Cottonseed protein peptide enzymolysis extraction method based on machine learning optimization and application of cottonseed protein peptide enzymolysis extraction method

Abstract

The invention belongs to the technical field of polypeptide, and discloses an enzymolysis extraction method of cottonseed protein peptide based on machine learning optimization and application of the enzymolysis extraction method. The method comprises the following steps: setting key factors influencing enzymolysis conditions; performing a single-factor experiment on each key factor, and selecting corresponding key factor variable values according to the single-factor experiment to form a key factor set; according to the key factor set, generating experimental samples of different enzymolysis condition combinations by adopting a full-factor experimental design, performing an experiment according to the experimental samples, and collecting corresponding actual response data; taking the experimental sample as the input of a machine learning model to obtain corresponding response data; performing multi-objective optimization by using a natural heuristic optimization algorithm in combination with the trained machine learning model to obtain a key factor variable value corresponding to an optimal experimental sample; according to the method, the optimal key factor variable value is effectively searched in a complex enzymolysis condition parameter space, and the cottonseed protein peptide extraction performance can be comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of polypeptides. More specifically, the present invention relates to an enzymatic extraction method of cottonseed protein peptides optimized based on machine learning and its application. Background Art

[0002] Cottonseed protein peptides are bioactive peptides with high nutritional value. It contains rich essential amino acids such as arginine and lysine, which play important roles in maintaining the physiological functions and growth and development of the human body. Traditional optimization of enzymatic hydrolysis conditions mainly relies on the experimental trial-and-error method. Researchers need to conduct a large number of experiments to change the enzymatic hydrolysis conditions one by one and observe the effects of different factors on the yield and biological activity of cottonseed protein peptides. This method is not only time-consuming and laborious but also requires a large amount of experimental materials and resources.

[0003] The patent application with the publication number CN117648571A discloses a data-driven enzymatic hydrolysis efficiency prediction and soft sensing method, including: the first stage: training an enzymatic hydrolysis efficiency prediction model according to the data collected during the enzymatic hydrolysis production process; including the following steps: S1, establishment and data preprocessing of the data set; S2, variable screening based on GRA; S3, input dimension reduction based on KPCA; S4, principal component clustering based on K-means++; S5, training the LSSVM model and using BOA for parameter optimization; the second stage: using the measured data during the enzymatic hydrolysis production process and using the trained model to predict the enzymatic hydrolysis efficiency. This technical solution can directly and accurately estimate the final enzymatic hydrolysis efficiency according to the production condition variables. A large number of experimental tests verify the effectiveness of this method. The obtained model has a small error and a high fitting degree, and has good performance.

[0004] Although the above technical solution can meet most scenarios, through research and practical application of the above method and the existing technology, it is found that the above method and the existing technology have at least the following partial defects:

[0005] There is a lack of a link to explore in detail the influence of each key factor on the enzymatic hydrolysis efficiency in different value ranges, which is likely to miss the value situations of different factors that affect the yield or biological activity of cottonseed protein peptides, making the factors considered in the subsequent constructed model and prediction results not comprehensive enough and unable to fully explore the potential optimal combinations of various factors; the comprehensive enzymatic hydrolysis condition combination method is not used, making it difficult to fully capture the complex synergistic or antagonistic relationships between different factors. There is a lack of consideration for multi-objective optimization, resulting in difficulty in balancing the relationships between multiple key indicators such as yield and biological activity, and unable to provide enzymatic hydrolysis condition suggestions with the optimal comprehensive performance that meet the needs of diverse practical application scenarios.

[0006] In view of this, the present invention proposes an enzymatic extraction method of cottonseed protein peptides optimized based on machine learning and its application to solve the above problems. Summary of the Invention

[0007] In order to overcome the above defects of the prior art and achieve the above object, the present invention provides the following technical solution: an enzymatic extraction method of cottonseed protein peptide optimized based on machine learning, comprising the following steps:

[0008] Set the key factors affecting the enzymatic hydrolysis conditions, and the key factors include the type of enzyme, enzyme addition amount, enzymatic hydrolysis time, substrate concentration, and pH value;

[0009] Take each key factor as a key factor variable respectively, conduct single-factor experiments, and according to the results of the single-factor experiments, extract the key factor variable values corresponding to the single-factor experiments in which the yield of cottonseed protein peptide is higher than the preset yield threshold or the biological activity index is higher than the preset index threshold to form a key factor set;

[0010] Generate experimental samples with different combinations of key factor variable values according to each key factor variable value in the key factor set by using a full-factor experimental design, conduct experiments according to the experimental samples, collect the experimental samples and the corresponding response data. The experimental samples include the corresponding type of enzyme, enzyme addition amount, enzymatic hydrolysis time, substrate concentration, and pH value. Among them, the type of enzyme is encoded, and the response data includes the yield of cottonseed protein peptide and the biological activity index;

[0011] Take the experimental samples as the input of the machine learning model to obtain the corresponding response data; use a nature-inspired optimization algorithm combined with the trained machine learning model for multi-objective optimization to obtain the key factor variable values corresponding to the optimal experimental samples.

[0012] Further, the method for obtaining the optimal experimental samples includes;

[0013] Step 1: Encode each key factor variable value in each experimental sample by using a binary coding method;

[0014] Step 2: Set the population size to N, each individual represents an experimental sample, number the individuals to obtain the corresponding individual numbers of each individual;

[0015] Step 3: Decode each individual in the population into the actual key factor variable values, take the experimental samples corresponding to the actual key factor variable values as the input of the trained machine learning model, and obtain the predicted response data, that is, the predicted yield of cottonseed protein peptide and the predicted biological activity index;

[0016] Step 4: Define the fitness function and calculate the fitness function of each individual;

[0017] Step 5: Conduct non-dominated sorting on the individuals according to the fitness function to obtain the non-dominated front sequence;

[0018] Step 6: Select parents from the non-dominated front sequence to generate a parent set;

[0019] Step 7: Repeat the step of selecting parent individuals from the parent set to generate offspring times to generate offspring. The step of selecting parent individuals to generate offspring is as follows:

