A broccoli isothiocyanate extraction process optimization method based on response surface-neural network

Abstract

The application discloses a broccoli isothiocyanate extraction process optimization method based on a response surface-nerve network, and belongs to the technical field of natural active ingredient extraction optimization. A three-wavelength correction quantitative detection method is established; single factor tests are used to screen three core variables, i.e., exogenous enzyme addition amount, enzymolysis temperature and enzymolysis time; a Box-Behnken response surface method is used to construct a quadratic regression model to analyze factor effects and interactions; a three-layer feedforward BP artificial nerve network is constructed based on the same test data, and a genetic algorithm is combined to carry out global optimization to obtain an optimal process, under which the actual isothiocyanate extraction amount is 7.84+ / -0.11 mg / g, and the predicted relative error is 2.49%, which is better than that of a single response surface model. Antioxidation detection shows that the DPPH clearance rate of 0.6 mg / mL extract is 92.08%, and the extract has good in-vitro antioxidant activity. The application overcomes the defects of single model, such as insufficient prediction accuracy and easy local optimum, and can be used for large-scale preparation of natural antioxidant feed additives, and provides a standardized composite modeling scheme for the broccoli isothiocyanate extraction process.

Description

Technical Field

[0001] This invention relates to the field of natural active ingredient extraction optimization technology, and more specifically to an optimization method for the extraction process of broccoli isothiocyanate based on response surface methodology (RSM). Background Technology

[0002] In livestock and poultry farming, factors such as heat stress, high-concentrate diets, and pathogen infections can easily disrupt the redox balance in animals, inducing oxidative stress, which in turn interferes with nutrient digestion and absorption, reduces feed utilization, and affects the quality of livestock products. Developing safe-to-source, well-defined natural antioxidants has become an important research direction in the current livestock industry.

[0003] Isothiocyanates (ITCs) in broccoli (Brassica oleracea L. var. italica Plenck) are natural antioxidants derived from the hydrolysis of glucosinolates (GSLs) by myrosinase. They are a class of multifunctional natural active ingredients with various effects, including antioxidant, anti-inflammatory, and antimicrobial properties. Optimizing a highly adaptable and efficient extraction process for broccoli ITCs is of great significance for the development of natural antioxidant feed additives.

[0004] Process optimization typically relies on advanced modeling methods to reveal the complex relationships between variables. The two most commonly used modeling methods are Response Surface Methodology (RSM) and Artificial Neural Networks (ANNs). RSM is a classic statistical method for multi-parameter optimization, capable of analyzing the interactions between factors through mathematical models to obtain optimal process parameters. However, while RSM, as a classic statistical method for multi-parameter optimization, can construct quadratic regression equations to analyze the main effects and interactions of factors, it has significant limitations in fitting complex nonlinear mapping relationships. Artificial Neural Networks do not require pre-defined function forms and have high accuracy in nonlinear fitting, but their model interpretability is poor, they are prone to getting trapped in local optima when used alone, and they lack statistical support. Existing single optimization models suffer from large prediction errors and insufficient optimization accuracy, making it difficult to simultaneously meet the dual needs of process law analysis and accurate prediction of optimal parameters.

[0005] In the extraction of ITCs from broccoli, the release and transformation of the target product ITCs face dual constraints: on the one hand, its precursor GSLs have poor chemical stability and are easily degraded by environmental factors such as heat, light, and metal ions; on the other hand, the activity of myrosinase, which catalyzes the hydrolysis of GSLs, is significantly regulated by multiple factors such as pH, temperature, and endogenous inhibitors, resulting in large fluctuations in enzymatic hydrolysis efficiency. In addition, the endogenous enzyme is partially inactivated during the freeze-thaw intervention process, further limiting the efficient transformation of ITCs. The aforementioned bottlenecks at the experimental level place higher demands on process optimization methods. However, existing studies have significant methodological limitations: RSM optimization is mostly limited to a single extraction stage (such as optimizing only the ethanol concentration or only the enzymatic hydrolysis time), failing to establish a coupled process model of enzymatic hydrolysis and extraction, and thus failing to reflect the true laws governing ITC generation under the interaction of multiple factors; ANN optimization generally directly applies a general network architecture and training parameters, without considering the Box function. Behnken's design, using small-sample experimental data for network depth, activation function, and training strategy selection, resulted in a significantly increased risk of overfitting and insufficient generalization ability. Therefore, constructing a coupled optimization method for broccoli ITCs extraction that integrates the advantages of response surface methodology and neural networks has become a critical scientific problem urgently needing to be solved to improve ITCs yield and process robustness. How to provide an optimization method for broccoli isothiocyanate extraction based on response surface methodology and neural networks is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides an optimization method for the extraction process of broccoli isothiocyanate based on response surface methodology (RSM) neural networks, aiming to solve the above-mentioned technical problems.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: An optimization method for broccoli isothiocyanate extraction process based on response surface methodology (RSM) neural networks includes the following steps: S1. Broccoli powder was obtained by freeze-drying broccoli, followed by pulverizing and sieving. Exogenous black mustard enzyme crude solution was prepared using white mustard seeds as raw material, and the black mustard enzyme activity was determined. S2. Weigh broccoli powder, add phosphate buffer and exogenous black myrosinase solution, and enzymatically hydrolyze in a water bath at 30~50℃ for 1~3 hours. Then add 75% ethanol solution and shake to extract. After centrifugation, take the supernatant and filter it through a filter membrane to obtain crude isothiocyanate extract. S3. Prepare isothiocyanate standard solution and plot standard curve. Calculate the corrected absorbance using the three-wavelength correction method at 239nm, 249nm, and 259nm. Determine the isothiocyanate concentration in the crude extract based on the standard curve and calculate the isothiocyanate extraction amount. S4. Determine the optimal material-liquid ratio and screen out three core optimization factors: exogenous enzyme addition amount, enzymatic hydrolysis temperature, and enzymatic hydrolysis time. S5. A quadratic regression model was constructed using Box-Behnken design, with the following factors: exogenous enzyme addition amount 6~10mL, enzymatic hydrolysis temperature 30~50℃, and enzymatic hydrolysis time 1.5~2.5h. The main effects and interactions of each factor were analyzed to obtain the theoretically optimal process parameters predicted by the response surface model. S6. Based on the Box-Behnken response surface experimental data, a three-layer feedforward BP neural network with 3 neurons in the input layer, 7-12 neurons in the hidden layer, and 1 neuron in the output layer was constructed. The experimental data was divided into training set, validation set, and test set to complete the network training and obtain a neural network prediction model for isothiocyanate extraction amount. S7. With the goal of maximizing the extraction yield of isothiocyanate, the predicted value of the BP neural network is used as the fitness function of the genetic algorithm for global optimization to obtain the optimal extraction process parameters. S8. Conduct multiple batches of parallel verification tests on the optimal extraction process obtained through optimization, and simultaneously determine the in vitro antioxidant activity of the obtained isothiocyanate extract.

