric biosensor probes symmetrically on the inner wall of the fermentation tank, adopt a three-electrode system, apply a stepped voltage through a potentiostat, collect redox current signals, and real-time detect metabolite concentration data in the microbial fermentation broth. Among them, the distance between the voltammetric biosensor probes is less than 30 cm, the sampling frequency is set to 1 Hz, the three-electrode system includes a working electrode, a reference electrode, and a counter electrode, and the metabolites include lactic acid and ethanol;

[0016] Embed a near-infrared spectroscopy analyzer in the fermentation broth circulation pipeline, install the light source and the detector at a 45° angle, calculate the absorbance value of the cells based on the Lambert-Beer law, construct a near-infrared spectroscopy absorbance sequence, and invert the cell density. Among them, the optical path is fixed at 10 mm, the calibration whiteboard is placed on the opposite side of the detection window, and baseline calibration is performed before starting the machine every day;

[0017] Using wavelet threshold denoising technology, perform three-layer wavelet decomposition on the original redox current signal. Through the soft threshold function, set the noise threshold as the square root logarithm of the noise standard deviation multiplied by the signal length to eliminate high-frequency noise. Based on the local outlier factor algorithm, calculate the density ratio of the data point neighborhood, mark the data deviating from the density distribution as abnormal, and interpolate and complete the abnormal data with a sliding window.

[0018] Perform sliding window smoothing filtering on the near-infrared spectrum absorbance sequence to eliminate the jitter noise of the near-infrared spectrum analyzer. Dynamically adjust the data sampling interval according to the fermentation broth flow rate. When the flow rate is greater than 1.5 m / s, enable the inertial compensation algorithm, and use the product of the previous moment's cell density value plus the compensation coefficient and the flow rate change as the current cell density value.

[0019] A further improvement of the technical solution of the present invention lies in that: in the biometric recognition module, the process of constructing the strain feature library includes:

[0020] For the preprocessed metabolite concentration data, according to a fixed time window Calculate the metabolic rates of lactic acid and ethanol respectively r and V y , use the dynamic time warping technology to align the metabolite concentration curves of different batches of strains, calculate the cumulative distance of the minimum path between the curves , and obtain the growth curve similarity , and its calculation process is as follows:

[0021] ;

[0022] ;

[0023] where, is the metabolic rate, is the metabolite concentration at the current moment, and are the lengths of curve A and curve B respectively. If the growth curve similarity is greater than 85%, it is marked as a strain with stable growth. If the growth curve similarity is less than 85%, it is marked as a strain with abnormal metabolism;

[0024] Perform principal component analysis on the near-infrared spectrum absorbance sequence, and extract the first three principal components PC1, PC2, and PC3. The calculation process is as follows:

[0025] ;

[0026] where k = 1, 2, and 3, is the weight coefficient of the kth principal component at wavelength i, is the absorbance value corresponding to the wavelength, is the number of spectral wavelengths in the range of 900 - 1700 nm. According to the clustering of the main component spatial distribution density, the fermentation stage is divided into the initial stage, logarithmic growth stage, and decline stage, and the corresponding main component threshold ranges are matched for different growth stages;

[0027] Encode the lactic acid and ethanol metabolic rates, growth curve similarity, PC1, PC2, and PC3 in the format of strain number - timestamp - feature vector, integrate them into a strain characteristic library. When adding new strain data, calculate its Euclidean distance from the existing strains in the strain characteristic library. If the Euclidean distance is greater than 2, create a new feature entry. After retraining the historical strain data, recalculate every 7 days , and update the main component eigenvalue in the characteristic library.

[0028] A further improvement of the technical solution of the present invention lies in: in the prediction and classification module, a fitness prediction model is constructed. The process of predicting the strain fitness includes:

[0029] Align the lactic acid and ethanol metabolic rates with the growth curve similarity according to the time step, construct a 3D input vector to cover the time series data of the entire 0 - 72 hours of fermentation process, and generate strain historical fitness labels based on the growth stability and product yield of the strain under specific culture conditions;

[0030] Adopt a bidirectional gated recurrent unit network architecture, set the dimensions of the forward hidden state and backward hidden state in its hidden layer to be 64 both, construct a fitness prediction model. The input layer of the fitness prediction model receives a three-dimensional feature vector, the hidden layer controls the information flow through the reset gate and update gate, the output layer concatenates the bidirectional hidden states and maps them to the strain fitness D through a fully connected layer. The loss function uses the mean square error to calculate the difference between the predicted strain fitness and the strain historical fitness label.

[0031] A further improvement of the technical solution of the present invention lies in: in the prediction and classification module, a microbial community metabolic network topology graph is constructed. The process of classifying according to the strain yield includes:

[0032] Set classification labels for the strains based on historical data. The strain classification labels include high yield, medium yield, low yield, and metabolic abnormality. Using the graph convolutional network architecture, take the first 3 - order main components PC1, PC2, and PC3 of each strain as the node feature vectors and input them into the input layer of the graph convolutional network architecture to construct a microbial community metabolic network topology graph. Construct an adjacency matrix based on the main component spatial distance. If the Euclidean distance of the main components of two strains is less than 1.5, it is defined that there is an edge between the nodes corresponding to the two strains;

[0033] The hidden layer of the microbial community metabolic network topology graph adopts two - layer graph convolutional operations, with 16 dimensions output for each layer. The node feature propagation is realized by normalizing the self - connected adjacency matrix and degree matrix. The output layer aggregates the node features through global max - pooling to generate a 3D probability vector ;

[0034] Set the upper limit of the strain yield probability threshold to 0.7 and the lower limit of the strain yield threshold to 0.5. If in the output 3D probability vector is greater than 0.7, then mark the strain as high-yield. If and is the highest among the three, then mark the strain as medium-yield. If and is the highest among the three, then mark the strain as low-yield;

[0035] If the difference between the probabilities of any two of them is less than 0.1, then determine the category according to the proximity of the principal component space distance to the historical high-yield cluster. If the three types of probabilities of the strain are all less than 0.5, then mark it as a candidate strain with abnormal metabolism. If the Euclidean distance of the principal component features of the strain is greater than 3, then temporarily store it as an unclassified strain, and dynamically adjust the upper and lower limits of the strain yield probability threshold based on the proportion of high-yield strains in the historical data.