[0020] Randomly select R parent individuals, obtain the key factor variable values corresponding to the R parent individuals for weighting, obtain a new set of key factor variable values corresponding to a new experimental sample, and label the new experimental sample as offspring;

[0021] Step 8: According to the preset mutation probability interval, perform a mutation operation on the key factor variable values corresponding to the offspring by using random flipping to obtain new offspring;

[0022] Step 9: Combine the parents, offspring, and new offspring to form a new population;

[0023] Step 10: Repeat Steps 4 - 9 using the new population until the maximum number of iterations is reached, then stop the iteration, and label the population obtained after stopping the iteration as the maximum population;

[0024] Step 11: Repeat Step 5 for the maximum population to obtain a new non-dominated front sequence, and select the first B population individuals from the new non-dominated front sequence in the sorted order as the optimal experimental samples.

[0025] Furthermore, the method for defining the fitness function includes:

[0026] Taking the biological activity index f1(x) as the first objective function of the fitness function, and taking the reciprocal f2(x) of the deviation between the predicted yield of cottonseed protein peptide and the target yield of cottonseed protein peptide as the second objective function of the fitness function.

[0027] Furthermore, the method for performing non-dominated sorting on the population individuals to obtain a non-dominated front sequence includes:

[0028] Step 5.1: Initialize the non-dominated front sequence where n F is the number of non-dominated fronts included in the non-dominated front sequence, is the n F th non-dominated front. Initialize the number of times each individual is dominated to 0, and obtain the individual number of each individual;

[0029] Step 5.2: Obtain the number of times individual i is dominated in the order of individual numbers, where i is the individual number; The method for obtaining the number of times individual i is dominated includes:

[0030] For individual i, traverse other individuals j in the population except individual i, where j is the individual number other than number i, and the individual number j ≠ i, and compare the performance of individual i and individual j on all objective functions; specifically as follows:

[0031] If all objective function values of individual j are not lower than all objective function values of individual i, and at least one objective function value is higher than the objective function value of individual i, then individual i is dominated by individual j, and record the number of times n that individual i is dominated i Increment by 1; all objective function values include the first objective function value and the second objective function value. Among them, the first objective function value of individual i and the first objective function value of individual j are compared correspondingly, and the second objective function value of individual i and the second objective function value of individual j are compared correspondingly;

[0032] If individual i is not dominated by any individual j, then the number of times n that individual i is dominated i Remains unchanged;

[0033] If all objective function values of individual j are not lower than all objective function values of individual i, but none of the objective function values is higher than that of individual i, that is, all objective function values of individual j are equal to all objective function values of individual i, then individual i and individual j are non-dominated with each other, and the number of times n that individual i is dominated i Remains unchanged;

[0034] Step 5.3: Repeat Step 5.2 for all individuals in the population, obtain the number of times each individual is dominated, and sort the individuals in ascending order according to the number of times each individual is dominated to obtain an individual sequence, and add the individual sequence to the non-dominated front sequence in turn. Among them, individuals with the same number of times dominated are added to the same non-dominated front.

[0035] Furthermore, the method for selecting parents from the non-dominated front sequence to generate a parent set includes:

[0036] Step 6.1: Calculate the fitness score value of the nth individual in each non-dominated front:

[0037] Step 6.2: Calculate the crowding degree of individuals in each non-dominated front according to the fitness score value of the individuals;

[0038] Step 6.3: Calculate the comprehensive score of each individual according to the comprehensive score formula, and select the individuals with a comprehensive score higher than the preset score threshold as parents.

[0039] Furthermore, the method for calculating the crowding degree of individuals in each non-dominated front includes:

[0040] Step 6.3.1: Initialize the crowding degree of individuals in each non-dominated front to 0;

[0041] Step 6.3.2: Set the crowding degree of the individuals included in the boundary individuals of the individual sequence to infinity;

[0042] Step 6.3.3: Calculate the crowding degree of the nth individual in the middle individual of the mth individual sequence in the individual sequence except the boundary individuals through the crowding degree formula;

[0043] Step 6.3.4: Repeat Step 6.3.2 to Step 6.3.3 to calculate the crowding degree of each population individual included in the middle individuals in the individual sequence except the boundary individuals in the individual sequence.

[0044] Furthermore, the method for obtaining the experimental sample data of the offspring includes:

[0045] Randomly select R parental individuals, obtain the key factors corresponding to the R parental individuals and perform weighting according to the weighting formula; count the new experimental samples corresponding to the key factor variable values of T key factor dimensions obtained by weighting, and label the new experimental samples as offspring.

[0046] Furthermore, the method for performing single-factor experiments on each of the key factors includes:

[0047] Step 1: Select the key factor to be processed;

[0048] Step 2: Keep other key factors unchanged, only change the key factor to be processed, set θ different gradients of the key factor to be processed, and measure the cottonseed protein peptide yield and bioactivity indexes with respect to the variable values corresponding to the key factor to be processed;

[0049] Step 3: Select the key factor variable values with the cottonseed protein peptide yield higher than the preset yield threshold or the bioactivity index higher than the preset index threshold to form a key factor set.

[0050] Furthermore, the method for generating experimental samples with different combinations of key factor variable values by using the full-factor experimental design includes:

[0051] Step 1: Obtain the key factor set corresponding to each key factor according to the single-factor experiment;

[0052] Step 2: Perform a full permutation and combination of all the key factor variable values in each key factor set to obtain the corresponding experimental samples.

[0053] Furthermore, the training method of the machine learning model includes:

[0054] Pre-collect A groups of experimental training data, where the experimental training data includes experimental samples and response data corresponding to the experimental samples;

[0055] Use each set of experimental training data as the input of a machine learning model. The machine learning model takes the response data corresponding to each set of experimental samples as the output, and uses the actual response data corresponding to each set of experimental samples as the prediction target. Use minimizing the sum of prediction errors of all response data as the training target. Train the machine learning model until the sum of prediction errors converges and then stop training. The machine learning model is a deep neural network model.