[0008] Furthermore, the preparation of the exogenous black mustard enzyme solution and the determination of its enzyme activity in S1 are specifically as follows: Take fresh white mustard seeds, crush and sieve them, weigh 5g of mustard seed powder and add it to 250mL of pre-cooled 0.1mol / L, pH6.5 phosphate buffer, sonicate for 15min under ice bath conditions, centrifuge at 5000r / min for 10min, and collect the supernatant to obtain exogenous black mustard enzyme solution. The method for determining the activity of black mustard enzyme is as follows: Take 1g of frozen mustard seed powder, add 6mL of pre-cooled 0.1mol / L, pH 6.5 phosphate buffer, grind into a homogenate, and centrifuge at 12000r / min for 15min at 4℃ to obtain crude enzyme solution; take 250μL of crude enzyme solution and mix with an equal volume of 2mmol / L potassium black mustard glucosinolate solution, react in a water bath at 37℃ for 15min, and terminate the reaction by heating in a boiling water bath for 5min; use a glucose assay kit to measure the absorbance at 510nm; the enzyme activity unit U is defined as the amount of enzyme required to catalyze the production of 1μmol of glucose per minute. The exogenous black mustard enzyme prepared had an activity of 0.984 U / gFW, an optimal pH of 6.5, and an optimal temperature of 40℃. It maintained high activity below 40℃, but its activity was rapidly lost above 70℃.

[0009] Further, S2 specifically involves: accurately weighing 0.5g of broccoli powder and placing it in a beaker, adding 2mL of 0.1mol / L, pH 6.5 phosphate buffer and a quantitative amount of exogenous myrosinase solution, performing enzymatic hydrolysis in a water bath, adding 75% ethanol solution at a material-to-liquid ratio of 1:20, shaking and extracting for 30min, centrifuging at 5000r / min for 10min, and passing the supernatant through a 0.45μm filter membrane to obtain crude isothiocyanate extract.

[0010] Furthermore, S3 specifically refers to: Accurately weigh isothiocyanate standard, dissolve it in anhydrous ethanol to prepare a 1 mg / mL stock solution, and serially dilute to obtain a series of standard working solutions of 5, 10, 20, 40, and 80 μg / mL; measure the absorbance at a wavelength of 249 nm, plot the standard curve, and obtain the linear regression equation y = 0.0097x + 0.2178, with a coefficient of determination R² = 0.9991; The formula for calculating the corrected absorbance is: , The formula for calculating the extraction yield of isothiocyanates is: , In the formula, C is the concentration of the crude isothiocyanate extract in μg / mL; V is the volume of the isothiocyanate filtrate in mL; and m is the mass of the broccoli powder in g.

[0011] Furthermore, the factor levels for the single-factor experiment in S4 were set as follows: the material-to-liquid ratio was set to 1:8, 1:12, 1:16, 1:20, 1:24, and 1:28 g / mL; the amount of exogenous enzyme added was set to 2, 4, 6, 8, 10, and 12 mL; the enzymatic hydrolysis temperature was set to 20, 30, 40, 50, 60, and 70℃; and the enzymatic hydrolysis time was set to 0.5, 1, 1.5, 2, 2.5, and 3 h. The optimal material-to-liquid ratio was determined to be 1:20 g / mL, and the amount of exogenous enzyme added, the enzymatic hydrolysis temperature, and the enzymatic hydrolysis time were selected as the three core factors for subsequent optimization.

[0012] Furthermore, S5 specifically includes: The factors and levels for the Box-Behnken experiment were set as follows: exogenous enzyme addition amount A, 6 mL at level -1, 8 mL at level 0, and 10 mL at level 1; enzymatic hydrolysis temperature B, 30℃ at level -1, 40℃ at level 0, and 50℃ at level 1; enzymatic hydrolysis time C, 1.5 h at level -1, 2 h at level 0, and 2.5 h at level 1; a total of 17 experiments were designed, with isothiocyanate extraction amount Y as the response value; The fitting yielded a quadratic regression equation: Y= 11.432 + 2.037A + 0.178B + 7.172C 0.004AB 0.123AC + 0.026BC 0.100A² 0.002B² 1.743 C² The model's coefficient of determination R² is 0.9758, the adjusted coefficient of determination adjR² is 0.9446, and the Adeq Precision is 16.9486.