[0036] A further improvement of the technical solution of the present invention lies in: in the strain selection decision module, the process of constructing a strain selection decision model and outputting the optimal strain number and culture parameters includes:

[0037] Dynamically adjust the weight coefficients of fitness and classification results according to the strain classification results, construct a strain selection decision model. If the strain is classified as high-yield, then give priority to the strain classification results. If the strain is classified as medium-yield and low-yield, then focus on fitness. Based on the weight coefficient of fitness and the weight coefficient of classification results, calculate the classification gain Z and the comprehensive score F, and the calculation process is as follows:

[0038] ;

[0039] ;

[0040] Among them, is the weight coefficient of fitness, is the weight coefficient of the strain classification result;

[0041] Set the upper and lower limits of the threshold of the comprehensive score to 0.8 and 0.6 respectively. If the comprehensive score ≥ 0.8, then the strain enters the preferred queue. If 0.6 ≤ comprehensive score < 0.8, then the strain enters the queue to be observed. If the comprehensive score < 0.6, then the strain is marked as a candidate for elimination. If the strain is marked as a candidate strain with abnormal metabolism, then push it to the warning monitoring module;

[0042] Introduce and record the pH and feeding rate of the strain under the optimal culture conditions The historical parameter mapping table is used to match the strain culture parameters with the same classification result and the highest fitness from the historical parameter mapping table, and update the pH and feeding rate. The calculation process is as follows:

[0043] ;

[0044] ;

[0045] Among them, and represent the reference pH and reference rate, which are their historical optimal values respectively.

[0046] A further improvement of the technical solution of the present invention lies in that: in the optimization training module, the process of optimizing the hyperparameters of the fitness prediction model, the microbial community metabolic network topology map, and the strain selection decision model includes:

[0047] The fitness prediction model dynamically adjusts its learning rate through the cosine annealing strategy. The initial learning rate is set to 0.001, the learning rate threshold is 0.0001, and it decays according to the cosine function as the number of training rounds increases. An early stopping mechanism is set. If the validation set loss of the fitness prediction model does not decrease for 10 consecutive rounds, the training is terminated to prevent overfitting;

[0048] Add the principal component features of the new strain to the topology map, recalculate the adjacency matrix. If the maximum Euclidean distance between the newly added node and the existing nodes exceeds 2, expand the network structure. When the number of newly added data reaches 100 per batch, use the incremental learning algorithm to update the parameters of the graph convolutional network, and use the cross-entropy loss function to optimize the microbial community metabolic network topology map;

[0049] When the mean square error between the actual yield of the preferred strain per batch and the comprehensive score predicted by the strain selection decision model exceeds 0.1, update the weight coefficient of the fitness according to the gradient descent method. Add the actual culture parameters and yield data of the preferred strain to the historical parameter mapping table, and update the reference pH value and reference rate value every 7 days.

[0050] A further improvement of the technical solution of the present invention lies in that: in the early warning monitoring module, the process of setting the lower limit of the microbial community metabolic rate warning and the abnormal fluctuation threshold of the cell density to trigger the strain elimination warning includes:

[0051] Based on the metabolic rate distribution of historical high-yield strains, select the 5th percentile as the lower limit of the microbial community metabolic rate warning. After adding high-yield strain data per batch, recalculate the lower limit of the microbial community metabolic rate warning according to the sliding window. If the deviation between the current value and the original threshold > 10%, update it as the microbial community metabolic rate threshold;

[0052] Calculate the percentage of the standard deviation to the mean of the bacterial density within the time window to obtain the coefficient of variation of the bacterial density. Set the abnormal fluctuation thresholds of the bacterial density to 25%, 20%, and 10% respectively for the initial stage, logarithmic growth phase, and decline phase of fermentation. If PC1 > 0.7 and PC2 < 0.3, it is determined that the bacterial community is in the initial stage of fermentation. If PC2 > 0.5 and PC3 < 0.2, it is determined that the bacterial community is in the logarithmic growth phase. If PC3 > 0.6 and PC1 < 0.4, it is determined that the bacterial community is in the decline phase. When the coefficient of variation of the bacterial density exceeds the abnormal fluctuation threshold of the bacterial density in the corresponding fermentation stage, it is regarded as an abnormal fluctuation;

[0053] For the strains in the initial stage of fermentation, if the initial metabolic rate is lower than the warning lower limit of the metabolic rate of the current bacterial community and the coefficient of variation of the bacterial density exceeds 25%, they are marked as growth-delayed strains and pushed to the manual review queue. If the strains in the initial stage of fermentation do not enter the logarithmic growth phase for more than 72 hours, a metabolic stagnation warning is triggered, and the fermentation is terminated and sampled for analysis;

[0054] Mark abnormal strains according to the warning lower limit of the bacterial community metabolic rate and the abnormal fluctuation threshold of the bacterial density, and re-verify the deviation degree of the principal component space. If the strains are confirmed to be abnormal, the culture is terminated and the culture parameters are adjusted, and the strain characteristic library is updated synchronously.