[0056] The loss function value of the machine learning model is the comprehensive loss function Loss, and the comprehensive loss function Loss is the weighted value of the mean square error MSE1 and MSE2.

[0057] Application method of the enzymatic extraction method of cottonseed protein peptides optimized by machine learning, using the enzymatic extraction method of cottonseed protein peptides optimized by machine learning in feed.

[0058] Technical effects and advantages of the enzymatic extraction method of cottonseed protein peptides optimized by machine learning and its application in the present invention:

[0059] Through single-factor experiments on multiple key factors, it is possible to comprehensively and meticulously screen out the value levels that have a positive impact on the yield or bioactivity index of cottonseed protein peptides in each factor. Use the value levels screened out by each key factor to generate experimental samples of different enzymatic hydrolysis condition combinations by full-factor experimental design. This method can comprehensively investigate the interaction between factors and systematically explore the entire enzymatic hydrolysis condition space. It is possible to discover the synergistic or antagonistic effects between key factors, thereby finding a better combination of enzymatic hydrolysis conditions. Use a nature-inspired optimization algorithm combined with the trained machine learning model for multi-objective optimization, which can effectively search for the optimal solution in the complex enzymatic hydrolysis condition parameter space. The finally obtained optimal enzymatic hydrolysis conditions can comprehensively improve the performance of cottonseed protein peptide extraction. It is possible to produce cottonseed protein peptides with higher yield and better bioactivity based on the optimal enzymatic hydrolysis conditions in practical applications. Description of the Drawings

[0060] Figure 1 It is a flowchart of the enzymatic extraction method of cottonseed protein peptides optimized by machine learning according to the present invention;

[0061] Figure 2 It is a flowchart of the method for conducting single-factor experiments on each key factor according to the present invention;

[0062] Figure 3 It is a flowchart of the method for obtaining the optimal enzymatic hydrolysis conditions according to the present invention;

[0063] Figure 4 It is a flowchart of the method for obtaining the non-dominated front sequence according to the present invention. Detailed Embodiments

[0064] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without making creative efforts belong to the scope of protection of the present invention.

[0065] Example 1

[0066] Please refer to Figure 1 As shown, the enzymatic hydrolysis extraction method of cottonseed protein peptides optimized based on machine learning in this embodiment includes:

[0067] Set the key factors affecting the enzymatic hydrolysis conditions, and the key factors include the type of enzyme, enzyme addition amount, enzymatic hydrolysis temperature, enzymatic hydrolysis time, substrate concentration, and pH value.

[0068] Different types of enzymes have different cleavage sites and specificities; for example, alkaline protease mainly acts on the peptide bond on the carboxyl side of basic amino acids in the peptide bond, while the cleavage site and action mode of neutral protease are different from it. In the extraction of cottonseed protein peptides, selecting a suitable enzyme can specifically cleave specific peptide bonds of cottonseed protein, thereby affecting the size, quantity, and structure of the generated peptide segments; a suitable enzyme type can also make cottonseed protein more effectively hydrolyzed into peptides with specific functions and properties; by optimizing the type of enzyme, the yield of cottonseed protein peptides can be improved. Different types of enzymes have different hydrolysis degrees on cottonseed protein under the same conditions, and selecting an enzyme with high hydrolysis efficiency can convert more protein into peptides.

[0069] The enzyme addition amount directly affects the rate and degree of the enzymatic hydrolysis reaction; within a certain range, as the enzyme addition amount increases, the contact opportunity between the enzyme and the substrate (cottonseed protein) increases, which can accelerate the hydrolysis reaction; this is because more enzyme molecules can act on the cleavage sites on the cottonseed protein molecules simultaneously, making the protein decompose into peptides faster; however, when the enzyme addition amount exceeds a certain limit, enzyme inhibition may occur; this is due to reasons such as excessive enzyme molecules competing with each other for the substrate or generating steric hindrance, resulting in the enzymatic hydrolysis efficiency not increasing or even decreasing; optimizing to obtain the optimal enzyme addition amount can save costs while ensuring the enzymatic hydrolysis effect.

[0070] The enzymatic hydrolysis temperature has a significant impact on the activity of the enzyme. Within the appropriate temperature range, the enzyme molecules have high activity and can rapidly catalyze the hydrolysis reaction of cottonseed protein. This is because the appropriate temperature can keep the enzyme molecules in a proper spatial conformation, which is conducive to binding with the substrate and exerting catalytic effects. When the temperature deviates from the optimal temperature, the enzyme activity will decrease. Excessive temperature may cause the enzyme to denature and inactivate because high temperature will destroy the spatial structure of the enzyme molecules, causing them to lose their catalytic function. Too low temperature will reduce the enzyme activity and slow down the reaction rate. Moreover, different enzymes have different optimal temperatures. Determining the optimal enzymatic hydrolysis temperature can maximize the activity of the enzyme and increase the yield of cottonseed protein peptides.

[0071] The enzymatic hydrolysis time determines the duration of the interaction between the enzyme and the substrate. In the initial stage of the enzymatic hydrolysis reaction, as the hydrolysis time increases, the degree of protein hydrolysis by the enzyme continuously deepens, and the yield of cottonseed protein peptides gradually increases. This is because the enzyme has enough time to act on the substrate and break it down into peptide segments. However, as time extends, when the enzymatic hydrolysis reaction reaches a certain degree, the reaction may tend to equilibrium. At this time, continuing to extend the enzymatic hydrolysis time, the yield may not increase significantly, and may even lead to peptide loss or quality decline due to further hydrolysis of peptides or other side reactions (such as peptide polymerization, etc.).

[0072] Optimizing the enzymatic hydrolysis time can improve production efficiency. If the enzymatic hydrolysis time is too long, it will increase production costs and energy consumption. For example, through experiments, it is found that after enzymatic hydrolysis for 6 - 8 hours, the yield of cottonseed protein peptides basically reaches the maximum value, and the yield changes little when the time is further extended. Therefore, controlling the enzymatic hydrolysis time within this range can shorten the production cycle and improve equipment utilization rate on the premise of ensuring the yield.

