Spinosad culture medium optimization method and device based on machine learning

Abstract

The invention discloses a spinosad culture medium optimization method and device based on machine learning, and belongs to the technical field of microbial fermentation. According to the method, deep learning and Bayesian optimization technologies are combined to optimize the ratio of the spinosad fermentation culture medium, firstly, the characteristics of the culture medium are learned through a deep neural network, and then model parameters are adjusted by using a Bayesian optimization algorithm, so that intelligent control and optimization of the ratio of the microbial culture medium are realized; the microorganism production efficiency and the product quality are improved. According to the method, technical support is provided for the fields of medicine production, food processing, environmental protection and the like through intelligent analysis and accurate prediction, and the method has remarkable market prospects and application value.

Description

Technical Field

[0001] This invention belongs to the field of microbial fermentation technology, specifically relating to a method and apparatus for optimizing spinosad culture medium based on machine learning. Background Technology

[0002] In microbial fermentation, the culture medium typically includes multiple components, such as carbon sources, nitrogen sources, and inorganic salts. These components not only provide the nutrients and energy needed for microbial growth but also significantly influence the metabolic processes and product formation of microorganisms. For example, glucose is a commonly used carbon source that is readily utilized by bacteria; while yeast extract provides a high-quality organic nitrogen source and growth factors. Therefore, optimizing the formulation of the fermentation medium is crucial in microbial culture and fermentation. A well-formulated culture medium can significantly improve the growth rate and product yield of microorganisms while reducing experimental time and resource waste.

[0003] Traditional methods for optimizing culture medium ratios are often inefficient and costly, frequently requiring multiple trials to find a suitable ratio. Traditional methods for optimizing culture medium ratios include the following types: 1) Traditional experimental design method Traditional optimization methods, such as Response Surface Methodology (RSM) and Orthogonal Array Design (OAD), are commonly used to optimize the composition of microbial culture media. These methods establish an experimental design matrix, select key factors, set appropriate experimental ranges, obtain response data through experiments, and then use mathematical models to predict the optimal culture medium ratio. The advantages of these traditional methods are their simplicity, intuitiveness, and strong operability, but they also have certain limitations. Because the experimental design of these methods often relies on human experience, in cases of high dimensionality and complex variables, some important factors or interactions may be overlooked, leading to less accurate optimization results.

[0004] 2) Optimization methods based on genetic algorithms Genetic Algorithms (GAs) are optimization methods that simulate the principles of natural selection and genetics, and have been applied in the optimization of microbial culture media in recent years. By simulating the process of natural selection, GAs optimize parameter combinations through operations such as selection, crossover, and mutation to find the optimal culture medium composition. While GAs have achieved some success in the optimization of microbial culture media, they also have some drawbacks. For example, GAs typically require a large number of iterations, resulting in slow optimization speeds, and they are prone to getting trapped in local optima. Furthermore, the performance of GAs depends heavily on the design of the fitness function, requiring experienced researchers to formulate a reasonable objective function.

[0005] 3) Machine learning-based optimization methods In recent years, machine learning algorithms have been widely used in the optimization of microbial culture media. Machine learning methods such as Support Vector Machines (SVM) and Decision Trees have been used to establish models of the relationship between culture medium formulations and microbial growth products. These methods can achieve efficient culture medium optimization by learning patterns from experimental data. For example, researchers use SVM to model microbial culture media and predict the optimal culture medium formulation based on historical data. While these methods improve optimization efficiency, they usually require a large amount of data to train the model and lack interpretability, resulting in a relatively weak understanding of the actual biological mechanisms during the optimization process.

