A method and system for optimizing the enzymatic hydrolysis process of polygonatum sibiricum based on machine learning

Abstract

The application provides a machine learning-based optimization method and system for protease hydrolysis of rhizoma polygonati, and belongs to the technical field of enzyme hydrolysis process optimization. The method uses literature data and subsequent internal experimental data as the training basis, obtains the optimized process conditions through a regression model and an optimization algorithm, experimentally verifies the optimized process conditions, and iteratively updates the model through the experimental results, so that the model gradually converges to the rhizoma polygonati protein enzyme hydrolysis special process space. The application uses the internal data backflow retraining method of rhizoma polygonati protein enzyme hydrolysis, so that the model recommendation gradually approaches the real enzyme hydrolysis characteristics of rhizoma polygonati protein, and can be directly extended and applied to the actual industrial production scene of rhizoma polygonati protein enzyme hydrolysis.

Description

Technical Field

[0001] This invention relates to the field of enzymatic hydrolysis process optimization technology, specifically to a method and system for optimizing the enzymatic hydrolysis process of Polygonatum odoratum based on machine learning. Background Technology

[0002] The research and development goal of the proteolytic process of Polygonatum sibiricum is to obtain a higher degree of hydrolysis (DH), better bioactive peptide generation capacity, and a stable process window with fewer experiments. Current mainstream research and development processes mainly rely on empirical screening, single-factor experiments, and response surface methodology, which can optimize local parameter ranges. However, when multiple variables such as enzyme type, enzyme dosage, reaction time, temperature, pH, and substrate differences exist simultaneously, the number of parameter combinations increases rapidly, correspondingly increasing experimental costs and analytical burden.

[0003] Existing publicly available literature has accumulated a wealth of research results on enzymatic hydrolysis processes, covering the correlation between substrate type, enzyme species, enzyme concentration, reaction time, and hydrolysis efficiency. However, this data is scattered across figures and texts in different publications, and the statistical definitions of the fields are inconsistent, making it difficult for researchers to directly compile it into a model training set, and also making it impossible to build a transferable predictive model adapted to the case of protamine.

[0004] This invention uses the experimental results from the paper "Optimizing Mung Bean and Soybean Hydrolysis for the Generation of Bioactive Peptides of Potential Functional Food Applications" as the original training data. It reconstructs structured samples, establishes a predictive model, and outputs optimized recommendation conditions through machine learning methods, providing a feasible prior modeling framework and basic technical support for optimizing the hydrolysis process of Polygonatum odoratum. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, this invention provides a machine learning-based method and system for optimizing the hydrolysis process of Polygonatum odoratum.

[0006] To achieve the above objectives, the specific technical solution of the present invention is as follows:

[0007] In a first aspect, the present invention provides a method for optimizing the hydrolysis process of Polygonatum sibiricum based on machine learning, comprising the following steps:

[0008] Acquire experimental data, perform field mapping, caliber verification, and standardization preprocessing on the experimental data according to a unified field system, and construct a structured dataset that can be input into the same training framework;

[0009] A main process optimization model and a candidate peptide screening auxiliary model are trained using a dataset. The main process optimization model performs regression prediction on single-objective or multi-objective responses, and the candidate peptide screening auxiliary model prioritizes active peptide candidates.

[0010] Optimal process conditions are obtained based on surrogate search, genetic search, and / or gradient-based local refinement; wherein, the surrogate search is used for candidate condition recommendation in stages with a small sample size, the genetic search is used for global parameter combination search, and the gradient-based optimization is used for local refinement near the candidate conditions;

[0011] The preferred process conditions are experimentally verified, and the verification results are incorporated into the dataset for further training. The model parameters are updated with the goal of minimizing the weighted comprehensive objective function. When the changes in recommended process conditions, the improvement in the comprehensive objective function, and the fluctuation in cross-validation performance are all below a preset threshold in continuous iterations, the model is determined to have converged to the dedicated process space for the hydrolysis of Polygonatum sibiricum.