[0013] Furthermore, S6 specifically includes: The constructed BP neural network is a three-layer feedforward network structure. The input layer contains three neurons, corresponding to the amount of exogenous enzyme added, the enzymatic hydrolysis temperature, and the enzymatic hydrolysis time, respectively. The output layer contains one neuron, corresponding to the amount of isothiocyanate extracted. Through trial and error screening, the number of hidden layer nodes was determined to be 10, and the network structure is 3-10-1. The network training adopted the Levenberg-Marquardt algorithm, and the Sigmoid function was selected as the activation function. The 17 sets of experimental data were divided into training set, validation set and test set.

[0014] Furthermore, the parameters of the genetic algorithm in S7 are set as follows: population size 30~100, crossover probability 0.6~0.9, mutation probability 0.05~0.2, maximum number of iterations 100~500, with the optimization objective being to maximize the extraction amount of isothiocyanate.

[0015] Furthermore, the process verification in S8 specifically involves conducting three parallel extraction experiments using both the response surface model optimal process and the neural network-genetic algorithm optimal process. The measured extraction yield of the response surface model optimal process was 7.73±0.12 mg / g, with a relative error of 3.65%. The measured extraction yield of the neural network-genetic algorithm optimal process was 7.84±0.11 mg / g, with a relative error of 2.49%.

[0016] Furthermore, the in vitro antioxidant activity assay in S8 includes the determination of DPPH free radical scavenging capacity and the determination of total antioxidant capacity: Determination of DPPH free radical scavenging capacity: Isothiocyanate solutions with mass concentrations of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6 mg / mL were prepared, with vitamin C of the same concentration as a positive control; a 0.04 g / L DPPH anhydrous ethanol stock solution was prepared; the reaction system contained 1 mL of sample solution, 1 mL of DPPH stock solution, and 1 mL of anhydrous ethanol. After reacting at a constant temperature in the dark for 30 min, the absorbance was measured at 517 nm, and the free radical scavenging rate was calculated. Total antioxidant capacity determination: The total antioxidant capacity of isothiocyanate solutions of different mass concentrations was determined using the FRAP method kit. The total antioxidant capacity increased with increasing concentration, reaching 0.76 mmol / L at a mass concentration of 0.6 mg / mL.

[0017] Compared to existing technologies, this invention conducted single-factor experiments to identify key variables affecting ITC extraction. Subsequently, a Box-Behnken design was employed to model and optimize key parameters including the material-to-liquid ratio, black mustard enzyme dosage, hydrolysis time, and hydrolysis temperature. Simultaneously, an ANN model was constructed and trained based on the same experimental data to compare the predictive accuracy and optimization performance of the RSM. Finally, the extraction process of ITCs was validated, and its antioxidant properties were evaluated. This invention provides an integrated modeling method for optimizing the extraction process of ITCs from broccoli, offering theoretical support for the development of green, natural antioxidant feed additives. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of a neural network model structure according to the present invention.

[0020] Figure 2 This is a schematic diagram of an ITCs standard curve according to the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the effect of the material-liquid ratio on the extraction yield of ITCs according to the present invention.

[0022] Figure 4 This is a schematic diagram illustrating the effect of the amount of exogenous enzyme added on the extraction yield of ITCs according to the present invention.

[0023] Figure 5 This is a schematic diagram illustrating the effect of enzymatic hydrolysis temperature on the extraction yield of ITCs according to the present invention.

[0024] Figure 6 This is a schematic diagram illustrating the effect of enzymatic hydrolysis time on the extraction yield of ITCs according to the present invention.

[0025] Figure 7 This is a schematic diagram of the response surface and contour lines of the interaction between the amount of exogenous enzyme added, the enzymatic hydrolysis time and the enzymatic hydrolysis temperature according to the present invention, wherein A and B represent the amount of exogenous enzyme added and the enzymatic hydrolysis temperature; C and D represent the amount of exogenous enzyme added and the enzymatic hydrolysis time; and E and F represent the enzymatic hydrolysis temperature and the enzymatic hydrolysis time.

[0026] Figure 8 This is a schematic diagram illustrating the training, verification, testing, and BP-ANN simulation output of fitted values ​​according to the present invention.

[0027] Figure 9 This is a schematic diagram of the DPPH free radical scavenging rate of an ITCs and VC solution according to the present invention.

[0028] Figure 10 This is a schematic diagram illustrating the antioxidant capacity of an ITCs and VC solution according to the present invention. Detailed Implementation

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] Broccoli and white mustard seeds were purchased from the Licheng District Agricultural Market in Jinan City; potassium myrosine glucuronide was purchased from Shanghai Maclean Biotechnology Co., Ltd.; DPPH (2,2-biphenyl-1-picrylhydrazyl) was purchased from Shanghai Maclean Biotechnology Co., Ltd.; the total antioxidant capacity assay kit (FRAP method) was purchased from Nanjing Jiancheng Bioengineering Institute; the glucose assay kit (GOD / POD colorimetric method) was purchased from Shanghai Beyotime Biotechnology Co., Ltd.; ITCs standards (purity ≥98%) were purchased from Shanghai Maclean Biotechnology Co., Ltd.; disodium hydrogen phosphate dodecahydrate and anhydrous sodium dihydrogen phosphate were purchased from Sinopharm Chemical Reagent Co., Ltd.; anhydrous ethanol was purchased from Tianjin Fuyu Fine Chemical Co., Ltd.; L(+)-ascorbic acid was purchased from Sinopharm Chemical Reagent Co., Ltd.; all reagents were of analytical grade.

[0031] UV-8000S UV-Vis spectrophotometer, Shanghai Yuanxi Instrument Co., Ltd.; 3K15-10797 high-speed centrifuge, SIGMA Centrifuge (Yangzhou) Co., Ltd.; HBS-1096A enzyme-linked immunosorbent assay (ELISA) analyzer, Nanjing Detie Experimental Equipment Co., Ltd.; DHG-9140 forced-air drying oven, Shanghai Yiheng Scientific Instrument Co., Ltd.; DK-8D electric thermostatic water bath, Shanghai Jinghong Experimental Equipment Co., Ltd.; CTFD-10T freeze dryer, Qingdao Yonghe Chuangxin Electronic Technology Co., Ltd.