[0006] In summary, existing multi-kill culture media suffer from problems such as long processing time, low efficiency, unstable optimization results of culture medium ratios, and difficulty in ensuring that the yield of each experiment is within a good range. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and apparatus for optimizing spinosad culture medium based on machine learning, so as to solve the problems of long time consumption and low efficiency of spinosad culture medium in the prior art, unstable optimization results of culture medium ratio, and difficulty in determining that the yield of each experiment is within a good range.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A machine learning-based method for optimizing spinosad culture medium includes: To obtain the components in the culture medium that affect spinosad production and the maximum content of each component; The components and their corresponding contents are determined using the pyDOE package, and the corresponding contents of the components are less than or equal to the highest contents. Based on the experimental results, a dataset of experimental combinations is obtained. Multiple DNN models are constructed, with the input of the test combination and the output of spinosad yield. The DNN models are trained using the training set in the dataset to obtain the features and patterns of the multiple DNN models. The DNN models are evaluated using the test set in the dataset to obtain the evaluation results of the DNN models. Clustering analysis is performed based on the features and patterns of DNN models to obtain several clusters, each of which consists of multiple DNN models of the same class. The DNN model with the best evaluation results is selected from each cluster, and the features of the selected DNN model are optimized using Bayesian methods to obtain the optimized formula for spinosad culture medium.

[0009] A further improvement of the present invention is that: Preferably, the experimental combination process for determining the components and their corresponding contents using the pyDOE package satisfies the minimum deviation sampling rule.

[0010] Preferably, the training set data of the multiple DNN models are different.

[0011] Preferably, the training set data size for each DNN model is 80% of the total data size of the dataset for the experimental combination.

[0012] Preferably, the clustering analysis is the K-means clustering method; the multiple DNN models of the same class have similar patterns and features.

[0013] Preferably, the evaluation result is determined by the coefficient of determination and the MSE value.

[0014] Preferably, the Bayesian optimization process involves inputting the selected DNN model features into the GpyOpt library, and obtaining the optimized formula for spinosad culture medium after iterative convergence.

[0015] Preferably, after inputting the DNN model features into the GpyOpt library, iterative optimization is performed using the EI harvest function and the MPI harvest function to obtain their respective candidate spinosad culture medium formulations; the optimal spinosad culture medium formulation is selected from the candidate optimized spinosad culture medium formulations as the optimized spinosad culture medium formulation.

[0016] Preferably, after obtaining the optimized formula of spinosad culture medium, the experiment is carried out again based on the optimized formula of spinosad culture medium, and the DNN model is repeatedly constructed to the Bayesian optimization process to obtain a further optimized formula of spinosad culture medium.

[0017] A spinosad culture medium optimization device based on machine learning, comprising: The acquisition module is used to obtain the components in the culture medium that affect spinosad production and the maximum content of each component. The experiment module is used to determine the components and experimental combinations of their corresponding contents using the pyDOE package, wherein the contents of the corresponding components are less than or equal to the maximum contents; and to obtain a dataset of experimental combinations based on the experimental results. The training module is used to construct multiple DNN models, where the input of each DNN model is a combination of experiments and the output is the yield of spinosad. The DNN models are trained using the training set in the dataset to obtain the features and patterns of the multiple DNN models. The DNN models are evaluated using the test set in the dataset to obtain the evaluation results of the DNN models. The clustering module is used to perform clustering analysis based on the features and patterns of DNN models to obtain several clusters, each of which consists of multiple DNN models of the same class. The optimization module is used to select the DNN model with the best evaluation results from each cluster, and then optimize the features of the selected DNN model through Bayesian optimization to obtain the optimized formula of spinosad culture medium.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention discloses a machine learning-based method for optimizing spinosad culture medium. The method involves obtaining the components affecting spinosad and their maximum content, then using the pyDOE package to obtain experimental combinations of components and their corresponding contents. After conducting the corresponding experiments, the results are obtained. A Deep Neural Network (DNN) model is trained using these results to obtain the features and patterns of the DNN model. Cluster analysis is performed based on pattern similarity to obtain clusters of the DNN model and corresponding evaluation results. Further, based on the evaluation results and clusters, Bayesian optimization techniques are used to obtain the optimized spinosad culture medium. This method combines deep neural networks and Bayesian optimization techniques, using a deep learning model to predict and optimize the ratio of microbial culture media, and further improving the accuracy and stability of the ratio through the Bayesian optimization algorithm. By combining microbial culture medium ratio optimization with artificial intelligence algorithms, this method applies deep neural network learning and Bayesian optimization techniques to the field of microbial culture medium ratios. By introducing deep learning technology, the content of spinosad produced by microorganisms in the culture medium can be analyzed and predicted more accurately, thereby achieving intelligent control and optimization of microbial culture conditions. The application of Bayesian optimization algorithms can help search for optimal solutions more efficiently during parameter tuning, improving the stability and efficiency of culture medium formulations. This method not only provides accurate predictions in high-dimensional, multivariate scenarios but also obtains optimal solutions with fewer trials, greatly improving optimization efficiency and accuracy, and thus increasing microbial production efficiency. The resulting optimized formulation can improve the growth rate of microorganisms and the yield and quality of metabolites, ultimately providing a more efficient and low-cost solution for biopharmaceutical production. Compared with traditional methods, this approach has the advantages of high efficiency, low cost, and stability, enabling more precise optimization of culture medium formulations.