[0055] A further improvement of the technical solution of the present invention lies in: in the sorting and regulation module, the process of visualizing the principal component distribution of strains through a three-dimensional scatter plot and marking high-yield clusters and generating a genetic characteristic comparison report includes:

[0056] Taking PC1, PC2, and PC3 of the strains as the three-dimensional coordinate axes, constructing a scatter plot, where each data point represents a strain. Identify high-yield clusters through a density clustering algorithm, calculate the vertex coordinates of the convex hull and generate the cluster boundary, and distinguish different strain classification labels with different color coding, and dynamically mark the spatial distribution of high-yield clusters, where red represents high-yield strains, yellow represents medium-yield strains, blue represents low-yield strains, and gray represents metabolically abnormal strains;

[0057] Statistically analyze the differences in the expression levels of metabolic pathway genes between high-yield clusters and all strains in the library, calculate the enrichment degree of metabolic pathways using hypergeometric tests, screen out significant pathways with an enrichment degree of metabolic pathways lower than 0.05, statistically analyze the mean and standard deviation of the key enzyme activities, mark highly expressed enzymes with activities greater than twice the standard deviation of the mean, and generate a comprehensive report including a three-dimensional scatter plot, a pathway enrichment table, and an enzyme activity comparison bar chart.

[0058] A further improvement of the technical solution of the present invention lies in: in the sorting and regulation module, the process of driving the automatic sorter and the fermentation parameter regulation system to perform strain optimization and pH and feeding rate adjustment includes:

[0059] Sort the strains in the preferred queue in descending order of the comprehensive score. First, select the strains ranked in the top 10% of the scores. If there are multiple strains in the same score range, sort them according to the proximity of the principal component space distance to the historical high-yield cluster. The sorter locates the target strains according to the strain numbers using microfluidic chip sorting technology. After sorting, the strains are automatically inoculated into a new fermenter, and the sorting timestamp and culture parameters are recorded synchronously.

[0060] If the strain is classified as high-yield, adjust the pH normally. If the strain is in the logarithmic growth phase, additionally increase the pH compensation amount by 0.05. Match the basic feeding rate according to the historical parameter mapping table. If the key enzyme activity is detected to be greater than 2 standard deviations of the mean, additionally increase the feeding rate by 10%.

[0061] Terminate the cultivation of the candidate strains with abnormal metabolism in the fermenter, turn off the feeding pump and start the in-tank cleaning program. Record the principal component characteristics and operation logs of the abnormal strains, push them to the strain characteristics library and mark them as eliminated data. After each batch of sorting, update the weight coefficient of the fitness according to the gradient descent method based on the difference between the actual yield and the predicted score. Remove the invalid parameter entries with an actual yield lower than 20% of the mean of the same type from the historical parameter mapping table every 7 days.

[0062] Due to the adoption of the above technical solution, the technical progress achieved by the present invention compared with the prior art is:

[0063] 1. The present invention provides a microbial fermentation strain breeding control system based on a neural network model, which significantly improves the accuracy and efficiency of strain breeding. Through intelligent processing and analysis, the breeding cycle is greatly shortened and the labor cost is reduced.

[0064] 2. The present invention provides a microbial fermentation strain breeding control system based on a neural network model, which realizes the full automation and real-time monitoring of the breeding process, effectively avoids human errors, and enhances the stability and reliability of the breeding results.

[0065] 3. The present invention provides a microbial fermentation strain breeding control system based on a neural network model. By continuously iteratively upgrading the model parameters through the optimization training module, it ensures that the system always maintains the optimal state, providing a solid foundation for the continuous innovation and technological upgrading of the microbial fermentation industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0067] Figure 1 It is a block diagram of the present invention. Detailed implementation manners

[0068] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0069] Embodiment, such as Figure 1 As shown, the present invention provides a microbial fermentation strain breeding control system based on a neural network model, including a cell data acquisition module, a biological feature recognition module, a prediction and classification module, a breeding decision module, an optimization and training module, an early warning and monitoring module, and a sorting and regulation module;

[0070] The cell data acquisition module collects and preprocesses microbial fermentation strain data, and obtains microbial fermentation strain data through a voltammetric biosensor probe and a near-infrared spectroscopy analyzer. Among them, the microbial fermentation strain data includes metabolite concentration data and cell density data in the microbial fermentation broth. Voltammetric biosensor probes are symmetrically deployed on the inner wall of the fermentation tank, and a three-electrode system is adopted. A stepped voltage is applied through a potentiostat to collect redox current signals and real-time detect metabolite concentration data in the microbial fermentation broth. Among them, the distance between the voltammetric biosensor probes is less than 30 cm, and the sampling frequency is set to 1 Hz. The three-electrode system includes a working electrode, a reference electrode, and a counter electrode. The metabolites include lactic acid and ethanol. A near-infrared spectroscopy analyzer is embedded in the fermentation broth circulation pipeline, and the light source and the detector are installed at a 45° angle. Based on the Lambert-Beer law, the absorbance value of the cells is calculated to construct a near-infrared spectroscopy absorbance sequence, and the cell density is inverted. Among them, the optical path is fixed at 10 mm, the calibration whiteboard is placed on the opposite side of the detection window, and baseline calibration is performed before starting up every day. The wavelet threshold denoising technology is used to perform three-layer wavelet decomposition on the original redox current signal. Through the soft threshold function, the noise threshold is set to the square root logarithm of the noise standard deviation multiplied by the signal length to eliminate high-frequency noise. Based on the local outlier factor algorithm, the density ratio of the data point neighborhood is calculated, and the data deviating from the density distribution is marked as abnormal, and the abnormal data is interpolated and complemented with a sliding window. The near-infrared spectroscopy absorbance sequence is smoothed by a sliding window to eliminate the jitter noise of the near-infrared spectroscopy analyzer. The data sampling interval is dynamically adjusted according to the flow rate of the fermentation broth. When the flow rate is greater than 1.5 m / s, the inertial compensation algorithm is enabled, and the cell density value at the previous moment plus the product of the compensation coefficient and the flow rate change is used as the current cell density value;