[0073] The substrate concentration affects the collision probability between the enzyme and the substrate. Within a certain range, as the concentration of cottonseed protein increases, the collision opportunities between the enzyme and the substrate increase, and the reaction rate will accelerate. This is because there are more substrate molecules around the enzyme molecules available for binding and catalytic hydrolysis. However, when the substrate concentration is too high, substrate inhibition may occur. This is because too many substrate molecules may surround the enzyme molecules, hindering the effective binding of the active center of the enzyme molecules to the substrate, or generating other factors unfavorable to the reaction, resulting in a decrease in enzymatic hydrolysis efficiency.

[0074] Optimizing the substrate concentration can balance the reaction rate and the yield. For example, by experiments, determine the appropriate range of cottonseed protein concentration, within which a higher peptide yield can be obtained. When the cottonseed protein concentration is too high, appropriately reducing the concentration can improve the enzymatic hydrolysis efficiency. When the concentration is too low, appropriately increasing the concentration can make full use of the enzyme activity and improve production efficiency.

[0075] The pH value has an important influence on the activity and stability of enzymes; enzyme molecules have specific active centers and spatial structures, and their activity depends on a certain pH environment. Different enzymes have different optimal pH ranges; at the optimal pH, the charge state and spatial conformation of the active center of the enzyme molecule are most conducive to binding with the substrate and catalyzing the reaction; for example, some proteases have the highest activity in the pH range of 7-8, because at this pH, the active groups of the enzyme molecule (such as amino groups, carboxyl groups, etc.) are in a suitable ionized state and can effectively interact with the substrate molecules; when the pH value deviates from the optimal pH, the activity of the enzyme will decrease. This is because the change in pH value will affect the charge distribution and spatial structure of the enzyme molecule, resulting in a weakened binding ability between the active center of the enzyme molecule and the substrate, or the denaturation of the enzyme molecule itself; extreme pH values may cause the enzyme to be completely inactivated.

[0076] Optimizing the pH value can improve the activity of the enzyme, thereby increasing the yield of cottonseed protein peptides. For example, by experiments, determine the optimal pH value of each enzyme during the enzymatic hydrolysis of cottonseed protein, and carry out the enzymatic hydrolysis reaction at this pH, which can maximize the activity of the enzyme.

[0077] Take each key factor as a key factor variable respectively, conduct single-factor experiments, and according to the results of the single-factor experiments, extract the key factor variable values corresponding to the single-factor experiments in which the yield of cottonseed protein peptides is higher than the preset yield threshold or the biological activity index is higher than the preset index threshold in the single-factor experiment results to form a key factor set.

[0078] Refer to Figure 2 , the methods for conducting single-factor experiments on each key factor respectively include:

[0079] Step 1: Select the key factor to be processed (such as enzyme type, substrate concentration, pH value, etc.).

[0080] Step 2: Keep other key factors unchanged, only change the key factor to be processed, and set θ groups of different gradients of the key factor to be processed (for example, if the selected key factor to be processed is the enzymatic hydrolysis temperature, set the enzymatic hydrolysis temperature gradients as 30°C, 35°C, 40°C, 45°C, 50°C, 55°C, 60°C), and measure the yield of cottonseed protein peptides and the biological activity index with the variable values corresponding to the key factor to be processed. In this way, the general influence trend of each factor on the enzymatic hydrolysis effect can be initially determined, and the possible better range of each factor can be found.

[0081] Step 3: Select the key factor variable values with the yield of cottonseed protein peptides higher than the preset yield threshold or the bioactivity index higher than the preset index threshold to form a key factor set (for example, when the enzymatic hydrolysis temperature is between 40 - 55 °C, it meets the condition that the yield of cottonseed protein peptides is higher than the preset yield threshold or the bioactivity index is higher than the preset index threshold. Then, in subsequent experiments, the enzymatic hydrolysis temperature range can be narrowed down to this interval to further optimize the values of other factors within this enzymatic hydrolysis temperature range).

[0082] In the above single-factor experiments, each group of single-factor experiments was also repeated multiple times (such as 3 - 5 times) to reduce experimental errors and ensure the reliability of the data. Through the single-factor experiments, the key factor value levels that have a significant impact on the yield and bioactivity index of cottonseed protein peptides can be preliminarily screened out; thus, appropriate key factor value levels can be selected for the full-factor experiment, reducing unnecessary experimental workload and making the full-factor experiment more targeted.

[0083] According to each key factor variable value in the key factor set, experimental samples with different combinations of key factor variable values are generated using a full-factor experimental design. Experiments are conducted based on the experimental samples, and the experimental samples and corresponding actual response data are collected. The experimental samples include the corresponding enzyme type, enzyme addition amount, enzymatic hydrolysis time, substrate concentration, and pH value. Among them, the enzyme type is encoded, and the response data includes the yield of cottonseed protein peptides and the bioactivity index.

[0084] The method of generating experimental samples with different combinations of enzymatic hydrolysis conditions using a full-factor experimental design includes:

[0085] Step 1: Obtain the key factor set corresponding to each key factor according to the single-factor experiment; for example, considering 3 different enzymes for the enzyme type, including enzyme A, enzyme B, and enzyme C, encoded as 1, 2, and 3 respectively, then the enzyme type includes three value levels of 1, 2, and 3; the enzyme addition amount includes four value levels of 0.5%, 1%, 1.5%, and 2%; the enzymatic hydrolysis temperature includes three value levels of 40 °C, 50 °C, and 60 °C; the enzymatic hydrolysis time includes three value levels of 4 hours, 6 hours, and 8 hours; the substrate concentration includes two value levels of 10% and 20%; the pH value includes three value levels such as 7, 8, and 9.

[0086] Step 2: Perform a full permutation and combination of all key factor variable values in each key factor set to obtain the corresponding experimental samples.