[0019] Furthermore, this method also helps reduce production costs and conserve resources, and has broad market application prospects. This invention also has the following advantages: 1) High optimization efficiency: Utilizing deep learning algorithms and Bayesian optimization, the optimal culture medium ratio can be found more quickly compared to traditional methods, significantly improving optimization efficiency.

[0020] 2) Low cost: Through precise optimization strategies, production costs can be effectively reduced, thereby improving the economic efficiency of the entire fermentation and cultivation process.

[0021] 3) Stable results: The method of this invention focuses on optimizing the accuracy and stability of the culture medium ratio, which can ensure that the output of microorganisms remains consistent in multiple experiments.

[0022] 4) Wide applicability: The application of this invention is not limited to the field of microbiology research, but can also be extended to multiple fields such as medicine, food processing, and environmental protection. By optimizing the microbial culture medium ratio, the production efficiency and quality of microbial products can be improved, providing technical support for industries such as drug development and food production. This method is applicable to multiple fields such as medicine, food processing, and environmental protection, and has significant market application prospects.

[0023] 5) Resource conservation: Intelligent analysis and accurate prediction reduce the waste of experimental time and resources, and improve the economic benefits of experiments. Attached Figure Description

[0024] Figure 1 This is a flowchart of a spinosad culture medium optimization method based on machine learning according to the present invention; Figure 2 This is a diagram of a spinosad culture medium optimization device based on machine learning according to the present invention; Figure 3 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0025] Hereinafter, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of that feature.

[0026] The synchronization method provided in this application can be applied to terminal devices such as mobile phones, tablets, wearable devices, in-vehicle devices, augmented reality (AR) / virtual reality (VR) devices, laptops, ultra-mobile personal computers (UMPCs), netbooks, and personal digital assistants (PDAs). This application does not impose any restrictions on the specific type of terminal device.

[0027] It should be noted that the terms "first," "second," etc., used in the specification and drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] Optimizing fermentation media plays a crucial role in microbial culture and fermentation. A well-formulated culture medium can improve microbial growth rates and product yields, thereby increasing production efficiency. By optimizing the medium, a more suitable environment can be provided for microorganisms, promoting their metabolic activities and growth, maximizing their potential production capacity. Furthermore, optimizing the medium can improve product purity and stability, enhancing product quality and meeting industrial production requirements. Most importantly, optimizing the medium reduces experimental time and resource waste, increasing experimental efficiency and providing a more reliable foundation for the industrialization of microbial fermentation and product production. Existing methods for preparing microbial culture media are often based on experience and trial and error, resulting in low efficiency and high costs. Therefore, developing an intelligent and precise method for optimizing culture medium formulation has significant theoretical and practical implications.