[0071] The biometric recognition module performs biometric recognition on the preprocessed microbial fermentation strain data, constructs a strain feature library, and for the preprocessed metabolite concentration data, within a fixed time window respectively calculates the metabolic rates of lactic acid and ethanol r and V y , uses the dynamic time warping technique to align the metabolite concentration curves of different batches of strains, and calculates the cumulative distance of the minimum path between the curves , to obtain the growth curve similarity , and its calculation process is as follows:

[0072] ;

[0073] ;

[0074] Among them, is the metabolic rate, is the metabolite concentration at the current moment, and are the lengths of curve A and curve B respectively. If the growth curve similarity is greater than 85%, it is marked as a strain with stable growth. If the growth curve similarity is less than 85%, it is marked as a strain with abnormal metabolism. Perform principal component analysis on the near-infrared spectral absorbance sequence, and extract the first 3 principal components PC1, PC2, and PC3. Its calculation process is as follows:

[0075] ;

[0076] Among them, k = 1, 2, and 3, is the weight coefficient of the kth principal component at wavelength i, is the absorbance value corresponding to the wavelength, is the number of spectral wavelengths in the range of 900 - 1700 nm. Cluster according to the spatial distribution density of the principal components, divide the fermentation stage into the initial stage, logarithmic growth stage, and decline stage, and match the corresponding principal component threshold ranges for different growth stages. Encode the lactic acid and ethanol metabolic rates, growth curve similarity, PC1, PC2, and PC3 in the format of strain number - timestamp - feature vector, and integrate them into the strain feature library. When adding new strain data, calculate its Euclidean distance from the existing strains in the strain feature library. If the Euclidean distance is greater than 2, create a new feature entry. After retraining the historical strain data, recalculate every 7 days , and update the principal component eigenvalue in the feature library;

[0077] The prediction classification module is used to combine the strain feature library to construct a fitness prediction model and a microbial community metabolic network topology graph, predict the strain fitness, classify according to the strain yield, align the lactic acid and ethanol metabolic rates with the growth curve similarity at time steps, construct a 3D input vector to cover the time series data of the whole fermentation process from 0 to 72 hours, generate the strain historical fitness label based on the growth stability and product yield of the strain under specific culture conditions, adopt a bidirectional gated recurrent unit network architecture, set the dimensions of the forward hidden state and the backward hidden state in its hidden layer to be both 64, construct a fitness prediction model. The input layer of the fitness prediction model receives a three-dimensional feature vector, the hidden layer controls the information flow through the reset gate and the update gate, the output layer concatenates the bidirectional hidden states and maps them to the strain fitness D through a fully connected layer. The loss function uses the mean square error to calculate the difference between the predicted strain fitness and the strain historical fitness label, sets the classification label for the strain based on historical data. The strain classification labels include high-yield, medium-yield, low-yield, and metabolic abnormality. Using the graph convolutional network architecture, the first 3 principal components PC1, PC2, and PC3 of each strain are used as node feature vectors to input into the input layer of the graph convolutional network architecture, construct a microbial community metabolic network topology graph, construct an adjacency matrix based on the principal component spatial distance. If the Euclidean distance of the principal components of two strains is less than 1.5, it is defined that there is an edge between the nodes corresponding to the two strains. The hidden layer of the microbial community metabolic network topology graph uses two layers of graph convolutional operations, each layer outputs 16 dimensions, and realizes node feature propagation through the normalization of the adjacency matrix and the degree matrix with self-connections added. The output layer aggregates the node features through global max pooling to generate a 3D probability vector , set the upper limit of the strain yield probability threshold to 0.7 and the lower limit of the strain yield threshold to 0.5. If in the output 3D probability vector is greater than 0.7, then mark this strain as high-yield. If and is the highest among the three, then mark this strain as medium-yield. If and is the highest among the three, then mark this strain as low-yield. If the probability difference between two of them is less than 0.1, then determine the category according to the proximity of the principal component spatial distance to the historical high-yield cluster. If the three probabilities of the strain are all less than 0.5, then mark it as a metabolic abnormality candidate strain. If the Euclidean distance of the principal component features of the strain is greater than 3, then temporarily store it as an unclassified strain, and dynamically adjust the upper and lower limits of the strain yield probability threshold based on the proportion of high-yield strains in historical data;

[0078] The strain selection decision-making module combines the strain fitness and the strain classification results with weights to construct a strain selection decision-making model, outputs the optimal strain number and culture parameters, dynamically adjusts the weight coefficients of fitness and classification results according to the strain classification results, constructs a strain selection decision-making model. If the strain is classified as high-yield, the strain classification result is given priority. If the strain is classified as medium-yield and low-yield, the fitness is emphasized. Based on the weight coefficient of fitness and the weight coefficient of classification results, the classification gain Z and the comprehensive score F are calculated. The calculation process is as follows:

[0079] ;

[0080] ;

[0081] Among them, is the weight coefficient of fitness, is the weight coefficient of the strain classification result. The upper and lower limits of the threshold of the comprehensive score are set to 0.8 and 0.6 respectively. If the comprehensive score ≥ 0.8, the strain enters the preferred queue. If 0.6 ≤ comprehensive score < 0.8, the strain enters the queue to be observed. If the comprehensive score < 0.6, the strain is marked as a candidate for elimination. If the strain is marked as a candidate for abnormal metabolism, it is pushed to the early warning monitoring module, and a historical parameter mapping table recording the pH and feeding rate of the strain under the optimal culture conditions is introduced. The culture parameters of the strain with the same classification result and the highest fitness are matched from the historical parameter mapping table, and the PH and feeding rate are updated. The calculation process is as follows:

[0082] ;

[0083] ;

[0084] Among them, and represent the reference pH and reference rate, which are their historical optimal values respectively;

[0085] Optimization training module, which optimizes the hyperparameters of the fitness prediction model, the microbial community metabolic network topology map, and the strain breeding decision model. The fitness prediction model dynamically adjusts its learning rate through the cosine annealing strategy, sets the initial learning rate to 0.001, the learning rate threshold to 0.0001, and decays according to the cosine function as the number of training rounds increases. A early stopping mechanism is set. If the validation set loss of the fitness prediction model does not decrease for 10 consecutive rounds, the training is terminated to prevent overfitting. Add the principal component features of the new strain to the topology map and recalculate the adjacency matrix. If the maximum Euclidean distance between the new node and the existing nodes exceeds 2, expand the network structure. When the number of newly added data reaches 100 per batch, use the incremental learning algorithm to update the parameters of the graph convolutional network. Use the cross-entropy loss function to optimize the microbial community metabolic network topology map. When the mean square error between the actual yield of the preferred strains per batch and the comprehensive score predicted by the strain breeding decision model exceeds 0.1, update the weight coefficient of the fitness according to the gradient descent method. Add the actual culture parameters and yield data of the preferred strains to the historical parameter mapping table, and update the reference pH value and reference rate value every 7 days;

[0086] Early warning monitoring module, which sets the lower limit of the microbial community metabolic rate warning and the threshold for abnormal fluctuations in cell density, thus triggering the strain elimination warning. Based on the historical metabolic rate distribution of high-yield strains, select the 5th percentile as the lower limit of the microbial community metabolic rate warning. After adding high-yield strain data per batch, recalculate the lower limit of the microbial community metabolic rate warning according to the sliding window. If the deviation between the current value and the original threshold > 10%, update it as the microbial community metabolic rate threshold. Calculate the percentage of the standard deviation of the cell density within the time window to the mean value to obtain the coefficient of variation of the cell density. Set the thresholds for abnormal fluctuations in cell density to 25%, 20%, and 10% for the initial stage, logarithmic growth phase, and decline phase of fermentation respectively. If PC1 > 0.7 and PC2 < 0.3, it is determined that the microbial community is in the initial stage of fermentation. If PC2 > 0.5 and PC3 < 0.2, it is determined that the microbial community is in the logarithmic growth phase. If PC3 > 0.6 and PC1 < 0.4, it is determined that the microbial community is in the decline phase. When the coefficient of variation of the cell density exceeds the threshold for abnormal fluctuations in cell density at the corresponding fermentation stage, it is regarded as abnormal fluctuations. For strains in the initial stage of fermentation, if the initial metabolic rate is lower than the current lower limit of the microbial community metabolic rate warning and the coefficient of variation of the cell density exceeds 25%, mark them as growth-delayed strains and push them to the manual review queue. If the strain does not enter the logarithmic growth phase for more than 72 hours during the initial stage of fermentation, trigger the metabolic stagnation warning, terminate the fermentation and sample for analysis. Mark abnormal strains according to the lower limit of the microbial community metabolic rate warning and the threshold for abnormal fluctuations in cell density, and re-verify the deviation degree of the principal component space. If the strain is confirmed to be abnormal, terminate the culture and adjust the culture parameters, and synchronously update the strain feature library;