[0012] Furthermore, the experimental data includes at least input fields and output response fields. For the master process optimization model, the input fields include enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, raw material or pretreatment state, and data source. The data source is used to distinguish between literature samples, internal experimental samples, and validation reflux samples, and is used as sample weight, batch correction, or confidence marker. The output response fields include response indicators such as degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. This invention maps literature samples and subsequent internal experimental samples of Polygonatum sibiricum hydrolysis into structured samples of "raw material / pretreatment state-enzyme type-temperature-pH-time-substrate concentration-enzyme-to-substrate ratio-data source-response indicator," forming a training data system suitable for Polygonatum sibiricum hydrolysis scenarios. For the candidate peptide screening auxiliary model, the input fields include characteristics such as molecular weight, hydrophobicity, net charge, secondary structure ratio, and amino acid composition of the hydrolysis product, and the output response fields are the activity candidate category or candidate priority.

[0013] Furthermore, the regression prediction of the target response includes regression prediction of a single target response or regression prediction of a multi-target response; wherein the single target response includes the degree of hydrolysis; and the multi-target response includes at least two of the following indicators: degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution.

[0014] This invention establishes a machine learning optimization method for the enzymatic hydrolysis process of Polygonatum sibiricum. This method uses literature data and subsequent internal experimental data as training foundations. Taking the master process optimization model as an example, the master process optimization model takes process conditions such as enzyme type, temperature, pH, time, substrate concentration, and enzyme-to-substrate ratio as inputs, and outputs indicators such as degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. The optimal process conditions are obtained through a regression model and optimization algorithm, and the model is iteratively updated through experimental result feedback. To ensure the computability and repeatability of the scheme, each enzymatic hydrolysis experiment is preferably represented as a sample pair consisting of an input condition vector and an experimental response vector, as shown below:

[0015]

[0016]

[0017] In the above input condition vector In the diagram, each input quantity corresponds to a specific enzyme type. Reaction temperature pH of the reaction Reaction time substrate concentration Enzyme base ratio Substrate or pretreatment state and data sources In the above response vector In the figure, each output quantity corresponds to the degree of hydrolysis in turn. Peptide yield Bioactivity indicators such as antioxidants and related indicators of molecular weight distribution .

[0018] Logical input variables Standardization process:

[0019]

[0020] in, This represents the original value of the i-th sample in the j-th numeric field; This indicates the mean of the field; This indicates the standard deviation of the field; This represents the standardized field value;

[0021] One-hot encoding is performed on categorical input variables to obtain one-hot encoding vectors. ), onehot ) and onehot ( ), and concatenate it with numerical standardized variables to form the model input vector. .

[0022] Furthermore, the main process optimization model can adopt an elastic network regression model, and use partial least squares regression model, support vector regression model, linear regression model or response surface quadratic regression model as a comparison model; the candidate peptide screening auxiliary model can adopt a support vector machine model, and use partial least squares discriminant model, logistic regression model or random forest model as a comparison model.

[0023] Furthermore, the main process optimization model is expressed as:

[0024]

[0025] in, Indicates by parameters A defined regression model; This represents the response vector predicted by the model.

[0026] Furthermore, the weighted synthesis objective function is expressed as:

[0027]

[0028] Among them, wDH, wyield, wAOX, and wMW are the target weights. For normalization function, This represents the target molecular weight distribution range.

[0029] Furthermore, the preferred process conditions are expressed as follows:

[0030]

[0031] X satisfies process constraints such as temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, and equipment operating range.

[0032] Furthermore, the acquisition function for the proxy search is expressed as:

[0033]

[0034] in, Predict the mean for the surrogate model; To predict uncertainty; To explore weights.

[0035] Furthermore, the population update of the genetic search is represented as:

[0036]

[0037] Where P(g) represents the candidate process group of the g-th generation; Mutate, Crossover, and Select represent mutation, crossover, and selection operations, respectively.

[0038] Furthermore, the gradient-based local refinement is expressed as:

[0039]

[0040] in, Step size; The gradient of the objective function; Indicates projection onto feasible process space .

[0041] Furthermore, the iterative update of the model can be represented as:

[0042]

[0043] Where D(t) is the training sample set for the t-th round; D(t) represents the newly added Polygonatum sibiricum experimental verification sample; These are the updated model parameters.