[0029] Based on the above reasons, this invention discloses a method combining deep learning and Bayesian optimization for optimizing the ratio of spinosad fermentation medium, specifically including the following steps; S1, obtain the components in the culture medium that affect spinosad production and the maximum content of each component; S2, using the pyDOE package to determine the components and the experimental combinations of their corresponding contents, wherein the contents of the corresponding components are less than or equal to the highest contents; based on the experimental results, obtain the dataset of experimental combinations; S3. Construct multiple DNN models, where the input of each DNN model is the experimental combination and the output is the spinosad yield; train the DNN models using the training set in the dataset to obtain the features and patterns of the multiple DNN models; evaluate the DNN models using the test set in the dataset to obtain the evaluation results of the DNN models. S4. Cluster analysis is performed based on the features and patterns of the DNN model to obtain several clusters, each of which consists of multiple DNN models of the same class. S5. Select the DNN model with the best evaluation results from each cluster, and optimize the features of the selected DNN model through Bayesian optimization to obtain the optimized formula of spinosad culture medium.

[0030] In some embodiments of the present invention, in S1, the common components in the culture medium that affect spinosad and the maximum content of each component can be obtained by consulting relevant literature. It should be understood that the method in this step is also applicable to the selection of the components and contents of culture media for other bacterial strains.

[0031] In some embodiments of the present invention, in S2, the experimental design matrix scheme is generated using the pyDOE library; the experimental design matrix scheme designs variables as components and the content of each component, wherein the content cannot exceed the maximum content of each component; different experimental groups contain different components; pyDOE is an experimental design package in Python, specifically used to design and generate various experimental design schemes, including full factorial experimental design, fractional experimental design, Latin hypercube experimental design, etc.; pyDOE provides a series of functions and tools that enable users to easily generate various experimental design matrices for optimizing experimental design and data collection processes.

[0032] As a preferred option, depending on the actual needs, the minimum content of some components can be non-zero, as they exist as necessary components of the culture medium.

[0033] As a preferred approach, PyPOE, during experiment design, adheres to the minimum bias sampling rule. The minimum bias sampling rule is a statistical sampling technique that selects a sample from the population such that the sample's statistical characteristics (such as mean and variance) are as close as possible to the corresponding characteristics of the population. This method is suitable for reducing sampling error and improving the accuracy of estimations.

[0034] In some embodiments of the present invention, after S2, based on the designed experimental combination, multiple culture media are constructed to conduct spinosad culture experiments. The culture time is set to a fixed value to obtain the spinosad yield under different culture media, thereby obtaining multiple sets of experimental datasets.

[0035] Furthermore, multiple sets of experimental datasets are divided into training and test sets. Preferably, 80% of the experimental data is used as the training set and 20% of the data is used as the test set.

[0036] In some embodiments of the present invention, in S3, multiple deep neural network (DNN) models are automatically generated through a machine learning platform. Each model is trained using a training set. It should be noted that although each DNN model is trained using a training set, the training set data used by different DNN models is different. That is, 80% of the different experimental results selected from all the above data are used as the training set.

[0037] After different DNN models are trained, their goodness of fit can be evaluated using the coefficient of determination (R-squared coefficient, R² test) and mean squared error (MSE) values ​​on the test dataset. The value of R² ranges from 0 to 1. The closer R² is to 1, the better the model fits the data, while the closer it is to 0 or less than 0, the worse the model fits the data. The smaller the MSE value, the better the fit.

[0038] In some embodiments of the present invention, each independent DNN model in S4 can learn different features and patterns through training, enabling them to exhibit different sensitivity patterns when responding to input data. By clustering the sensitivity patterns of these models, models with similar response patterns and features can be grouped into the same cluster, thereby forming a holistic classification and understanding of model behavior and performance, and obtaining the corresponding clustering. The above classification process is implemented using the k-means clustering algorithm of unsupervised learning.