[0087] The method of obtaining the number of experimental samples includes:

[0088] Multiply the value levels of each key factor, i.e., 3 * 4 * 3 * 3 * 2 * 3 = 648, to obtain 648 experimental samples. This means that 648 different experimental samples need to be experimented on, and each experimental sample is an experimental group. The purpose of doing this is to comprehensively and systematically study the effects of each key factor and the interactions between key factors on the yield and bioactivity indexes of cottonseed protein peptides. Through the full-factor experiment, very detailed and comprehensive data can be obtained, enabling accurate analysis of the main effects and interaction effects of each key factor, providing a sufficient basis for subsequent optimization.

[0089] Use the experimental samples as the input of the machine learning model to obtain the corresponding response data; use the nature-inspired optimization algorithm combined with the trained machine learning model for multi-objective optimization to obtain the key factor variable values corresponding to the optimal experimental samples.

[0090] The training method of the machine learning model includes:

[0091] Pre-collect the experimental training data of group A. The experimental training data includes experimental samples and the response data corresponding to the experimental samples;

[0092] Use each group of experimental training data as the input of the machine learning model. The machine learning model takes the response data corresponding to each group of experimental samples as the output, and the actual response data corresponding to each group of experimental samples as the prediction target; use minimizing the sum of the prediction errors of all response data as the training target; train the machine learning model until the sum of the prediction errors reaches convergence and then stop training; the machine learning model is a deep neural network model;

[0093] The loss function value of the machine learning model is the comprehensive loss function Loss, and the comprehensive loss function Loss is the weighted value of the mean square error MSE1 and MSE2;

[0094] The formula for the comprehensive loss function: Loss = α1 × MSE1 + α2 × MSE2;

[0095]

[0096] Among them, α1 is the weight coefficient of the mean square error MSE1; α2 is the weight coefficient of the cross-entropy loss function H; a is the a-th group of experimental samples; A is the number of groups of experimental samples; is the yield of cottonseed protein peptides corresponding to the a-th group of experimental samples; z a is the actual yield of cottonseed protein peptides corresponding to the a-th group of experimental samples; The bioactivity index corresponding to the a-th group of experimental samples; h a is the actual bioactivity index corresponding to the a-th group of experimental samples.

[0097] The nature-inspired optimization algorithm is preferably the non-dominated sorting genetic algorithm. The non-dominated sorting genetic algorithm is used to combine with the trained machine learning model for multi-objective optimization to obtain the key factor variable values corresponding to the optimal experimental samples.

[0098] Referring to Figure 3 , the method for obtaining experimental samples includes:

[0099] Step 1: Encode each key factor variable value in each experimental sample using the binary encoding method; for example, for the enzyme addition amount in the range of 0-2%, if binary encoding is used, it can be divided into a certain precision (such as 0.01%), and a binary string of a certain length is used to represent the enzyme addition amount.

[0100] Step 2: Set the population size to N. Each individual represents an experimental sample, and the individuals are numbered to obtain the individual number corresponding to each individual. The population size N can be set according to the number of experimental samples, such as 800 individuals.

[0101] Step 3: Decode each individual in the population into the actual key factor variable values, and use the actual key factor variable values as the input of the trained machine learning model to obtain the predicted response data, that is, the predicted yield of cottonseed protein peptides and the predicted values of bioactivity indexes.

[0102] Step 4: Define the fitness function, calculate the fitness function of each individual. Take the bioactivity index f1(x) as the first objective function of the fitness function, and take the reciprocal of the deviation between the predicted yield of cottonseed protein peptides and the target yield of cottonseed protein peptides as the second objective function of the fitness function, where x is the vector of key factor variable values; Y is the predicted yield of cottonseed protein peptides; Y target is the target yield of cottonseed protein peptides; among them, the higher the bioactivity index, the greater the value of the first objective function; the smaller the deviation between the predicted yield of cottonseed protein peptides and the target yield, the greater the value of the second objective function.

[0103] Step 5: Perform non-dominated sorting on the individuals according to the fitness function to obtain the non-dominated front sequence.

[0104] Referring to Figure 4 , the method for performing non-dominated sorting on the population individuals to obtain the non-dominated front sequence includes:

[0105] Step 5.1: Initialize the non-dominated front sequence where n F is the number of non-dominated fronts included in the non-dominated front sequence, is the n F th non-dominated front. Initialize the number of times each individual is dominated to 0, and obtain the individual number of each individual;

[0106] Step 5.2: Obtain the number of times individual i is dominated in the order of individual numbers, where i is the individual number; the method for obtaining the number of times individual i is dominated includes:

[0107] For individual i, traverse other individuals j in the population except individual i, where j is the individual number other than number i, and the individual number j ≠ i, and compare the performances of individual i and individual j on all objective functions; specifically as follows:

[0108] If all objective function values of individual j are not lower than all objective function values of individual i, and at least one objective function value is higher than the objective function value of individual i, then individual i is dominated by individual j, and record the number of times n that individual i is dominated i Increment by 1; all objective function values include the first objective function value and the second objective function value. Among them, the first objective function value of individual i and the first objective function value of individual j are compared correspondingly, and the second objective function value of individual i and the second objective function value of individual j are compared correspondingly;

[0109] If individual i is not dominated by any individual j, then the number of times n that individual i is dominated i Remains unchanged;

[0110] If all objective function values of individual j are not lower than all objective function values of individual i, but none of the objective function values is higher than that of individual i, that is, all objective function values of individual j are equal to all objective function values of individual i, then individual i and individual j are non-dominated with each other, and the number of times n that individual i is dominated i Remains unchanged;