[0044] This invention employs standardized processing formulas to uniformly process input variables of different dimensions, and uses one-hot encoding to convert enzyme species, pretreatment states, and data sources into model-recognizable features. A prediction formula establishes a mapping relationship between process conditions and target responses. A weighted comprehensive objective function balances high hydrolysis degree, high peptide yield, high activity, and target molecular weight distribution, where each weight coefficient characterizes the importance of different optimization objectives. For the application scenario of protamine, the weight coefficients can be adjusted according to the research and development objectives; for example, increasing the weight corresponding to activity when activity is the priority objective, and increasing the weight corresponding to yield when yield is the priority objective. A normalization function is used to eliminate dimensional differences between different output indicators. The search and update formulas correspond to surrogate search, genetic search, gradient-based local refinement, and experimental backflow retraining, respectively. For single-objective scenarios, only the hydrolysis degree regression model can be retained; for multi-objective scenarios, multiple output sub-models can be trained simultaneously.

[0045] The optimal process condition solution formula is used to determine the preferred parameter combination within the feasible process space; the search formula based on the acquisition function corresponds to a Bayesian search strategy driven by a surrogate model; the population update formula corresponds to a genetic global optimization strategy; and the local update formula corresponds to a gradient-based local refinement strategy. These three optimization methods can be selected based on the data stage, or combined in the order of "surrogate search to generate candidate conditions, genetic search to expand global candidates, and gradient-based optimization for local refinement." When new experimental results for Polygonatum sibiricum are obtained, the incremental samples are incorporated into the existing dataset, and the main process optimization model and the candidate peptide screening auxiliary model are retrained to obtain updated model parameters, thus forming a closed-loop iterative mechanism jointly driven by literature data and internal experimental data.

[0046] Furthermore, the multi-objective optimization module calculates predicted values ​​for degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution for candidate process conditions under given constraints of temperature, pH, reaction time, substrate concentration, and enzyme-to-substrate ratio. Subsequently, each predicted value is normalized and a comprehensive score is formed according to preset weights. The candidate process conditions are then ranked according to the comprehensive scores, thereby outputting the optimal process conditions that balance high activity and high yield within the same constraint set.

[0047] Secondly, the present invention provides a machine learning-based optimization system for the hydrolysis process of Polygonatum sibiricum, comprising:

[0048] The data acquisition module is used to collect the input conditions and output responses of the Polygonatum odoratum protein hydrolysis experiment. The data acquisition module includes a data access unit, a field mapping unit, a calibration unit, and a data storage unit. The data access unit is responsible for accessing literature samples, internal experimental samples, and validation reflux samples. The field mapping unit is responsible for mapping raw fields to raw material or pretreatment state, enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, data source, and response index. The calibration unit is responsible for standardizing temperature units, time units, concentration units, and response index names. The data storage unit is responsible for constructing a trainable sample set.

[0049] Field standardization module: Used to uniformly process categorical features, numerical features, missing values, and dimensional differences. Its function is to eliminate the differences in the scope between experiments from different sources, so that different samples can be input into the same regression model or optimization model.

[0050] The machine learning modeling module is used to establish the mapping relationship between process conditions and target responses. This module includes a feature encoding unit, a regression modeling unit, a classification modeling unit, and a model evaluation unit. The feature encoding unit is responsible for one-hot encoding, numerical standardization, and missing value imputation. The regression modeling unit is responsible for predicting the degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. The classification modeling unit takes peptide molecular weight, hydrophobicity, net charge, secondary structure ratio, and optional amino acid composition features as input and outputs activity candidate categories or candidate priorities. The model evaluation unit is responsible for cross-validation, independent test set validation, accuracy, F1 score, and error index calculation.

[0051] The optimization module is used to search for optimal parameter combinations within the feasible process space. The optimization module includes a surrogate search unit, a global search unit, and a local refinement unit. The surrogate search unit corresponds to Bayesian search, used to balance prediction mean and uncertainty in small sample stages. The global search unit corresponds to genetic search, used to search for global candidates within the combination space of enzyme species, temperature, pH, time, substrate concentration, and enzyme-to-substrate ratio. The local refinement unit corresponds to gradient optimization, used to refine the neighborhood of high-scoring candidate conditions.