[0039] It should be understood that in this step, the number of DNN models in each cluster is variable and is determined based on the specific similarity.

[0040] In some embodiments of the present invention, in S5, optimization using the Bayesian algorithm is replaced by optimization using the GpyOpt library. The GpyOpt library is a Python library for Bayesian optimization, implementing the Bayesian optimization algorithm based on Gaussian processes. The GpyOpt library is primarily used to optimize functions with continuous parameters. By establishing a Gaussian process model of the parameters, it continuously updates the posterior distribution of the parameters to find the optimal solution. In exploring the composition and optimal ratio of culture media, since the composition and ratio of the culture media are usually continuous parameters, the Bayesian optimization algorithm can effectively explore the parameter space and find the optimal composition and ratio. By establishing a Gaussian process model, the GpyOpt library can gradually converge to the optimal ratio scheme through continuous trial and observation, thereby helping to optimize the culture media formulation. Therefore, using the GpyOpt library can improve the efficiency and accuracy of parameter optimization and is a suitable tool for exploring the composition and optimal ratio of culture media. Specifically, the DNN model with the best evaluation result is selected from each cluster formed above, imported into the GpyOpt library file for optimization, and the optimized formulation of spinosad culture medium is obtained through the final optimization result.

[0041] As a preferred approach, this invention selects two commonly used acquisition functions from the GPyOpt library, 'EI' and 'MPI', for optimization. Expected Improvement ('EI') finds the sampling point with the greatest potential improvement around the current best point. It considers the balance between predicted values ​​and confidence intervals, determining the next most promising improvement location by measuring the difference between the current best value and the predicted value. 'EI' focuses more on potential improvement space when selecting the next sampling point, tending to select points in unexplored areas, thus facilitating global search. Maximum Probability of Improvement (MPI), on the other hand, selects the point with the highest probability of improvement, i.e., the point with the highest probability of becoming a potential optimal solution. MPI focuses on selecting the point most likely to improve the objective function value, thus finding the optimal solution more quickly. After multiple iterations of the two optimization functions, multiple candidate culture medium components and ratios are obtained, from which several optimal components and ratios that maximize spinosad yield are selected.

[0042] In some embodiments of the present invention, after S5, the culture medium ratio combination that maximizes the predicted spinosad yield can be experimentally verified again. The results can be repeated from S2 to S5. Throughout the process, the DNN model is retrained, and the culture medium ratio is continuously adjusted and optimized through experimental verification until a satisfactory ratio result is obtained. This process experimentally verifies the optimized culture medium ratio, continuously adjusting and optimizing parameters to ensure the accuracy and stability of the culture medium ratio.

[0043] A second aspect of the present invention discloses a spinosad culture medium optimization device based on machine learning, comprising: The acquisition module is used to obtain the components in the culture medium that affect spinosad production and the maximum content of each component. The experiment module is used to determine the components and their corresponding contents in experimental combinations using the pyDOE package, wherein the contents of the components are less than or equal to the maximum contents; and to obtain a dataset of experimental combinations based on the experimental results.

[0044] The training module is used to construct multiple DNN models, where the input of each DNN model is a combination of experiments and the output is the yield of spinosad. The DNN models are trained using the training set in the dataset to obtain the features and patterns of the multiple DNN models. The DNN models are evaluated using the test set in the dataset to obtain the evaluation results of the DNN models. The clustering module is used to perform clustering analysis based on the features and patterns of DNN models to obtain several clusters, each of which consists of multiple DNN models of the same class. The optimization module is used to select the DNN model with the best evaluation results from each cluster, and then optimize the features of the selected DNN model through Bayesian optimization to obtain the optimized formula of spinosad culture medium.