[0111] Step 5.3: Repeat Step 5.2 for all individuals in the population to obtain the number of times each individual is dominated, and sort the individuals in ascending order according to the number of times each individual is dominated to obtain an individual sequence. Then add the individual sequence to the non-dominated front sequence in turn. Among them, individuals with the same number of times dominated are added to the same non-dominated front. For example, sorting the individuals according to the number of times dominated gives the individual sequence: 0(3), 0(53), 1(45), 2(15), 3(4), 3(54), 4(76), 5(32), 6(41), 6(69), 7(85), 8(46), 9(13), 10(5)……; where the number in the parentheses is the individual number. Add the individual sequence to the non-dominated front sequence in turn. Among them, individuals numbered 3 and 53 are added to the first non-dominated front F1, the individual numbered 45 is added to the second non-dominated front F2, the individual numbered 15 is added to the third non-dominated front F3, the individuals numbered 4 and 54 are added to the fourth non-dominated front F4, the individual numbered 76 is added to the fifth non-dominated front F5, the individual numbered 32 is added to the sixth non-dominated front F6, the individuals numbered 41 and 69 are added to the seventh non-dominated front F7, the individual numbered 85 is added to the eighth non-dominated front F8, the individual numbered 46 is added to the ninth non-dominated front F9, the individual numbered 13 is added to the tenth non-dominated front F 10 In, the individual numbered 5 is added to the eleventh non-dominated front F 11 In, and the remaining individual sequence is added to the corresponding non-dominated front sequence by analogy.

[0112] Through this non-dominated sorting method, the individuals in the population can be divided into different levels according to their superiority and inferiority relationships in the multi-objective space, providing a basis for subsequent operations such as selection, crossover, and mutation to achieve multi-objective optimization.

[0113] Step 6: Select parents from the non-dominated front sequence to generate a parent set.

[0114] The method of selecting parents from the non-dominated front sequence includes:

[0115] Step 6.1: Calculate the fitness score value S of the nth individual in each non-dominated front f,n :

[0116] S f,n =β1f1(x)+β2f2(x);

[0117] Among them, β1 and β2 are the weight coefficients of the bioactivity index f1(x) and the deviation f2(x) between the predicted yield of cottonseed protein peptides and the target yield, respectively;

[0118] Step 6.2: Calculate the crowding degree of individuals in each non-dominated front according to the fitness score value of the individual.

[0119] Step 6.3: Calculate the comprehensive score of each individual according to the comprehensive score formula, and select the individuals with a comprehensive score higher than the preset score threshold as the parental generation. The comprehensive score formula is as follows:

[0120] S K = ω f S f,K + ω d d m,K ;

[0121] Wherein, S K is the comprehensive score of the individual with the individual number K; S f,K is the fitness score value of the individual with the individual number K; d m,K is the crowding degree of the individual with the individual number K; ω f and ω d are the weight coefficients of the fitness score value and the crowding degree respectively.

[0122] The method for calculating the crowding degree of individuals in each of the non-dominated fronts includes:

[0123] Step 6.3.1: Initialize the crowding degree of individuals in each non-dominated front to 0;

[0124] Step 6.3.2: Set the crowding degree of the individuals included in the boundary individuals of the individual sequence to infinity;

[0125] Step 6.3.3: Calculate the crowding degree of the nth individual v m,n (v1 < v m < v N ) in the middle of the mth individual sequence in the individual sequence except the boundary individuals, where v1 is the first individual sequence and v N is the Nth individual sequence. The crowding degree formula is as follows:

[0126]

[0127] Wherein, d m,n is the crowding degree of the nth individual in the middle of the mth individual sequence; S f (v m+1,n ) is the fitness value of the nth individual in the middle of the (m + 1)th individual sequence; S f (v m-1,n ) is the fitness value of the nth individual in the middle of the (m - 1)th individual sequence; and are the maximum and minimum values of the fitness score value of the individual respectively; is the feature weight; d e (v m,n ) is the feature distance of the nth individual of the middle individual in the mth individual sequence; τ t (v m,n ) is the value of the tth key factor variable of the nth individual of the middle individual in the mth individual sequence; τ t (v m+1,n ) is the value of the tth key factor variable of the nth individual of the middle individual in the (m + 1)th individual sequence; τ t (v m-1,n ) is the value of the tth key factor variable of the nth individual of the middle individual in the (m - 1)th individual sequence; T is the total number of key factor dimensions;

[0128] Step 6.3.4: Repeat Step 6.3.2 to Step 6.3.3 to calculate the crowding degree for each population individual included in the middle individuals of the individual sequence except for the boundary individuals in the individual sequence.

[0129] Step 6.4: Calculate the comprehensive score of each individual according to the comprehensive score formula, and select the individuals with comprehensive scores higher than the preset score threshold as the parent generation; the comprehensive score formula is as follows:

[0130] S K = ω f S f,K + ω d d m,K ;

[0131] Among them, S K is the comprehensive score of the individual with individual number K; S f,K is the fitness score value of the individual with individual number K; d m,K is the crowding degree of the individual with individual number K; ω f and ω d are the weight coefficients of the fitness score value and the crowding degree respectively, which can be adjusted according to the optimization requirements.

[0132] Step 7: Repeat times the step of selecting parent individuals from the parent generation set to generate offspring, and generate offspring. The step of selecting parent individuals to generate offspring is specifically as follows:

[0133] Randomly select R parent individuals, obtain the weighted key factor variable values corresponding to the R parent individuals, obtain a new experimental sample corresponding to a group of new key factor variable values, and label the new experimental sample as the offspring;

[0134] The method for obtaining the experimental sample data of the offspring includes:

[0135] Randomly select R parental individuals, obtain the key factors corresponding to the R parental individuals, and perform weighting according to the weighting formula. The weighting formula is as follows:

[0136]

[0137] Among them, Z c is the key factor variable value of the c-th key factor dimension obtained by weighting; y rc is the key factor variable value of the c-th key factor dimension of the r-th parental individual; ρ rc is the weight coefficient of y rc . According to the above weighting formula, count the new experimental samples corresponding to the key factor variable values of T key factor dimensions obtained by weighting, and mark the new experimental samples as offspring.