[0052] Verification and Iteration Module: This module receives new experimental results from Polygonatum odoratum and completes closed-loop updates. Its function is to feed the latest experimental results back to the sample library and retrain the main process optimization model and the candidate peptide screening auxiliary model to obtain updated model parameters, so that the model gradually converges to the Polygonatum odoratum-specific process space.

[0053] The core of this invention lies in: converting enzymatic hydrolysis process samples from different sources into structured vectors that can be processed by machine learning through a unified field system that includes data sources, raw material or pretreatment status, and response indicators; establishing a mapping relationship between process parameters and multi-objective responses through a master process optimization model; associating the optimization results of process conditions with active peptide candidate verification objects through a candidate peptide screening auxiliary model; outputting optimal parameters within the feasible process space through a phased search mechanism of surrogate search, genetic search, and gradient local refinement; and gradually approximating the actual enzymatic hydrolysis characteristics of Polygonatum protein through internal data reflux and retraining. The candidate peptide screening auxiliary model also employs an experimental backflow retraining mechanism for iterative updates: in the same batch where the optimized process conditions are experimentally verified, the molecular weight, hydrophobicity, net charge, secondary structure ratio, and amino acid composition of the candidate peptides identified in the enzymatic hydrolysis products are calculated, and their activity is measured to determine the active candidate category or candidate priority, forming new labeled candidate peptide samples; the newly added candidate peptide samples are then incorporated into the candidate peptide dataset, and the auxiliary model is retrained; when the change in candidate peptide priority ranking and the fluctuation of classification performance indicators (accuracy, F1 score) are both below a preset threshold in continuous iterations, the auxiliary model is considered to have converged. Thus, the same batch of Polygonatum odoratum enzymatic hydrolysis verification experiments simultaneously provides new training samples for both the main process optimization model and the candidate peptide screening auxiliary model, enabling the two models to be updated collaboratively in closed-loop iteration.

[0054] Thirdly, this invention provides a general-purpose computer device, including: a processor, a memory, and an input / output interface. The memory stores a program; when the program is processed by the processor, it executes the machine learning-based optimization method for the hydrolysis process of Polygonatum sibiricum. Specifically, the program includes: data acquisition, standardization, modeling, optimization, and iterative update procedures; the memory is also used to store sample sets, model parameters, and optimization results; the input / output interface is used to receive experimental data and output recommended process conditions. The above hardware resources and modules form a one-to-one execution relationship, thereby ensuring that this technical solution can run stably on the computing device.

[0055] Fourthly, the present invention provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the aforementioned machine learning-based optimization method for the hydrolysis process of Polygonatum odoratum.

[0056] Compared with the prior art, the advantages of the present invention are:

[0057] 1. The machine learning optimization method for the enzymatic hydrolysis process of Polygonatum sibiricum provided by this invention uses literature data and subsequent internal experimental data as training foundations, takes process conditions as input, and outputs indicators such as degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. Optimal process conditions are obtained through a master process optimization model and optimization algorithm. Unlike model construction methods that only perform conventional reflux retraining, this invention sets up a unified field system at the data layer, including "raw material / pretreatment state - enzyme type - temperature - pH - time - substrate concentration - enzyme-to-substrate ratio - data source - response index". At the model layer, it sets up a master process optimization model and a candidate peptide screening auxiliary model. At the optimization layer, it sets up a phased search mechanism of surrogate search, genetic search, and gradient-based local refinement. At the validation layer, it sets convergence criteria based on the recommended process variation, the improvement of the comprehensive objective function, and the performance fluctuation of cross-validation. This combination enables the enzymatic hydrolysis samples of Polygonatum sibiricum to be transferred from literature data to internal experimental data, and allows the model recommendations to gradually approach the actual enzymatic hydrolysis characteristics of Polygonatum sibiricum protein.

[0058] 2. The method of this invention sets up two types of interconnected models: the main process optimization model takes enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, raw material or pretreatment state, and data source as inputs, and outputs process responses such as degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution; the candidate peptide screening auxiliary model takes peptide molecular weight, hydrophobicity, net charge, secondary structure ratio, and optional amino acid composition characteristics as inputs, and outputs the active candidate category or candidate priority. This dual-model structure can couple the two judgment links of "whether the process conditions are optimal" and "whether the product peptide has the target activity", improving the pertinence and effectiveness of subsequent experimental verification. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of a machine learning-based method for optimizing the hydrolysis process of Polygonatum odoratum according to the present invention.