[0045] This invention combines deep learning and Bayesian optimization techniques to optimize the formulation of spinosad fermentation medium. The method first uses a deep neural network to learn the characteristics of the culture medium and then employs a Bayesian optimization algorithm to adjust the model parameters, thereby achieving intelligent control and optimization of the microbial culture medium formulation to improve microbial production efficiency and product quality. Through intelligent analysis and accurate prediction, this method provides technical support for fields such as pharmaceutical production, food processing, and environmental protection, demonstrating significant market prospects and application value.

[0046] The following is a further explanation with reference to specific embodiments: Example S1, Obtain the components in the culture medium that affect spinosad and the maximum content of each component. A review of relevant literature revealed that for spinosad, commonly used components in the culture medium can be categorized into carbon sources, nitrogen sources, inorganic salts, and other reference components. Carbon sources include glucose (a commonly used carbon source, easily utilized by bacteria); corn syrup (rich in carbon, promoting the accumulation of fermentation products); and starch (a relatively complex carbohydrate, helping to prolong fermentation time). Nitrogen sources include cottonseed flour (a commonly used organic nitrogen source, providing abundant protein); yeast extract (a high-quality organic nitrogen source, also providing vitamins and other growth factors); and ammonium sulfate ((NH4)2SO4) (an inorganic nitrogen source, low in cost and with good stability). Inorganic salts include calcium carbonate (CaCO3), which regulates the pH of the culture medium and provides calcium ions; ferric sulfate (FeSO4), which provides iron, promoting cell growth and metabolism; dipotassium hydrogen phosphate (K2HPO4), which provides phosphorus to cells and buffers the pH of the culture medium; and magnesium sulfate (MgSO4), whose magnesium ions participate in various enzymatic reactions. Other culture medium components include sucrose, which serves as another carbon source and can stimulate antibiotic production; soybean meal, a source of plant protein that has been used in the fermentation medium for spinosad; fish meal, which provides a variety of amino acids and is often used as a supplementary nitrogen source in the fermentation industry; potato starch, a complex carbon source that can gradually release carbohydrates and extend production time; and maltose, a relatively mild sugar carbon source that is beneficial for controlling the fermentation rate.

[0047] In this embodiment, cottonseed meal, yeast extract, corn syrup, ammonium sulfate, glucose, calcium carbonate, ferric sulfate, potassium sulfate, magnesium sulfate, starch, and soybean meal were selected as the key components of the fermentation medium. These components are rich in nutrients and can effectively support the growth of microorganisms, providing them with essential nutrients, carbon sources, and minerals, thus optimizing the fermentation process and increasing the yield of bioproducts. Furthermore, the cost-effectiveness and availability of these components make them suitable for industrial applications, and related studies have confirmed their importance in fermentation media.

[0048] Furthermore, in this embodiment, the maximum concentrations of the 11 components selected are set as follows: The following ingredients were used in a spinosad yield experiment: cottonseed meal 60 g / L, yeast extract 20 g / L, corn syrup 45 g / L, ammonium sulfate (NH4)2SO4 2 g / L, glucose 100 g / L, calcium carbonate (CaCO3) 7 g / L, ferric sulfate (FeSO4) 1 g / L, potassium sulfate (K2HPO4) 1 g / L, magnesium sulfate (MgSO4) 4.8 g / L, starch 110 g / L, and soybean meal 30 g / L. The yield of spinosad was measured using the soil bacterium *Saccharopolyspora spinosa*, and the yield after 8 hours was used as the target variable in the subsequent training model.

[0049] S2. The experimental design method selects the Least Bias Sampling (LHS) parameter of the random design function of the pyDOE package to randomly generate 128 combinations that satisfy the LHS rule. The variables include the components and contents of the culture medium. Each experimental combination includes different culture medium components and their corresponding contents. The contents of each culture medium component cannot exceed the above maximum contents in order to ensure that the experimental data are sampled evenly and distributed reasonably, and finally generate the culture medium dataset.

[0050] Use 80% of the data in the dataset as the training set and 20% as the test set.