[0032] After freeze-drying broccoli, the broccoli was pulverized and sieved. 0.5 g of broccoli powder was accurately weighed and placed in a beaker. 2 mL of 0.1 mol / L, pH 6.5 phosphate buffer and 6 mL of exogenous myrosinase solution were added, and the mixture was incubated in a 40°C water bath for 1 h. Then, 75% ethanol solution at a specific material-to-liquid ratio (g / mL) was added, and the mixture was extracted by shaking for 30 min. After centrifugation at 5000 rpm for 10 min, the supernatant was filtered through a 0.45 μm filter membrane to obtain the crude ITCs extract.

[0033] Take fresh mustard seeds, crush and sieve them, and accurately weigh 5g into a beaker. Then, add 250mL of pre-cooled 0.1mol / L phosphate buffer (pH 6.5) and sonicate for 15min in an ice bath. After treatment, centrifuge at 5000r / min for 10min and collect the supernatant. This crude enzyme solution should be prepared and used immediately, or stored for a short period at -20℃.

[0034] The black mustard enzyme activity was determined according to the method of CHEN

[12] . 1 g of frozen mustard seed powder was added to 6 mL of pre-cooled phosphate buffer (0.1 mol / L, pH 6.5), and after homogenization, it was centrifuged at 12000 r / min for 15 min at 4 °C. The supernatant was taken as the crude enzyme solution. 250 μL of the crude enzyme solution was mixed with an equal volume of 2 mmol / L potassium black mustard glucosinolate solution, and reacted in a water bath at 37 °C for 15 min. The reaction was then terminated by heating in a boiling water bath for 5 min. The glucose content was determined using a glucose detection kit, and the absorbance was measured at 510 nm.

[0035] Enzyme activity unit (U) is defined as the amount of enzyme required to catalyze the production of 1 μmol of glucose per minute. The exogenous black mustard enzyme activity was measured to be 0.984 U / gFW.

[0036] The basic enzymatic properties of the exogenous myrosinase (EC 3.2.3.1) used in this invention are as follows: Optimal pH: The optimal pH for the exogenous enzyme used in this study is 6.5.

[0037] Optimal temperature: The optimal temperature for the exogenous enzyme used in this study is 40℃.

[0038] Thermal stability: Black mustard enzyme maintains high activity below 40℃, but loses activity rapidly above 70℃. In this invention, the optimized enzymatic hydrolysis temperature range is 30~50℃, which is within the enzyme's relatively stable temperature range.

[0039] Accurately weigh 10 mg of ITCs standard sample and dissolve it thoroughly in 10 mL of anhydrous ethanol to prepare a stock solution with a mass concentration of 1 mg / mL. After shaking well, the stock solution was serially diluted to prepare a series of standard working solutions with mass concentrations of 5, 10, 20, 40, and 80 μg / mL. The absorbance was measured at a wavelength of 249 nm using a UV spectrophotometer. A standard curve was plotted with ITCs concentration on the x-axis and the corresponding absorbance value on the y-axis, and the linear regression equation was calculated.

[0040] To eliminate interference from impurities, a three-wavelength correction method at 239 nm, 249 nm, and 259 nm was used to calculate the corrected absorbance. This wavelength combination is a classic wavelength system for the quantitative detection of ITCs, which can effectively improve the specificity and accuracy of detection.

[0041] Corrected absorbance calculation formula: (1), The formula for calculating the extraction yield of isothiocyanates is: (2), (2) In the formula, C is the concentration of the crude ITCs extract in μg / mL; V is the volume of the ITCs filtrate in mL; and m is the mass of the broccoli powder in g.

[0042] Single-factor experiments were conducted to investigate the effects of four factors on the extraction yield of ITCs: material-to-liquid ratio (1:8, 1:12, 1:16, 1:20, 1:24, 1:28), amount of exogenous enzyme added (2, 4, 6, 8, 10, 12 mL), enzymatic hydrolysis temperature (20, 30, 40, 50, 60, 70℃), and enzymatic hydrolysis time (0.5, 1, 1.5, 2, 2.5, 3 h).

[0043] Based on the results of the single-factor experiments, three significant influencing factors—the amount of exogenous enzyme added (A), the enzymatic hydrolysis temperature (B), and the enzymatic hydrolysis time (C)—were selected for a Box-Behnken response surface methodology experiment, with the amount of ITCs extracted (Y) as the response value. The levels of each factor are shown in Table 1.

[0044] Table 1. Factors and Levels in Response Surface Experiment

[0045] Based on Box-Behnken experimental data, a backpropagation (BP) artificial neural network was constructed using the Neural Network Toolbox in MATLAB R2021a. Preliminary experiments comparing the mean squared error (MSE) of 4–15 hidden layer nodes determined the optimal number of hidden layer nodes to be 10. A 3-10-1 three-layer feedforward BP neural network was constructed: the input layer contained the amount of exogenous enzyme added, the hydrolysis temperature, and the hydrolysis time; the output layer contained the amount of ITCs extracted.

[0046] The Levenberg-Marquardt (LM) algorithm was used for network training, and the Sigmoid function was selected as the activation function. The 17 sets of experimental data were divided into training set, validation set and test set according to the proportions of 70%, 15% and 15%, respectively. The minimum MSE and the maximum correlation coefficient (R) were used as evaluation indicators to determine the optimal ANN model.