[0060] Figure 2 The results of different master process optimization models in Example 1 on five-fold cross-validation and independent test sets;

[0061] Figure 3 The results show the comparison of candidate peptide screening model data in Example 3;

[0062] Figure 4 Output results for the process optimization module. Detailed Implementation

[0063] To enable those skilled in the art to clearly and completely understand the technical solution of the present invention, the present invention will be further described in detail below with reference to embodiments. Obviously, the embodiments described herein are only for explaining the present invention and are not intended to limit the scope of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0064] This invention provides a machine learning-based method for optimizing the hydrolysis process of Polygonatum sibiricum, such as... Figure 1 As shown, it includes the following steps:

[0065] Acquire experimental data, perform field mapping, caliber verification, and standardization preprocessing on the experimental data according to a unified field system, and construct a structured dataset that can be input into the same training framework;

[0066] A main process optimization model and a candidate peptide screening auxiliary model are trained using a dataset. The main process optimization model performs regression prediction on single-objective or multi-objective responses, and the candidate peptide screening auxiliary model prioritizes active peptide candidates.

[0067] Optimal process conditions are obtained based on surrogate search, genetic search, and / or gradient-based local refinement; wherein, the surrogate search is used for candidate condition recommendation in stages with a small sample size, the genetic search is used for global parameter combination search, and the gradient-based optimization is used for local refinement near the candidate conditions;

[0068] The preferred process conditions are experimentally verified, and the verification results are incorporated into the dataset for further training. The model parameters are updated with the goal of minimizing the weighted comprehensive objective function. When the changes in recommended process conditions, the improvement in the comprehensive objective function, and the fluctuation in cross-validation performance are all below a preset threshold in continuous iterations, the model is determined to have converged to the dedicated process space for the hydrolysis of Polygonatum sibiricum.

[0069] In one embodiment, the experimental data includes at least input fields and output response fields. For the master process optimization model, the input fields include enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, raw material or pretreatment state, and data source. The data source is used to distinguish between literature samples, internal experimental samples, and validation reflux samples, and is used as sample weight, batch correction, or confidence level markers. The output response fields include response indicators such as degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. For the candidate peptide screening auxiliary model, the input fields include characteristics such as molecular weight, hydrophobicity, net charge, secondary structure ratio, and amino acid composition of the enzymatic hydrolysis products, and the output response fields are the activity candidate category or candidate priority.

[0070] In one embodiment, the regression prediction of the target response includes regression prediction of a single target response or regression prediction of a multi-target response; the single target response includes the degree of hydrolysis; the multi-target response includes at least two of the following indicators: degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution; the multi-target optimization module calculates predicted values ​​of the degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution for candidate process conditions under given constraints of temperature, pH, reaction time, substrate concentration, and enzyme-to-substrate ratio; then, each predicted value is normalized and a comprehensive score is formed according to a preset weight; the candidate process conditions are ranked according to the comprehensive score, thereby outputting the optimal process conditions that balance high activity and high yield within the same constraint set.

[0071] In the following specific embodiments, this invention organizes experimental data from the literature to construct a structured training dataset, and then expands the data to form a dual-module data system comprising process optimization and candidate peptide screening. The resulting dataset contains 50 process optimization samples and 24 candidate peptide screening samples, comprehensively covering both process condition parameters and the physicochemical characteristics of peptides. Different machine learning models are trained using this dataset, and their performance is evaluated through cross-validation and independent test sets. The results show that the ElasticNet regression model used in this invention achieves an R-value in five-fold cross-validation. 2 A score of 0.6635 was achieved, with an R score on the independent test set. 2 A score of 0.9542 indicates that this performance is superior to the partial least squares regression (PLS) model in the literature (cross-validation R). 2 0.6454, independent test set R 2 The cross-validation R-value was 0.9454, which is also significantly better than the response surface methodology (RSM) quadratic regression model (cross-validation R-value). 2 0.4152, independent test set R 2 (R = 0.3884). Five-fold cross-validation R of the Advanced SVR model. 2The accuracy was 0.5916, lower than the 0.6635 of the elastic network model; therefore, the elastic network regression model was chosen in the final deployment stage to obtain better overall performance. Meanwhile, in the candidate peptide screening auxiliary task, the support vector machine (SVC) model proposed in this invention, after four-fold hierarchical cross-validation, achieved an accuracy of 0.7500 and an F1 score of 0.7685, higher than the accuracy of 0.7083 and F1 score of 0.7586 of the PLS model in the literature. These results indicate that the method provided by this invention can simultaneously improve the accuracy of process condition prediction and the effectiveness of candidate bioactive peptide screening.