[0051] S3. Construct a deep neural network (DNN) model, and train and learn it by inputting the above culture medium feature data.

[0052] The model was built using the open-source machine learning platform TensorFlow (v2.6) with a script written in Python 3.7. Eleven selected culture medium components were used as input parameters to shape the model, and the training data was fitted by minimizing the mean squared error (MSE). Specific training parameters (epochs=1000, optimizer='adam', loss='mse') and more parameter settings are detailed in the model code documentation. A total of 50 independent DNN models were generated for subsequent analysis. During the training of these 50 independent DNN models, each model used 80% of the total data set, but the specific data was not identical. This dataset design ensured that each independent DNN model learned different features and patterns, resulting in different sensitivity patterns when responding to input data.

[0053] Based on the coefficient of determination (R² test) and MSE value of the test dataset (Note: The R² metric (R-squared coefficient) is typically used to evaluate the goodness of fit of a model; its value ranges from 0 to 1. The closer R² is to 1, the better the model fits the data; conversely, the closer it is to 0 or less than 0, the worse the model fits the data), in machine learning, Mean Squared Error (MSE) is commonly used as an evaluation metric for regression model performance. MSE is the average of the squared differences between predicted and true values. MSE values ​​are usually non-negative; the smaller the value, the smaller the difference between the model's prediction and the actual value. Therefore, when comparing different models, the model with a smaller MSE is generally considered more accurate; simply put, smaller is better, representing a closer prediction to the true value. Within each cluster, the representative model with the best accuracy is selected for further optimization of the matching parameters using Bayesian optimization.

[0054] S4, by clustering the sensitivity patterns of these models, models with similar response patterns and features can be grouped into the same cluster, thus forming a holistic classification and understanding of model behavior and performance. The k-means clustering algorithm for unsupervised learning was selected to cluster the 50 independent DNN models generated above. Using the scikit-learn program (a machine learning library), a total of 6 clusters were obtained. The results of obtaining the 50 DNN models and their corresponding mean squared error (MSE) values, root mean squared error (RMSE) values, coefficients of determination (R² test) for the test dataset, and coefficients of determination (R² test) for the training dataset, along with clustering information, are shown in the table below: Table 1. Model Accuracy Evaluation Table

[0055] Note: The bolded lines represent representative models from each selected cluster. These models will proceed to the next step of Bayesian optimization (excluding models in cluster 3 that do not have any good fits). The parameters of the representative deep neural network models will be adjusted and optimized to obtain the optimal culture medium ratio.

[0056] S5 uses the Bayesian optimization algorithm to optimize the parameters of the deep neural network model.

[0057] Five representative models (dnn_model6, dnn_model13, dnn_model20, dnn_model32, dnn_model42) were selected after model clustering and imported into the GpyOpt library. Bayesian optimization was performed on these representative models using two commonly used harvest functions: EI and MPI. Finally, Bayesian optimization (BO) was applied to these five representative models, using two common evaluation methods: BO-EI and BO-MPI. At least 50 iterations were performed, generating a total of 5*2*50=500 candidates. From these, 10 optimal culture medium ratios that maximize spinosad yield were selected. The maximum estimated spinosad yield range for all models is 1250.3~1658.2 mg / L. The optimal culture medium ratios are detailed in the Excel file. Compared to the previous highest of 784.72 mg / L... Based on the optimal culture medium ratio obtained using a Bayesian algorithm optimization model, the yield of mg / L spinosad is expected to increase by 1.38 to 2.11 times. The specific details are as follows: Table 2 Optimal Culture Medium Ratio

[0058] S6, Validation and Optimization: The culture medium ratio combination that maximizes the predicted yield of spinosad can be tested again. The results can be used to retrain the DNN model through the entire process. The culture medium ratio is continuously adjusted and optimized through experimental validation until a satisfactory ratio result is obtained.