[0087] The sorting and regulation module visualizes the distribution of the main components of strains and the genetic characteristic report through a three-dimensional scatter plot, triggers the automatic sorter to perform strain optimization and synchronously regulate the culture parameters. Taking the PC1, PC2, and PC3 of the strains as the three-dimensional coordinate axes, a scatter plot is constructed. Each data point represents a strain. The high-yield strain clusters are identified through the density clustering algorithm, the vertex coordinates of the convex hull are calculated and the cluster boundaries are generated. Different strain classification labels are distinguished by different color coding, and the spatial distribution of the high-yield clusters is dynamically marked. Among them, red represents high-yield strains, yellow represents medium-yield strains, blue represents low-yield strains, and gray represents metabolically abnormal strains. The differences in the expression levels of metabolic pathway genes between the high-yield strain clusters and the whole library of strains are statistically analyzed, the hypergeometric test is used to calculate the metabolic pathway enrichment degree, and the significant pathways with a metabolic pathway enrichment degree lower than 0.05 are screened. The mean and standard deviation of the key enzyme activities are statistically analyzed, and the highly expressed enzymes with activities greater than twice the standard deviation of the mean are marked. A comprehensive report containing a three-dimensional scatter plot, a pathway enrichment table, and a bar chart of enzyme activity comparison is generated. The strains in the optimization queue are sorted in descending order according to the comprehensive score, and the top 10% of the strains with the highest scores are preferentially sorted. If there are multiple strains in the same score interval, they are sorted according to the proximity of the main component spatial distance to the historical high-yield cluster. The sorter locates the target strain using the microfluidic chip sorting technology according to the strain number. After sorting, the strain is automatically inoculated into a new fermenter, and the sorting timestamp and culture parameters are synchronously recorded. If the strain is classified as high-yield, the pH is normally adjusted. If the strain is in the logarithmic growth phase, an additional pH compensation amount of 0.05 is added. The basic feeding rate is matched according to the historical parameter mapping table. If it is detected that the key enzyme activity is greater than twice the standard deviation of the mean, the feeding rate is additionally increased by 10%. The culture of the metabolically abnormal candidate strains in the fermenter is terminated, the feeding pump is turned off and the in-tank cleaning program is started. The main component characteristics and operation logs of the abnormal strains are recorded, pushed to the strain characteristic library and marked as eliminated data. After each batch of sorting, according to the difference between the actual yield and the predicted score, the weight coefficient of the fitness is updated by the gradient descent method. Every 7 days, the invalid parameter entries with an actual yield lower than 20% of the mean of the same type are removed from the historical parameter mapping table.

[0088] First, through the bacterial data acquisition module, the system will automatically collect and preprocess the bacterial data during the microbial fermentation process to ensure the accuracy and availability of the data. Subsequently, the biometric recognition module will conduct in-depth analysis on the preprocessed data, identify the biological characteristics of the bacteria, and construct a comprehensive bacterial characteristic library. Then, the prediction and classification module will utilize the bacterial characteristic library to construct a fitness prediction model and a topological map of the microbial metabolic network to predict the fitness and yield of the strains, and classify the strains according to the yield. The strain selection decision-making module will comprehensively consider the fitness and classification results of the strains, and construct a strain selection decision-making model through a weighted combination method, and finally output the numbers of the optimal strains and their corresponding culture parameters. During use, the optimization training module will continuously optimize the hyperparameters of the fitness prediction model, the topological map of the microbial metabolic network, and the strain selection decision-making model to improve the prediction accuracy and selection efficiency of the system. At the same time, the early warning monitoring module will monitor the microbial metabolic rate and cell density in real time. Once the set early warning lower limit is exceeded or abnormal fluctuations occur, the system will immediately trigger a strain elimination warning to ensure the stability and safety of the selection process. Finally, the sorting and regulation module will visualize the principal component distribution and genetic characteristic report of the strains through a three-dimensional scatter plot to help users intuitively understand the genetic background and principal component distribution of the strains, thereby triggering the automatic sorter to execute strain optimization and synchronously regulating the culture parameters to achieve the automation and intelligence of strain selection.

[0089] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A microbial fermentation strain selection and breeding control system based on a neural network model, characterized in that: It includes bacterial data collection module, biological feature recognition module, prediction and classification module, breeding decision module, optimization training module, early warning monitoring module and sorting and regulation module; The bacterial cell data collection module collects and pre-processes the microbial fermentation strain data; The biometric identification module performs biometric identification on the pre-processed microbial fermentation strain data to construct a strain feature library; The prediction and classification module is used to combine the strain feature library, construct a fitness prediction model and a bacterial community metabolic network topology map, predict strain fitness, and classify according to strain yield; The breeding decision module performs a weighted combination of strain fitness and strain classification results, constructs a strain breeding decision model, and outputs the optimal strain number and culture parameters; The optimization training module optimizes the hyperparameters of the fitness prediction model, the bacterial community metabolic network topology map and the bacterial strain breeding decision model; The early warning monitoring module sets the lower limit of bacterial metabolic rate warning and the threshold of abnormal fluctuation of bacterial density, thereby triggering the strain elimination warning; The sorting and regulation module visualizes the main component distribution and genetic characteristic report of the strain through a three-dimensional scatter plot, triggers the automatic sorter to perform strain optimization and synchronously regulates the culture parameters.

2. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 1, characterized in that: In the bacterial cell data collection module, the collection and preprocessing process of microbial fermentation strain data includes: The microbial fermentation strain data is obtained by using a voltammetric biosensor probe and a near infrared spectrometer, wherein the microbial fermentation strain data includes metabolite concentration data and bacterial cell density data in the microbial fermentation liquid; The voltammetric biosensor probes are symmetrically arranged on the inner wall of the fermentation tank. A three-electrode system is used to apply step voltage through a constant potential instrument to collect redox current signals and detect the concentration data of metabolites in the microbial fermentation broth in real time. The metabolites include lactic acid and ethanol. A near-infrared spectrometer was embedded in the fermentation liquid circulation pipeline. The light source and the detector were installed at a 45° angle. The bacterial absorbance value was calculated based on the Lambert-Beer law, and the near-infrared spectrum absorbance sequence was constructed to invert the bacterial density. The wavelet threshold denoising technology was used to perform three-layer wavelet decomposition on the original redox current signal, and the near-infrared spectral absorbance sequence was smoothed by sliding window filtering. The data sampling interval was dynamically adjusted according to the fermentation liquid flow rate. When the flow rate was greater than 1.5 m / s, the inertia compensation algorithm was enabled, and the product of the cell density value at the previous moment plus the compensation coefficient and the flow rate change was taken as the current cell density value.

3. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 2, characterized in that: In the biometric identification module, the process of constructing a strain feature library includes: For the pre-processed metabolite concentration data, according to the fixed time window Calculate the metabolic rates of lactate and ethanol separately r and V y Dynamic time warping technology was used to align the metabolite concentration curves of different batches of strains and calculate the minimum path cumulative distance between the curves. , get the growth curve similarity , if the growth curve similarity is greater than 85%, it is marked as a growth-stable strain, and if the growth curve similarity is less than 85%, it is marked as a metabolically abnormal strain; The principal component analysis of the near-infrared spectral absorbance sequence was performed to extract the first three principal components PC1, PC2 and PC3. The fermentation stage was divided into the initial stage, logarithmic growth stage and decay stage according to the spatial distribution density of the principal components, and the corresponding principal component threshold range was matched for different growth stages. The lactate and ethanol metabolic rates, growth curve similarity, PC1, PC2 and PC3 are encoded in the format of strain number-timestamp-feature vector and integrated into the strain feature library. When new strain data is added, the Euclidean distance between it and the existing strains in the strain feature library is calculated. If the Euclidean distance is greater than 2, a new feature entry is created. After the historical strain data is retrained, it is recalculated every 7 days. , update the principal component eigenvalues ​​in the feature library.

4. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 3 is characterized in that: In the prediction and classification module, a fitness prediction model is constructed, and the process of predicting strain fitness includes: The lactate and ethanol metabolic rates and growth curve similarities were aligned by time step to construct a 3D input vector, covering the time series data of 0-72 hours during the fermentation process, and the strain historical fitness label was generated based on the growth stability and product yield of the strain under specific culture conditions; A bidirectional gated recurrent unit network architecture was adopted, and the dimensions of the forward hidden state and the backward hidden state in the hidden layer were set to 64. A fitness prediction model was constructed. The input layer of the fitness prediction model received a three-dimensional feature vector, and the hidden layer controlled the information flow through the reset gate and the update gate. The output layer concatenated the bidirectional hidden states and mapped them to the strain fitness D through the fully connected layer. The loss function used the mean square error to calculate the difference between the predicted strain fitness and the strain historical fitness label.

5. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 4, characterized in that: In the prediction and classification module, the process of constructing a bacterial community metabolic network topology map and classifying according to strain yield includes: Based on historical data, classification labels are set for strains, and the strain classification labels include high-yield, medium-yield, low-yield and metabolic abnormalities. The graph convolutional network architecture is used to input the first three principal components PC1, PC2 and PC3 of each strain as node feature vectors into the input layer of the graph convolutional network architecture to construct a bacterial community metabolic network topology map. An adjacency matrix is ​​constructed based on the principal component space distance. If the principal component Euclidean distance of two strains is less than 1.5, it is defined that there is an edge between the nodes corresponding to the two strains. The hidden layer of the bacterial community metabolic network topology map uses two layers of graph convolution operations, each layer outputs 16 dimensions, and node feature propagation is achieved by adding self-connected adjacency matrix and degree matrix normalization. The output layer aggregates node features through global maximum pooling to generate a 3D probability vector ; The upper and lower limits of the strain yield probability threshold are set to 0.7 and 0.5 respectively. If the output 3D probability vector If greater than 0.7, the strain is marked as high-yielding. and If the highest of the three, the strain is marked as medium-yielding. and If it is the highest among the three, the strain is marked as low-yielding; like If the probability difference between the two is less than 0.1, the category is determined according to the proximity of the principal component space distance to the historical high-yield cluster. If the three-category probability of the strain is less than 0.5, it is marked as a candidate strain for metabolic abnormality. If the Euclidean distance of the principal component characteristics of the strain is greater than 3, it is temporarily stored as an unclassified strain, and the upper and lower limits of the strain yield probability threshold are dynamically adjusted based on the proportion of high-yield strains in historical data.

6. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 5, characterized in that: In the breeding decision module, the process of constructing a strain breeding decision model and outputting the optimal strain number and culture parameters includes: According to the strain classification results, the weight coefficients of fitness and classification results are dynamically adjusted to build a strain selection decision model. If the strain is classified as high-yield, the strain classification results are given priority. If the strain is classified as medium-yield and low-yield, fitness is emphasized. Based on the weight coefficient of fitness and the weight coefficient of classification results, the classification gain Z and comprehensive score F are calculated. The upper and lower limits of the comprehensive score threshold are set to 0.8 and 0.6 respectively. If the comprehensive score is ≥0.8, the strain enters the preferred queue. If 0.6≤comprehensive score<0.8, the strain enters the observation queue. If the comprehensive score is <0.6, the strain is marked as an elimination candidate. If the strain is marked as a metabolic abnormality candidate strain, it is pushed to the early warning monitoring module. Introduce the recording of pH and feed rate of strains under optimal culture conditions The historical parameter mapping table is used to match the strain culture parameters with the same classification result and the highest fitness from the historical parameter mapping table, and the pH and feed rate are updated.

7. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 6, characterized in that: In the optimization training module, the process of optimizing the hyperparameters of the fitness prediction model, the bacterial community metabolic network topology map, and the bacterial strain selection decision model includes: The fitness prediction model dynamically adjusts its learning rate through the cosine annealing strategy. The initial learning rate is set to 0.001, the learning rate threshold is set to 0.0001, and it decays according to the cosine function as the number of training rounds increases. The early stopping mechanism is set. If the validation set loss of the fitness prediction model does not decrease for 10 consecutive rounds, the training is terminated to prevent overfitting. The principal component features of the new strain are added to the topological map, and the adjacency matrix is ​​recalculated. If the maximum Euclidean distance between the new node and the existing node exceeds 2, the network structure is expanded. When the new data in each batch reaches 100, the parameters of the graph convolutional network are updated using the incremental learning algorithm, and the cross entropy loss function is used to optimize the topological map of the bacterial community metabolic network. When the mean square error between the actual yield of each batch of preferred strains and the comprehensive score predicted by the strain selection decision model exceeds 0.1, the fitness weight coefficient is updated according to the gradient descent method, the actual culture parameters and yield data of the preferred strains are added to the historical parameter mapping table, and the benchmark pH value and benchmark rate value are updated every 7 days.

8. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 7, characterized in that: In the early warning monitoring module, the process of setting the early warning lower limit of bacterial metabolism rate and the threshold of abnormal fluctuation of bacterial density to trigger the strain elimination early warning includes: Based on the distribution of metabolic rates of historical high-yield strains, the 5th percentile was selected as the lower limit of bacterial metabolic rate warning. After each batch of high-yield strain data was added, the lower limit of bacterial metabolic rate warning was recalculated according to the sliding window. If the current value deviated from the original threshold by more than 10%, it was updated to the bacterial metabolic rate threshold. The percentage of the standard deviation of the cell density within the time window to the mean was calculated to obtain the cell density variation coefficient. The cell density abnormal fluctuation thresholds were set at 25%, 20% and 10% according to the initial fermentation stage, logarithmic growth stage and decay stage, respectively. If PC1>0.7 and PC2<0.3, the bacterial community was judged to be in the initial fermentation stage. If PC2>0.5 and PC3<0.2, the bacterial community was judged to be in the logarithmic growth stage. If PC3>0.6 and PC1<0.4, the bacterial community was judged to be in the decay stage. When the cell density variation coefficient exceeded the cell density abnormal fluctuation threshold of the corresponding fermentation stage, it was regarded as an abnormal fluctuation. For strains in the early stage of fermentation, if the initial metabolic rate is lower than the lower limit of the current bacterial community metabolic rate warning and the bacterial density variation coefficient exceeds 25%, it will be marked as a growth-delayed strain and pushed to the manual review queue. If the strain does not enter the logarithmic growth phase for more than 72 hours in the early stage of fermentation, a metabolic stagnation warning will be triggered, and the fermentation will be terminated and samples will be collected for analysis; Abnormal strains are marked according to the lower warning limit of bacterial metabolic rate and the threshold of abnormal fluctuation of bacterial density, and the deviation of the principal component space is verified for a second time. If the strain is confirmed to be abnormal, the culture is terminated and the culture parameters are adjusted, and the strain feature library is updated synchronously.

9. The microbial fermentation strain selection and breeding control system based on a neural network model according to claim 8, characterized in that: In the sorting and control module, the process of visualizing the main component distribution of strains and marking high-yield bacterial clusters through a three-dimensional scatter plot and generating a genetic characteristic comparison report includes: The PC1, PC2 and PC3 of the strains were used as three-dimensional coordinate axes to construct a scatter plot. Each data point represented a strain. The high-yield bacterial clusters were identified by the density clustering algorithm. The coordinates of the convex hull vertices were calculated and the cluster boundaries were generated. Different strain classification labels were distinguished by different color codes, and the spatial distribution of high-yield clusters was dynamically annotated. Red indicated high-yield strains, yellow indicated medium-yield strains, blue indicated low-yield strains, and gray indicated metabolically abnormal strains. The differences in metabolic pathway gene expression between the high-yield bacterial cluster and the strains in the entire library were statistically analyzed. The metabolic pathway enrichment was calculated using the hypergeometric test. Significant pathways with metabolic pathway enrichment lower than 0.05 were screened. The mean and standard deviation of key enzyme activities were statistically analyzed. Highly expressed enzymes with activities greater than 2 times the standard deviation of the mean were marked. A comprehensive report containing a three-dimensional scatter plot, a pathway enrichment table, and an enzyme activity comparison bar chart was generated.

10. A microbial fermentation strain selection and breeding control system based on a neural network model according to claim 9, characterized in that: In the sorting control module, the process of driving the automatic sorter and the fermentation parameter control system to perform strain optimization and pH and feed rate adjustment includes: The strains in the preferred queue are arranged in descending order according to the comprehensive scores, and the strains with the top 10% scores are selected first. If there are multiple strains in the same score interval, they are sorted according to the distance in the main component space and the proximity to the historical high-yield cluster. The sorter locates the target strain according to the strain number using microfluidic chip sorting technology. After sorting, the strain is automatically inoculated into a new fermentation tank, and the sorting timestamp and culture parameters are recorded simultaneously; If the strain is classified as high-yielding, the pH is adjusted normally. If the strain is in the logarithmic growth phase, an additional pH compensation of 0.05 is added, and the basic feeding rate is matched according to the historical parameter mapping table. If the key enzyme activity is detected to be greater than 2 times the standard deviation of the mean, the feeding rate is increased by an additional 10%; Terminate the cultivation of candidate strains with metabolic abnormalities in the fermentation tank, turn off the feed pump and start the tank cleaning program, record the main component characteristics and operation logs of the abnormal strains, push them to the strain feature library and mark them as eliminated data, and after each batch is sorted, update the fitness weight coefficient according to the gradient descent method based on the difference between the actual yield and the predicted score, and remove invalid parameter entries with actual yields lower than 20% of the same category average from the historical parameter mapping table every 7 days.

Images