[0072] Example 1: Training of a process optimization model based on DH data from the original paper

[0073] The DH bar chart in Fig. 3 of the original paper "Optimizing Mung Bean and Soybean Hydrolysis for the Generation of Bioactive Peptides of Potential Functional Food Applications" was selected as the original data source. Fifty samples were obtained through analysis, covering two substrates, four enzymes, two enzyme concentration levels, three reaction times, and a control group. The results of different models on the five-fold cross-validation and independent test set are shown in Table 1 and [Table data would be inserted here]. Figure 2 As shown.

[0074] Table 1: Results of different models on five-fold cross-validation and independent test sets.

[0075]

[0076] The results in Table 1 show that the elastic network regression model used in this invention achieves the best results on both five-fold cross-validation and the independent test set, where CVR... 2 The value is 0.6635, Test R 2 The value is 0.9542, which is significantly better than the PLS and RSM models in the literature.

[0077] Example 2: Recommendation Condition Output

[0078] The discrete process combinations from the original paper were input into the final model and the conditions were sorted. The top-ranked process conditions are shown in Table 2.

[0079] Table 2: Top-ranked process conditions in model output

[0080]

[0081] Table 2 shows that the top-ranked recommended conditions are Mung bean, Bromelain, 10.0%, and 12 h, with a predicted degree of hydrolysis of 39.632 and an observed degree of hydrolysis of 46.475. This recommended region is consistent with the high-performance process window in the original paper. These results demonstrate that this invention has completed a full closed loop from data extraction from the original paper, construction of training / test sets, model training, baseline comparison, condition recommendation, and webpage visualization, providing a directly reusable algorithmic framework for subsequent internal validation of the Polygonatum odoratum hydrolysis process.

[0082] Example 3: Candidate Peptide Screening Auxiliary Module

[0083] Twenty-four synthetic peptides from the study "Valorization of Turkey Cruor Through Peptic Hydrolysis: Peptidomic and Machine Learning Approaches for the Identification of Antimicrobial Peptides" were selected as an augmented dataset. The fields included molecular weight, proportion of hydrophobic residues, net charge, proportion of random coils, proportion of α-helices, and proportion of extended strands. The peptides were labeled with whether they exhibited antifungal or antiyeast activity.

[0084] Under four-fold hierarchical cross-validation, the proposed SVC model achieved an accuracy of 0.7500 and an F1 score of 0.7685, which is higher than the accuracy of 0.7083 and F1 score of 0.7586 of the PLS model in the literature (see...). Figure 3 The F1 score is the harmonic mean of precision and recall in the classification model, used to comprehensively evaluate the performance of candidate active peptide recognition when the number of positive and negative samples is limited or the class distribution is unbalanced. These results indicate that this auxiliary module can serve as a candidate screener for active peptides after the hydrolysis of Polygonatum odoratum, and can work in conjunction with the main process optimization model.

[0085] Example 4: Process Optimization Module

[0086] After the main model was trained, three optimization strategies were applied to the discrete process mesh. Bayesian-style surrogate search, Genetic-style global search, and Gradient-style local refinement all targeted regions with high performance conditions, demonstrating that this technique can converge to near-optimal process combinations from different optimization perspectives. The results are shown in Table 3. Figure 4 As shown.