[0059] The method of this invention has been experimentally verified and demonstrated good feasibility. The optimized culture medium ratio, verified experimentally, maximized the predicted yield of the fermentation medium, indicating that the model adjustment and optimization were successful. Furthermore, through experimental verification and continuous optimization, the accuracy and stability of the culture medium ratio were ensured, production costs were reduced, and production efficiency was improved. Throughout the process, the combination of a deep neural network model and Bayesian optimization effectively explored the composition and optimal ratio of the culture medium. The final optimized ratio is expected to increase the yield of spinosad by 1.38 to 2.11 times, thereby achieving a significant increase in yield and improved production efficiency.

[0060] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing spinosad culture medium based on machine learning, characterized in that, include: To obtain the components in the culture medium that affect spinosad production and the maximum content of each component; The components and their corresponding contents are determined using the pyDOE package, and the corresponding contents of the components are less than or equal to the highest contents. Based on the experimental results, a dataset of experimental combinations is obtained. Multiple DNN models are constructed, with the input of the test combination and the output of spinosad yield. The DNN models are trained using the training set in the dataset to obtain the features and patterns of the multiple DNN models. The DNN models are evaluated using the test set in the dataset to obtain the evaluation results of the DNN models. Clustering analysis is performed based on the features and patterns of DNN models to obtain several clusters, each of which consists of multiple DNN models of the same class. The DNN model with the best evaluation results is selected from each cluster, and the features of the selected DNN model are optimized using Bayesian methods to obtain the optimized formula for spinosad culture medium.

2. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, The experimental combination process for determining the components and their corresponding contents using the pyDOE package satisfies the minimum deviation sampling rule.

3. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, The training data for the various DNN models are different.

4. The method for optimizing spinosad culture medium based on machine learning according to claim 13, characterized in that, The training set data size for each of the DNN models is 80% of the total data size of the dataset for the experimental combination.

5. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, The clustering analysis is the K-means clustering method; the multiple DNN models of the same class have similar patterns and features.

6. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, The evaluation results are determined by the coefficient of determination and the MSE value.

7. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, The Bayesian optimization process involves inputting the selected DNN model features into the GpyOpt library, and obtaining the optimized formula for spinosad culture medium after iterative convergence.

8. The method for optimizing spinosad culture medium based on machine learning according to claim 7, characterized in that, After inputting the DNN model features into the GpyOpt library, iterative optimization was performed using the EI harvest function and the MPI harvest function to obtain their respective candidate spinosad culture medium formulations. The optimal spinosad culture medium formulation was selected from the candidate spinosad culture medium formulations as the optimized spinosad culture medium formulation.

9. The method for optimizing spinosad culture medium based on machine learning according to claim 1, characterized in that, After obtaining the optimized formula of spinosad culture medium, experiments were conducted again based on the optimized formula of spinosad culture medium. The DNN model was repeatedly constructed to the Bayesian optimization process to obtain a further optimized formula of spinosad culture medium.

10. A spinosad culture medium optimization device based on machine learning, characterized in that, include: The acquisition module is used to obtain the components in the culture medium that affect spinosad production and the maximum content of each component. The experiment module is used to determine the components and experimental combinations of their corresponding contents using the pyDOE package, wherein the contents of the corresponding components are less than or equal to the maximum contents; and to obtain a dataset of experimental combinations based on the experimental results. The training module is used to construct multiple DNN models, where the input of each DNN model is a combination of experiments and the output is the yield of spinosad. The DNN models are trained using the training set in the dataset to obtain the features and patterns of the multiple DNN models. The DNN models are evaluated using the test set in the dataset to obtain the evaluation results of the DNN models. The clustering module is used to perform clustering analysis based on the features and patterns of DNN models to obtain several clusters, each of which consists of multiple DNN models of the same class. The optimization module is used to select the DNN model with the best evaluation results from each cluster, and then optimize the features of the selected DNN model through Bayesian optimization to obtain the optimized formula of spinosad culture medium.

Images