[0047] The trained ANN model is combined with GA for global optimization: the population size is set to 50, the crossover probability is 0.8, the mutation probability is 0.1, and the maximum number of iterations is 200. The optimal process parameters are searched with the goal of maximizing the extraction of ITCs.

[0048] To determine antioxidant activity, accurately weigh 4.00 mg of DPPH powder, dissolve it in anhydrous ethanol, and dilute to a final volume of 100 mL in a brown volumetric flask to prepare a DPPH stock solution with a concentration of 0.04 g / L (approximately 0.101 mmol / L). Store the stock solution in a sealed container at low temperature and protected from light for later use. Dissolve ITCs samples in ultrapure water to prepare a series of solutions with mass concentrations of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6 mg / mL. Prepare VC solutions with the same concentration gradient, using vitamin C (VC) as a positive control. For the sample group, add 1.0 mL of sample solution to 1.0 mL of DPPH stock solution and 1.0 mL of anhydrous ethanol; for the blank group, add 1.0 mL of sample solution to 1.0 mL of anhydrous ethanol and 1.0 mL of ultrapure water; for the control group, add 1.0 mL of ultrapure water to 1.0 mL of DPPH stock solution and 1.0 mL of anhydrous ethanol.

[0049] All reaction systems were incubated at a constant temperature in the dark for 30 min, and the absorbance of each tube was measured at a wavelength of 517 nm. Three replicates were performed for each concentration, and the DPPH radical scavenging rate was calculated using the following formula: Clearance rate (%) = , In the formula, The absorbance of the sample. This represents the absorbance of the blank sample. The absorbance is the control absorbance. Using vitamin C as a positive control, solutions with ITCs concentrations of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6 mg / mL were prepared. Three replicates were performed for each concentration. The procedures were strictly followed according to the kit instructions from Nanjing Jiancheng Biotechnology Institute. The T-AOC activity of the samples and vitamin C solutions was measured to evaluate their in vitro antioxidant activity.

[0050] All experiments were performed in triplicate, and results are expressed as mean ± standard deviation. One-way ANOVA was performed using SPSS 27 software, with a significance level set at P < 0.05. Response surface methodology and data analysis were performed using Design-Expert 11 software, ANN-GA analysis was performed using MATLAB R2021a software, and graphs were generated using GraphPad Prism 8 software.

[0051] Depend on Figure 2 We can see that, through linear regression, we obtain the linear regression equation: y = 0.0097x + 0.2178, with a coefficient of determination R0. 2 =0.9991 indicates that the standard curve has excellent linearity in the range of 5~80μg / mL, and the quantitative method is accurate and reliable, which can meet the needs of subsequent sample content detection.

[0052] Depend on Figure 3 It was found that as the solid-liquid ratio increased from 1:8 (g / mL) to 1:20 (g / mL), the extraction yield gradually increased, reaching a peak at 1:20–1:24 (g / mL), indicating the best extraction effect. When the solid-liquid ratio continued to increase to 1:28 (g / mL), the excessive solvent volume in the system may lead to a decrease in the concentration of the active ingredient and increase the solute diffusion distance, thereby reducing mass transfer efficiency and ultimately causing a slight decrease in the extraction yield. Considering both cost and extraction effect, the optimal solid-liquid ratio was determined to be 1:20 (g / mL).

[0053] Depend on Figure 4 It was found that within the range of 2–8 mL of exogenous enzyme addition, the increased content of myrosinase accelerated the hydrolysis rate of glucosinolates, and the amount of ITCs generated continuously increased, reaching a maximum at 8 mL. When the enzyme addition exceeded 8 mL, the substrate was rapidly consumed, and the high concentration of enzyme altered the system microenvironment, inducing side reactions and reducing the amount of ITCs extracted. Therefore, the optimal amount of exogenous enzyme addition is 8 mL.

[0054] Depend on Figure 5 It can be seen that when the temperature increases from 20℃ to 40℃, the activity of enzyme molecules increases with increasing temperature, the hydrolysis reaction rate accelerates, and the extraction yield of ITCs continues to rise; 40℃ is the suitable temperature for enzymatic reaction, where the extraction yield reaches its peak; above 40℃, black mustard enzyme protein denatures and becomes inactive, and since ITCs are heat-sensitive substances, they are easily decomposed and volatilized at high temperatures, resulting in a gradual decrease in extraction yield. The optimal enzymatic hydrolysis temperature is 40℃.

[0055] Depend on Figure 6It can be seen that when the enzymatic hydrolysis time is from 0.5 h to 2.0 h, the substrate is continuously hydrolyzed, ITCs are continuously generated and dissolved, and the extraction yield gradually increases; after the reaction time exceeds 2.0 h, the product is in the reaction system for a long time and undergoes degradation and oxidation, and the extraction yield slowly decreases. The optimal enzymatic hydrolysis time is 2.0 h.

[0056] Analysis of variance showed that the model's F-value was 31.34 (P < 0.0001), indicating that the overall model differences were highly significant. The model's coefficient of determination (R²) was 0.9758, and the adjusted coefficient of determination (adj R²) was 0.9446, indicating that the model has good fitting and predictive abilities. Furthermore, the AdeqPrecision was 16.9486, significantly higher than the critical value of 4, further confirming the model's excellent signal-to-noise ratio and high reliability.

[0057] Based on the results of response surface methodology (RSM), the importance of each factor on the extraction yield of ITCs was ranked as follows: hydrolysis time (C) > amount of exogenous enzyme added (A) > hydrolysis temperature (B). Regarding interactions: the AC and BC interactions were significant (P < 0.05), while the AB interaction was not significant (P > 0.05).