[0087] Table 3: Output Results of Process Optimization Module

[0088]

[0089] The implementation of the above embodiments of the present invention is based on programmed processing by a device with processor functionality. Therefore, in engineering practice, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, on the basis of the above embodiments, the embodiments of the present invention provide a machine learning-based optimization system for the hydrolysis process of Polygonatum sibiricum, including: a data acquisition module: used to collect the input conditions and output responses of the hydrolysis experiment of Polygonatum sibiricum; the data acquisition module includes a data access unit, a field mapping unit, a caliber verification unit, and a data storage unit; wherein, the data access unit is responsible for accessing literature samples, internal experimental samples, and validation reflux samples; the field mapping unit is responsible for mapping the original fields to raw material or pretreatment state, enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, data source, and response index; the caliber verification unit is responsible for unifying the temperature unit, time unit, concentration unit, and response index name; the data storage unit is responsible for constructing a trainable sample set. A field standardization module: used to uniformly process categorical features, numerical features, missing values, and dimensional differences, its function is to eliminate caliber differences between experiments from different sources, so that different samples can be input into the same regression model or optimization model. Machine Learning Modeling Module: This module establishes the mapping relationship between process conditions and target responses. It includes a feature encoding unit, a regression modeling unit, a classification modeling unit, and a model evaluation unit. The feature encoding unit is responsible for one-hot encoding, numerical standardization, and missing value imputation. The regression modeling unit predicts the degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. The classification modeling unit takes peptide molecular weight, hydrophobicity, net charge, secondary structure ratio, and optional amino acid composition features as input and outputs activity candidate categories or candidate priorities. The model evaluation unit is responsible for cross-validation, independent test set validation, accuracy, F1 score, and error index calculation. Optimization Module: This module searches for optimal parameter combinations within the feasible process space. It includes a surrogate search unit, a global search unit, and a local refinement unit. The surrogate search unit corresponds to Bayesian search; the global search unit corresponds to genetic search; and the local refinement unit corresponds to gradient optimization. The verification and iteration module is used to receive new experimental results of Polygonatum odoratum and complete closed-loop updates. Its function is to feed the latest experimental results back to the sample library and retrain the main process optimization model and the candidate peptide screening auxiliary model to obtain the updated model parameters. When the changes in recommended process conditions, the improvement of the comprehensive objective function and the fluctuation of cross-validation performance are all lower than the preset threshold in continuous iterations, it is determined that the model gradually converges to the Polygonatum odoratum-specific process space.

[0090] The method described in this invention is implemented using electronic devices; therefore, it is necessary to introduce the relevant electronic devices. To this end, this invention provides a general-purpose computer device, including: a processor, a memory, and an input / output interface. The memory stores a program; when the program is processed by the processor, it executes the machine learning-based optimization method for the hydrolysis process of Polygonatum sibiricum. Specifically, the program includes: data acquisition, standardization, modeling, optimization, and iterative update procedures; the memory is also used to store sample sets, model parameters, and optimization results; the input / output interface is used to receive experimental data and output recommended process conditions. The above hardware resources and modules form a one-to-one execution relationship, thereby ensuring that this technical solution can run stably on a computing device. This invention also provides a non-transitory computer-readable storage medium storing computer instructions, which cause the computer to execute the machine learning-based optimization method for the hydrolysis process of Polygonatum sibiricum.

[0091] The above detailed embodiments describe the implementation of the present invention; however, the present invention is not limited to the specific details described in the above embodiments. Within the scope of the claims and technical concept of the present invention, various simple modifications and changes can be made to the technical solution of the present invention, and these simple modifications all fall within the protection scope of the present invention.

Claims

1. A method for optimizing the hydrolysis process of Polygonatum odoratum based on machine learning, characterized in that, Includes the following steps: Acquire experimental data, perform field mapping, caliber verification, and standardization preprocessing on the experimental data according to a unified field system, and construct a structured dataset that can be input into the same training framework; A main process optimization model and a candidate peptide screening auxiliary model are trained using a dataset. The main process optimization model performs regression prediction on single-objective or multi-objective responses, and the candidate peptide screening auxiliary model prioritizes active peptide candidates. Optimal process conditions are obtained based on surrogate search, genetic search, and / or gradient local refinement. The preferred process conditions were experimentally verified, and the verification results were incorporated into the dataset for further training. The model parameters were updated with the goal of minimizing the weighted comprehensive objective function. When the changes in recommended process conditions, the improvement in the comprehensive objective function, and the fluctuation in cross-validation performance are all below the preset threshold during continuous iterations, the model is determined to have converged to the dedicated process space for the hydrolysis of Polygonatum sibiricum.