[0138] Step 8: According to the preset mutation probability interval, perform a mutation operation on the key factor variable values corresponding to the offspring using the random flipping method to obtain new offspring;

[0139] Step 9: Combine the parental individuals, offspring, and new offspring to form a new population;

[0140] Step 10: Repeat Steps 4-9 using the new population until the maximum number of iterations is satisfied, then stop the iteration, and mark the population obtained after stopping the iteration as the maximum population;

[0141] Step 11: Repeat Step 5 for the maximum population to obtain a new non-dominated front sequence, and select the first B population individuals from the new non-dominated front sequence as the optimal experimental samples according to the sorting order.

[0142] Example 2

[0143] Please refer to Figure 1 As shown, this example further improves the design on the basis of Example 1. In Example 1, based on the preset mutation probability interval combined with the random flipping method, it may be relatively blind when exploring the solution space, without considering the quality of the current solution and the evolutionary stage of the population. It may frequently perform large-scale and unreasonable mutations on some individuals that are already in a better region, resulting in the algorithm being difficult to converge to a better solution; moreover, the mutation operation lacks adaptive adjustment of individual fitness and population evolutionary status. When the population is already close to the optimal solution region, it still performs random flipping mutations according to a fixed probability interval, which may destroy the better solution structure that has been found, causing the population to oscillate near the optimal solution and the convergence speed to slow down; therefore, this example provides an enzymatic extraction method of cottonseed protein peptides optimized based on machine learning, including: mutating the offspring individuals based on self-regulated mutation probability and self-regulated mutation step size.

[0144] A method for mutating offspring based on self - adjusted mutation probability and self - adjusted mutation step size includes:

[0145] For the l - th generation of offspring, perform mutation based on the adaptive mutation probability and the adaptive mutation step size:

[0146] Calculate the adaptive mutation probability according to the fitness score value of the offspring:

[0147]

[0148] Among them, p G (X) is the mutation probability of offspring X; S f,X is the fitness score value of offspring X; is the average fitness score value; μ X is the relative fitness score value of offspring X; p G1 is the preset upper limit of the mutation probability; p G2 is the preset lower limit of the mutation probability; μ min is the lower limit of the relative fitness score; μ max is the upper limit of the relative fitness score; when the relative fitness score value of offspring X is relatively low (lower than μ min ), the mutation probability takes a higher value p G1 to increase the exploration of the poorer offspring; when the relative fitness score value of offspring X is relatively high (higher than μ max ), the mutation probability takes a lower value p G2 to avoid destroying the better offspring; when the relative fitness score value of offspring X is in the middle range, the mutation probability linearly adjusts from p G1 to p G2 .

[0149] Calculate the mutation step size according to the fitness score value of the offspring as follows:

[0150]

[0151] Among them, λ G (X, I) is the mutation step size of offspring X in the c - th key factor; λ G1 is the preset upper limit of the mutation step size; λ G2 is the preset lower limit of the mutation step size;

[0152] For all L generations of offspring, perform mutation based on the mutation probability and the mutation step size:

[0153] Define the convergence index ρ as follows:

[0154]

[0155] Among them, is the fitness of the optimal individual in the l - th generation; is the fitness of the optimal individual of the (l - 1)-th generation;

[0156] Calculate the mutation probability based on the convergence index δ:

[0157]

[0158] where is the mutation probability of the offspring of the l-th generation; is the mutation probability of the offspring of the (l - 1)-th generation; w1 and w2 are probability adjustment coefficients; ε is a preset convergence threshold;

[0159] Calculate the mutation step size of the offspring of the l-th generation on the c-th key factor based on the convergence index δ

[0160]

[0161] where is the mutation step size of the offspring of the (l - 1)-th generation on the c-th key factor; w1 and w2 are step size adjustment coefficients.

[0162] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

[0163] Finally: The above description is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An enzymatic extraction method of cottonseed protein peptides based on machine learning optimization, characterized in that: The steps include: Set the key factors that affect the enzymatic hydrolysis conditions, including the type of enzyme, enzyme addition amount, enzymatic hydrolysis time, substrate concentration and pH value; Each key factor is used as a key factor variable to conduct a single factor experiment. According to the single factor experiment results, the key factor variable values ​​corresponding to the single factor experiment in which the yield of cottonseed protein peptide is higher than the preset yield threshold or the biological activity index is higher than the preset index threshold in the single factor experiment results are extracted to form a key factor set; According to each key factor variable value in the key factor set, a full factorial experimental design is used to generate experimental samples with different combinations of key factor variable values, and experiments are conducted according to the experimental samples, and the experimental samples and corresponding actual response data are collected. The experimental samples include the type of corresponding enzyme, enzyme addition amount, enzymolysis time, substrate concentration and pH value, wherein the type of enzyme is coded and represented, and the response data include the yield and biological activity index of cottonseed protein peptide; The experimental samples are used as the input of the machine learning model to obtain the corresponding response data; the nature-inspired optimization algorithm is combined with the trained machine learning model to perform multi-objective optimization to obtain the key factor variable values ​​corresponding to the optimal experimental samples.

2. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 1, characterized in that: Methods for obtaining optimal experimental samples include; Step 1: Use binary coding method to encode the value of each key factor variable in each experimental sample; Step 2: Set the population size to N, each individual represents an experimental sample, number the individuals, and obtain the individual number corresponding to each individual; Step 3, decoding each individual in the population into an actual key factor variable value, using the experimental samples corresponding to the actual key factor variable value as the input of the trained machine learning model, and obtaining the response data prediction value, i.e., the yield prediction value and the biological activity index prediction value of the cottonseed protein peptide; Step 4: Define the fitness function and calculate the fitness function of each individual; Step 5: Perform non-dominated sorting on the individuals according to the fitness function to obtain a non-dominated frontier sequence; Step 6: Select a parent from the non-dominated frontier sequence to generate a parent set; Step 7: Repeat the steps of selecting parent individuals from the parent set to generate offspring θ times to generate θ offspring. The steps of selecting parent individuals to generate offspring are as follows: Randomly select R parent individuals, obtain the key factor variable values ​​corresponding to the R parent individuals, and weight them to obtain a new set of experimental samples corresponding to a new set of key factor variable values, and mark the new experimental samples as offspring; Step 8: According to the preset mutation probability interval, the key factor variable values ​​corresponding to the offspring are mutated by random flipping to obtain new offspring; Step 9: Merge the parent generation, offspring generation and new offspring generation to form a new population; Step 10: Repeat steps 4-9 with the new population until the maximum number of iterations is met, then stop the iteration and mark the population obtained after the iteration is stopped as the maximum population; Step 11: Repeat step 5 for the largest population to obtain a new non-dominated frontier sequence, and select the first B population individuals from the new non-dominated frontier sequence in sorted order as the optimal experimental samples.

3. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 2, characterized in that: The method of defining the fitness function includes: The biological activity index f1(x) is used as the first objective function of the fitness function, and the inverse f2(x) of the deviation between the yield prediction value of the cottonseed protein peptide and the target yield of the cottonseed protein peptide is used as the second objective function of the fitness function.

4. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 3, characterized in that: The method of performing non-dominated sorting on the individuals of the population to obtain a non-dominated frontier sequence includes: Step 5.1: Initialize the non-dominated frontier sequence Among them, n F is the number of non-dominated frontiers contained in the non-dominated frontier sequence, For nth F non-dominated frontiers, initialize the number of times each individual is dominated to 0, and obtain the individual number of each individual; Step 5.2, obtaining the number of times individual i is dominated according to the order of individual numbers, where i is the individual number; the method for obtaining the number of times individual i is dominated includes: For individual i, traverse other individuals j in the population except individual i, where j is the individual number other than individual i, individual number j≠i, and compare the performance of individual i and individual j on all objective functions; the details are as follows: If all objective function values ​​of individual j are not lower than all objective function values ​​of individual i, and at least one objective function value is higher than the objective function value of individual i, then individual i is dominated by individual j, and the number of times individual i is dominated is recorded. i Add 1; all objective function values ​​include the first objective function value and the second objective function value, wherein the first objective function value of individual i is compared with the first objective function value of individual j, and the second objective function value of individual i is compared with the second objective function value of individual j; If individual i is not dominated by any individual j, then the number of times individual i is dominated is n i constant; If all objective function values ​​of individual j are not lower than all objective function values ​​of individual i, but none of them is higher than that of individual i, that is, all objective function values ​​of individual j are equal to all objective function values ​​of individual i, then individuals i and j are mutually non-dominated, and the number of times individual i is dominated is n i constant; Step 5.3: Repeat step 5.2 for all individuals in the population, obtain the number of times each individual is dominated, and sort the individuals in ascending order according to the number of times each individual is dominated, obtain the individual sequence, and add the individual sequence to the non-dominated frontier sequence in sequence, where individuals with the same number of times dominated are added to the same non-dominated frontier.

5. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 4, characterized in that: The method for selecting a parent generation from the non-dominated frontier sequence to generate a parent generation set comprises: Step 6.

1. Calculate the fitness score of the nth individual in each non-dominated frontier: Step 6.2, calculate the crowding degree of each individual in the non-dominated frontier according to the fitness score value of the individual; Step 6.3: Calculate the comprehensive score of each individual according to the comprehensive score formula, and select individuals with comprehensive scores higher than the preset score threshold as parents.

6. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 5, characterized in that: The method for calculating the crowding degree of each individual in the non-dominated front includes: Step 6.3.1, initialize the crowding degree of each individual in the non-dominated front to 0; Step 6.3.2, set the crowding degree of the individuals included in the boundary individuals of the individual sequence to be infinite; Step 6.3.3, calculate the crowding degree of the nth individual in the middle of the mth individual sequence except the boundary individuals in the individual sequence by using the crowding degree formula; Step 6.3.4: Repeat steps 6.3.2 to 6.3.3 to calculate the crowding degree of the population individuals contained in each intermediate individual in the individual sequence except the boundary individuals.

7. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 6, characterized in that: The method for obtaining the experimental sample data of the offspring includes: R parent individuals are randomly selected, and the key factors corresponding to the R parent individuals are obtained and weighted according to a weighting formula; new experimental samples corresponding to the key factor variable values ​​of T key factor dimensions are obtained by statistical weighting, and the new experimental samples are marked as offspring.

8. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 1, characterized in that: The method of conducting a single factor experiment on each of the key factors comprises: Step 1: Select the key factors to be processed; Step 2, other key factors are fixed unchanged, only the key factor to be treated is changed, different gradients of the key factors to be treated are set in groups θ, and the yield and bioactivity index of cottonseed protein peptides are measured along with the variable values ​​corresponding to the key factors to be treated; Step 3, selecting key factor variable values ​​whose cottonseed protein peptide yield is higher than a preset yield threshold or whose biological activity index is higher than a preset index threshold to form a key factor set.

9. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 1, characterized in that: The method of using the full factorial experimental design to generate experimental samples with different combinations of key factor variable values ​​includes: Step 1: Obtain the key factor set corresponding to each key factor according to the single factor experiment; Step 2: Perform full permutations and combinations on all key factor variable values ​​in each key factor set to obtain the corresponding experimental samples.

10. The method for enzymatic extraction of cottonseed protein peptides based on machine learning optimization according to claim 1, characterized in that: The training method of the machine learning model includes: Collecting a group of experimental training data in advance, the experimental training data includes experimental samples and response data corresponding to the experimental samples; Each group of experimental training data is used as the input of the machine learning model, and the machine learning model uses the response data corresponding to each group of experimental samples as output, and the actual response data corresponding to each group of experimental samples as the prediction target; minimizing the sum of the prediction errors of all response data is used as the training target; the machine learning model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the machine learning model is a deep neural network model; The loss function value of the machine learning model is the comprehensive loss function Loss, and the comprehensive loss function Loss is the weighted value of the mean square error MSE1 and MSE2.

11. An application method of the enzymatic extraction method of cottonseed protein peptides based on machine learning optimization, characterized in that: Application of the enzymatic extraction method of cottonseed protein peptides based on machine learning optimization as described in any one of claims 1 to 10 in feed.

Images