[0058] By fitting the model, a quadratic regression equation with the extracted amount (Y) as the response value is obtained: Y= 11.432 + 2.037A + 0.178B + 7.172C 0.004AB 0.123AC + 0.026BC 0.100A² 0.002B² 1.743 C² In the formula, Y is the amount of ITCs extracted (mg / g), A is the amount of exogenous enzyme added (mL), B is the enzymatic hydrolysis temperature (°C), and C is the enzymatic hydrolysis time (h). Analysis of the model regression equation revealed the optimal extraction process for ITCs as follows: exogenous enzyme addition of 8.138 mL, enzymatic hydrolysis temperature of 41.809℃, and enzymatic hydrolysis time of 2.078 h. Under these conditions, the predicted extraction yield of ITCs was 8.023 mg / g. To improve the feasibility of the experiment and considering practical and economic factors, the process was adjusted to: exogenous enzyme addition of 8.1 mL, enzymatic hydrolysis temperature of 42℃, and enzymatic hydrolysis time of 2.1 h.

[0059] Table 2 Box-Behnken Experimental Design and Results

[0060] Table 3. Results of ANOVA for the regression model

[0061] Note: * indicates a statistically significant difference (P < 0.05); ** indicates a highly statistically significant difference (P < 0.01); NS indicates no significant difference (P > 0.05).

[0062] Depend on Figure 7 The response surfaces of AC and BC are steep, and the contour lines are elliptical, indicating that the interaction between AC and BC on the extraction of ITCs is significant (P<0.05). The response surface of AB is gentler, indicating that the interaction term of AB on the extraction of ITCs is not significant (P>0.05), which is consistent with the results of the analysis of variance.

[0063] After constructing a BP-ANN model based on RSM experimental data, the network's fitting performance was evaluated. Figure 8 It can be seen that the correlation coefficients of the BP-ANN model training set, validation set, test set, and all samples are 0.99619, 0.98353, 0.98223, and 0.98166, respectively, all close to 1, indicating excellent model fitting performance and accurate reflection of the nonlinear relationship between process parameters and extraction yield. Through global optimization using GA, the theoretically optimal parameters are: exogenous enzyme addition of 7.7756 mL, enzymatic hydrolysis temperature of 40.0178℃, and enzymatic hydrolysis time of 2.1640 h, with a predicted extraction yield of 8.0399 mg / g. Combined with practical production rounding parameters: exogenous enzyme addition of 7.8 mL, enzymatic hydrolysis temperature of 40℃, and enzymatic hydrolysis time of 2.2 h.

[0064] This invention employs a trial-and-error method to determine the optimal number of hidden layer nodes in a BP-ANN. With a fixed input layer (3 neurons, corresponding to the amount of exogenous enzyme added, enzyme digestion temperature, and enzyme digestion time) and output layer (1 neuron, corresponding to the amount of ITCs extracted), preliminary experiments were conducted with the number of hidden layer nodes ranging from 7 to 12. For each node setting, training was repeated 10 times to eliminate biases introduced by random initialization. The mean squared error (MSE) of the validation set was used as the primary evaluation metric, while the coefficient of determination (R²) was also considered. Results showed that when the number of hidden layer nodes was 10, the validation set MSE was the lowest, and the MSE of the test set was close to that of the training set, indicating no significant overfitting. When the number of nodes continued to increase (≥11), the validation set MSE actually increased, indicating that the network began to exhibit overfitting. Therefore, a 3-10-1 network structure was determined to be the optimal network structure. Table 4 Comparison of BP neural network training performance under different numbers of hidden layer nodes

[0065] RSM model: coefficient of determination R² = 0.9758, adjusted R² = 0.9446, root mean square error RMSE = √(mean square of residuals) = √(0.0564 / 7) = 0.0898 mg / g.

[0066] ANN Model: The BP-ANN model (3-10-1 structure) was constructed based on the same experimental data. The correlation coefficients R of the training set, validation set, test set and all samples were 0.99619, 0.98353, 0.98223 and 0.98166, respectively; the coefficient of determination R² of all samples was 0.98166²≈0.9637, and the root mean square error RMSE was 0.0819 mg / g.

[0067] Comparative conclusions: The R² of the ANN model (0.9637) is higher than that of the RSM model (0.9758, adjusted to 0.9446), and the RMSE (0.0819 mg / g) is lower than that of the RSM model (0.0898 mg / g), indicating that the ANN model is superior to the RSM model in both fitting accuracy and predictive ability.

[0068] To verify the reliability of the RSM and ANN-GA optimization processes, three parallel experiments were conducted under the optimal extraction conditions determined by both methods. The predictive ability of the models was evaluated by comparing the relative errors between the predicted and measured values. As shown in Table 5, the measured and predicted values ​​of both the RSM and ANN-GA models were quite close, with relative errors of 3.65% and 2.49%, respectively, both at low levels, indicating that the established models have good prediction accuracy and reliability. Compared with the single RSM model, the actual extraction amount of the ANN-GA model is closer to the predicted value, indicating that the model has good predictive ability. This demonstrates that ANN-GA has better optimization ability and higher prediction accuracy in multi-factor optimization processes.

[0069] Table 5. Validation experiments of RSM and RSM-ANN-GA for optimal extraction conditions.

[0070] Depend on Figure 9 It was found that within the concentration range of 0.1–0.6 mg / mL, the DPPH free radical scavenging rate of ITCs continuously increased with increasing concentration, reaching as high as 92.08% at a concentration of 0.6 mg / mL. At the same concentration, the scavenging ability of ITCs was slightly lower than that of the positive control VC, but the trend was consistent, proving that ITCs possess certain free radical scavenging activity.

[0071] Depend on Figure 10 It can be seen that the T-AOC of ITCs increases with increasing mass concentration. At 0.6 mg / mL, the T-AOC of ITCs reaches 0.76 mmol / L, which is still lower than that of VC solution, but still shows a good concentration-dependent antioxidant effect.