2. The method for optimizing the enzymatic hydrolysis process of Polygonatum odoratum based on machine learning according to claim 1, characterized in that, The experimental data includes at least input fields and output response fields. For the master process optimization model, the input fields include enzyme type, temperature, pH, time, substrate concentration, enzyme-to-substrate ratio, raw material or pretreatment state, and data source. The output response fields include one or more of the following: degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. For the candidate peptide screening auxiliary model, the input fields include the molecular weight, hydrophobicity, net charge, secondary structure ratio, and amino acid composition of the enzymatic hydrolysis product. The output response fields are the activity candidate category or candidate priority.

3. The method for optimizing the enzymatic hydrolysis process of Polygonatum odoratum based on machine learning according to claim 2, characterized in that, The main process optimization model includes elastic network regression model, partial least squares regression model, support vector regression model, linear regression model, or response surface quadratic regression model.

4. The method for optimizing the enzymatic hydrolysis process of Polygonatum odoratum based on machine learning according to claim 2, characterized in that, The candidate peptide screening auxiliary models include support vector machine classification models, partial least squares discriminant models, logistic regression models, or random forest models.

5. The method for optimizing the hydrolysis process of Polygonatum odoratum based on machine learning according to claim 1, characterized in that, The regression prediction of the target response includes regression prediction of a single target response or regression prediction of a multi-target response; wherein, the single target response includes the degree of hydrolysis; and the multi-target response includes at least two of the following indicators: degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution.

6. The method for optimizing the enzymatic hydrolysis process of Polygonatum odoratum based on machine learning according to claim 5, characterized in that, In the optimization stage of regression prediction for multi-objective response, under given constraints of temperature, pH, reaction time, substrate concentration and enzyme-to-substrate ratio, predicted values ​​of degree of hydrolysis, peptide yield, antioxidant activity and molecular weight distribution are calculated for candidate process conditions. Subsequently, each predicted value is normalized and a comprehensive score is formed according to a preset weight. The candidate process conditions are ranked according to the comprehensive score, thereby outputting the optimal process conditions that take into account both high activity and high yield within the same constraint set.

7. A machine learning-based optimization system for the enzymatic hydrolysis process of Polygonatum odoratum, used to implement the machine learning-based optimization method for the enzymatic hydrolysis process of Polygonatum odoratum as described in any one of claims 1-6, characterized in that, include: Data acquisition module: used to collect the input conditions and output responses of the Polygonatum odoratum protein hydrolysis experiment; Field standardization module: used to uniformly process categorical features, numerical features, missing values, and dimensional differences; Machine learning modeling module: used to establish the mapping relationship between process conditions and target response; Optimization module: Used to search for optimal parameter combinations within the feasible process space; Verification and Iteration Module: Used to receive new experimental results of Polygonatum and complete closed-loop updates, so that the model gradually converges to the Polygonatum-specific process space.

8. The machine learning-based optimization system for the hydrolysis process of Polygonatum odoratum according to claim 7, characterized in that, The machine learning modeling module includes a feature encoding unit, a regression modeling unit, a classification modeling unit, and a model evaluation unit. The feature encoding unit is responsible for one-hot encoding, numerical standardization, and missing value imputation. The regression modeling unit is responsible for predicting the degree of hydrolysis, peptide yield, antioxidant activity, and molecular weight distribution. The classification modeling unit takes peptide molecular weight, hydrophobicity, net charge, secondary structure ratio, and optional amino acid composition features as input and outputs activity candidate categories or candidate priorities. The model evaluation unit is responsible for cross-validation, independent test set validation, accuracy, F1 score, and error index calculation.

9. A general-purpose computer device, characterized in that, include: Processor, memory, and input / output interface, wherein the memory stores a program; When the program is processed by the processor, it executes the machine learning-based optimization method for the hydrolysis process of Polygonatum odoratum as described in any one of claims 1-6.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to execute the machine learning-based optimization method for the hydrolysis process of Polygonatum odoratum as described in any one of claims 1-6.

Images