[0072] This invention has the following advantages over other algorithms: (1) Complementarity with RSM: As a classic statistical method, RSM can quantitatively describe the main effects and interaction effects of each factor through a quadratic regression equation, and the model has strong interpretability. However, RSM has inherent limitations in capturing complex nonlinear relationships. ANN does not require a pre-defined function form and can approximate any nonlinear mapping. Using the two together can achieve the complementary advantages of "RSM analyzing trends + ANN accurately fitting".

[0073] (2) Comparison with PSO-SVR: Although Particle Swarm Optimization for Support Vector Regression (PSO-SVR) has good generalization ability, SVR is sensitive to kernel function selection, has complex parameter tuning, and is easily affected by initial parameter values ​​under small sample conditions. In contrast, BP-ANN combined with the Levenberg-Marquardt algorithm has a fast training convergence speed and is not sensitive to initial parameters.

[0074] (3) Synergistic advantages with GA-ANN: Genetic algorithms (GA) have strong global search capabilities, do not rely on gradient information, and are suitable for multi-objective optimization. Combining GA with ANN (ANN-GA), using the trained ANN model as the fitness function to evaluate candidate solutions, can effectively avoid getting trapped in local optima. Literature reports show that ANN-GA outperforms RSM in both prediction accuracy and actual extraction rate in extraction process optimization. For example, in cottonseed protein extraction research, the average percentage error (MPE) of the ANN model was 0.673%, far lower than the 2.56% of RSM.

[0075] (4) Empirical support for this study: The verification experiment shows that the relative error of the actual extraction amount of the ANN-GA model is 2.49%, which is lower than that of RSM (3.65%), directly proving the superiority of this scheme in the extraction of ITCs from broccoli.

[0076] This study used broccoli as raw material and employed exogenous black myrosinase to assist in the enzymatic hydrolysis and ethanol extraction of ITCs. The extraction process was systematically optimized through single-factor experiments, RSM, and RSM-ANN composite models. The results showed that various process conditions affected the ITC extraction yield to varying degrees, with enzymatic hydrolysis time being the key dominant factor. Compared to the single response surface methodology, the RSM-ANN model exhibited smaller prediction errors and superior process optimization performance.

[0077] The optimal extraction conditions were: a material-to-liquid ratio of 1:20, an exogenous enzyme addition of 7.8 mL, a hydrolysis temperature of 40℃, and a hydrolysis time of 2.2 h. Under these conditions, the actual extracted yield of ITCs was 7.84 ± 0.11 mg / g. Antioxidant activity assays showed that the obtained ITCs extract possessed strong free radical scavenging ability and a certain total antioxidant capacity, exhibiting good in vitro antioxidant activity, thus providing a theoretical basis for the development of natural antioxidant feed additives.

[0078] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing the extraction process of isothiocyanates from broccoli based on response surface methodology (RSM), characterized in that, Includes the following steps: S1. Broccoli powder was obtained by freeze-drying broccoli, followed by pulverizing and sieving. Exogenous black mustard enzyme crude solution was prepared using white mustard seeds as raw material, and the black mustard enzyme activity was determined. S2. Weigh broccoli powder, add phosphate buffer and exogenous black myrosinase solution, and enzymatically hydrolyze in a water bath at 30~50℃ for 1~3 hours. Then add 75% ethanol solution and shake to extract. After centrifugation, take the supernatant and filter it through a filter membrane to obtain crude isothiocyanate extract. S3. Prepare isothiocyanate standard solution and plot standard curve. Calculate the corrected absorbance using the three-wavelength correction method at 239nm, 249nm, and 259nm. Determine the isothiocyanate concentration in the crude extract based on the standard curve and calculate the isothiocyanate extraction amount. S4. Determine the optimal material-liquid ratio and screen out three core optimization factors: exogenous enzyme addition amount, enzymatic hydrolysis temperature, and enzymatic hydrolysis time. S5. A quadratic regression model was constructed using Box-Behnken design, with the following factors: exogenous enzyme addition amount 6~10mL, enzymatic hydrolysis temperature 30~50℃, and enzymatic hydrolysis time 1.5~2.5h. The main effects and interactions of each factor were analyzed to obtain the theoretically optimal process parameters predicted by the response surface model. S6. Based on the Box-Behnken response surface experimental data, a three-layer feedforward BP neural network with 3 neurons in the input layer, 7-12 neurons in the hidden layer, and 1 neuron in the output layer was constructed. The experimental data was divided into training set, validation set, and test set to complete the network training and obtain a neural network prediction model for isothiocyanate extraction amount. S7. With the goal of maximizing the extraction yield of isothiocyanate, the predicted value of the BP neural network is used as the fitness function of the genetic algorithm for global optimization to obtain the optimal extraction process parameters. S8. Conduct multiple batches of parallel verification tests on the optimal extraction process obtained through optimization, and simultaneously determine the in vitro antioxidant activity of the obtained isothiocyanate extract.

2. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, The preparation and enzyme activity determination of the exogenous black mustard enzyme solution in S1 are as follows: Take fresh white mustard seeds, crush and sieve them, weigh 5g of mustard seed powder and add it to 250mL of pre-cooled 0.1mol / L, pH6.5 phosphate buffer, sonicate for 15min under ice bath conditions, centrifuge at 5000r / min for 10min, and collect the supernatant to obtain exogenous black mustard enzyme solution. The method for determining the activity of black mustard enzyme is as follows: Take 1g of frozen mustard seed powder, add 6mL of pre-cooled 0.1mol / L, pH 6.5 phosphate buffer, grind into a homogenate, and centrifuge at 12000r / min for 15min at 4℃ to obtain crude enzyme solution; take 250μL of crude enzyme solution and mix with an equal volume of 2mmol / L potassium black mustard glucosinolate solution, react in a water bath at 37℃ for 15min, and terminate the reaction by heating in a boiling water bath for 5min; use a glucose assay kit to measure the absorbance at 510nm; the enzyme activity unit U is defined as the amount of enzyme required to catalyze the production of 1μmol of glucose per minute. The exogenous black mustard enzyme prepared had an activity of 0.984 U / gFW, an optimal pH of 6.5, and an optimal temperature of 40℃. It maintained high activity below 40℃, but its activity was rapidly lost above 70℃.

3. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, S2 specifically involves: accurately weighing 0.5g of broccoli powder and placing it in a beaker, adding 2mL of 0.1mol / L, pH 6.5 phosphate buffer and a quantitative amount of exogenous myrosinase solution, performing enzymatic hydrolysis in a water bath, adding 75% ethanol solution at a material-to-liquid ratio of 1:20, shaking and extracting for 30min, centrifuging at 5000r / min for 10min, and passing the supernatant through a 0.45μm filter membrane to obtain crude isothiocyanate extract.

4. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, Specifically, S3 is: Accurately weigh isothiocyanate standard, dissolve it in anhydrous ethanol to prepare a 1 mg / mL stock solution, and serially dilute to obtain a series of standard working solutions of 5, 10, 20, 40, and 80 μg / mL; measure the absorbance at a wavelength of 249 nm, plot the standard curve, and obtain the linear regression equation y = 0.0097x + 0.2178, with a coefficient of determination R² = 0.9991; The formula for calculating the corrected absorbance is: ， The formula for calculating the extraction yield of isothiocyanates is: ， In the formula, C is the concentration of the crude isothiocyanate extract in μg / mL; V is the volume of the isothiocyanate filtrate in mL; and m is the mass of the broccoli powder in g.

5. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, The factor levels for the single-factor experiment in S4 were set as follows: the material-to-liquid ratio was 1:8, 1:12, 1:16, 1:20, 1:24, and 1:28 g / mL; the amount of exogenous enzyme added was 2, 4, 6, 8, 10, and 12 mL; the enzymatic hydrolysis temperature was 20, 30, 40, 50, 60, and 70℃; and the enzymatic hydrolysis time was 0.5, 1, 1.5, 2, 2.5, and 3 h. The optimal material-to-liquid ratio was determined to be 1:20 g / mL, and the amount of exogenous enzyme added, the enzymatic hydrolysis temperature, and the enzymatic hydrolysis time were selected as the three core factors for subsequent optimization.

6. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, Specifically, S5 is: The factors and levels for the Box-Behnken experiment were set as follows: exogenous enzyme addition amount A, 6 mL at level -1, 8 mL at level 0, and 10 mL at level 1; enzymatic hydrolysis temperature B, 30℃ at level -1, 40℃ at level 0, and 50℃ at level 1; enzymatic hydrolysis time C, 1.5 h at level -1, 2 h at level 0, and 2.5 h at level 1; a total of 17 experiments were designed, with isothiocyanate extraction amount Y as the response value; The fitting yielded a quadratic regression equation: Y= 11.432+2.037A+0.178B+7.172C 0.004AB 0.123AC+0.026BC 0.100A² 0.002B² 1.743C², The model's coefficient of determination R² is 0.9758, the adjusted coefficient of determination adjR² is 0.9446, and the Adeq Precision is 16.9486.

7. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, Specifically, S6 is: The constructed BP neural network is a three-layer feedforward network structure. The input layer contains three neurons, corresponding to the amount of exogenous enzyme added, the enzymatic hydrolysis temperature, and the enzymatic hydrolysis time, respectively. The output layer contains one neuron, corresponding to the amount of isothiocyanate extracted. Through trial and error screening, the number of hidden layer nodes was determined to be 10, and the network structure is 3-10-1. The network training adopted the Levenberg-Marquardt algorithm, and the Sigmoid function was selected as the activation function. The 17 sets of experimental data were divided into training set, validation set and test set.

8. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, The parameters of the genetic algorithm in S7 are set as follows: population size 30~100, crossover probability 0.6~0.9, mutation probability 0.05~0.2, maximum number of iterations 100~500, with the optimization objective being to maximize the extraction amount of isothiocyanate.

9. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, The process verification in S8 specifically involves conducting three parallel extraction experiments using both the response surface model optimal process and the neural network-genetic algorithm optimal process. The measured extraction yield of the response surface model optimal process was 7.73±0.12 mg / g, with a relative error of 3.65%. The measured extraction yield of the neural network-genetic algorithm optimal process was 7.84±0.11 mg / g, with a relative error of 2.49%.

10. The method for optimizing the extraction process of broccoli isothiocyanate based on response surface methodology according to claim 1, characterized in that, The in vitro antioxidant activity assay in S8 includes DPPH free radical scavenging capacity assay and total antioxidant capacity assay: Determination of DPPH free radical scavenging capacity: Isothiocyanate solutions with mass concentrations of 0.1, 0.2, 0.3, 0.4, 0.5, and 0.6 mg / mL were prepared, with vitamin C of the same concentration as a positive control; a 0.04 g / L DPPH anhydrous ethanol stock solution was prepared; the reaction system contained 1 mL of sample solution, 1 mL of DPPH stock solution, and 1 mL of anhydrous ethanol. After reacting at a constant temperature in the dark for 30 min, the absorbance was measured at 517 nm, and the free radical scavenging rate was calculated. Total antioxidant capacity determination: The total antioxidant capacity of isothiocyanate solutions of different mass concentrations was determined using the FRAP method kit. The total antioxidant capacity increased with increasing concentration, reaching 0.76 mmol / L at a mass concentration of 0.6 mg / mL.

Images