Extraction, process optimization and application of momordica grosvenori polysaccharide

Abstract

The invention discloses extraction, process optimization and application of momordica grosvenori polysaccharide, and belongs to the technical field of natural product extraction. Aiming at the problems of large prediction deviation and insufficient optimization robustness caused by dependence on a single model in the prior art, two prediction models of a response surface method and a neural network are established in parallel, and key process parameters are respectively used as variables to predict the total polysaccharide content, so that two groups of optimization process parameters are obtained; through experimental verification and calculation of a relative error between a predicted value and an actual value, a process parameter with a higher actual measurement value in an error range is finally selected as a final scheme. The method can significantly improve the extraction yield and process reliability of the momordica grosvenori polysaccharide, and is suitable for industrial extraction and application in antioxidant products.

Description

Technical Field

[0001] This invention belongs to the field of natural product extraction technology, specifically relating to the extraction, process optimization, and application of monk fruit polysaccharide. Background Technology

[0002] In the field of natural product active ingredient extraction technology, optimizing the extraction process is a key step in improving yield and ensuring product quality stability. As a bioactive component, the extraction of monk fruit polysaccharide typically involves complex interactions of multiple process parameters, such as solvent composition, solid-liquid ratio, temperature, and time. Traditional process optimization methods largely rely on single-step modeling optimization strategies, such as response surface methodology. This method, by establishing a mathematical model between process parameters and response values, can effectively predict optimal process conditions within a certain range.

[0003] However, in practical applications, it has been found that optimization results obtained based on a single mathematical model sometimes deviate from subsequent actual verification experiments. This deviation may stem from limitations of the model itself, such as the limited fitting ability of response surface methodology to highly nonlinear relationships, or the experimental design failing to fully cover the actual optimal region. When the optimal conditions predicted by the model are at the boundary of actual production feasibility, this deviation may lead to poor stability and reproducibility of the optimized process. Although researchers can screen for better conditions by increasing the number of experimental verifications, this approach significantly increases time and material costs, reducing the efficiency of process optimization.

[0004] On the other hand, machine learning methods such as artificial neural networks have shown potential in handling complex nonlinear systems, and are expected to discover optimization regions that are difficult to find using traditional methods. However, their predictive performance is heavily dependent on the quality and quantity of training data. Under conditions of small sample data, the model is prone to overfitting or insufficient generalization ability. This means that relying solely on the "optimal solution" given by a single neural network model may also be unstable. Therefore, how to effectively evaluate the reliability of the output results of different optimization models, and how to select efficient, stable, and easily industrially implementable combinations of process parameters from the various potential optimization schemes given by the models, has become a practical problem in improving the robustness of process optimization. Existing methods lack a systematic verification and comparison mechanism to avoid the decision risks that may be brought about by a single model, thus making it difficult to guarantee that the optimization results have both high yield and high reliability. Summary of the Invention

[0005] One object of the present invention is to solve at least the above-mentioned problems and to provide at least the advantages that will be described later.

[0006] Another objective of this invention is to provide a method for optimizing the extraction process of monk fruit polysaccharides based on model validation and comparison. This method can establish and experimentally verify two prediction models with different principles in parallel, and select the process parameters with higher measured yield and smaller prediction error from the optimization schemes given by the models. This improves the polysaccharide yield while enhancing the reliability and robustness of the optimization results.

[0007] To achieve these objectives and other advantages of the present invention, a method for optimizing the extraction process of mogroside polysaccharides based on model validation comparison is provided, comprising the following steps: S1: Mix L-proline, the hydrogen bond acceptor, and citric acid, the hydrogen bond donor, in a molar ratio of 1:1 to 1:3, and add water to make the water content reach 20% to 40%. Stir at 70°C to 90°C to form a transparent liquid and obtain a eutectic solvent. S2: Mix the monk fruit powder with the eutectic solvent at a liquid-to-solid ratio of 1:20 to 1:40, and extract under ultrasonic conditions of 40°C to 80°C and ultrasonic time of 30 to 150 minutes. After filtration, alcohol precipitation, centrifugation, washing, reconstitution, and freeze-drying, monk fruit polysaccharide is obtained. The total polysaccharide content of the monk fruit polysaccharide is determined by the phenol-sulfuric acid method. S3: Using the molar ratio and water content in step S1 and the liquid-to-solid ratio in step S2 as variables, and the total polysaccharide content measured in step S2 as the response value, a first prediction model is established using the response surface methodology, and the first set of optimized process parameters and the corresponding first predicted total polysaccharide content are calculated through this model. S4: Using the molar ratio and water content in step S1, and the liquid-to-material ratio, ultrasonic temperature, and ultrasonic time in step S2 as input variables, and the total polysaccharide content measured in step S2 as the output variable, a second prediction model is established using a multilayer perceptron neural network. The second set of optimized process parameters and the corresponding second predicted total polysaccharide content are then calculated using this model. S5: Steps S1 and S2 are repeated using the first set of optimized process parameters and the second set of optimized process parameters, respectively, to obtain the first actual total polysaccharide content and the second actual total polysaccharide content. Calculate the first relative error between the first predicted total polysaccharide content and the first actual total polysaccharide content, and calculate the second relative error between the second predicted total polysaccharide content and the second actual total polysaccharide content; If both the first relative error and the second relative error are within the error range, the optimized process parameters with the highest measured total polysaccharide content are selected as the final extraction process for monk fruit polysaccharides.

[0008] Preferably, the hydrogen bond acceptor is selected from L-proline, betaine, and choline chloride, and the hydrogen bond donor is selected from citric acid, malic acid, oxalic acid, ethylene glycol, urea, glycerol, 1,3-butanediol, and lactic acid.

[0009] Preferably, the specific steps for establishing the first prediction model using the response surface methodology in step S3 are as follows: S3.1: Using the molar ratio A, water content B, and liquid-to-solid ratio C as three factors, each factor is set to three levels: low, medium, and high. The values ​​corresponding to the low, medium, and high levels of each factor are as follows: molar ratio 1:1, 1:2, 1:3; water content 20%, 30%, 40%; liquid-to-solid ratio 1:20, 1:30, 1:40. Based on the Box-Behnken design principle, an experimental scheme containing 12 factorial points and 5 center points, for a total of 17 experimental points, is generated. S3.2: Following the experimental plan in step S3.1, repeat steps S1 and S2 to obtain 17 different combinations of process parameters and their corresponding first actual total polysaccharide content; S3.3: Using the molar ratio A, water content B, and liquid-to-solid ratio C as independent variables, and the first actual total polysaccharide content as the dependent variable Y, a multiple quadratic regression method is used to fit the experimental data, establishing a second-order polynomial regression model containing linear terms, quadratic terms, and pairwise interaction terms of the independent variables. Its form is as follows: Y = β0 + β1A + β2B + β3C + β 12 AB+β 13 AC +β 23 BC +β 11 A 2 + β 22 B 2 + β 33 C 2 , Where β is the regression coefficient of each term in the model; S3.4: Perform an analysis of variance on the second-order polynomial regression model established in step S3.3, and calculate the F-value, P-value, and coefficient of determination R. 2 With the adjusted coefficient of determination R 2 When the p-value is less than 0.01 and the p-value of the lack-of-fit term is greater than 0.05, the model is considered significant and well-fitted, and the second-order polynomial regression model is confirmed as the effective first prediction model. S3.5: Based on the confirmed effective first prediction model, the numerical optimization algorithm is used to optimize within the range of the values ​​of the factors, with the goal of maximizing the total polysaccharide content, to solve for the first set of optimized process parameters and the corresponding first predicted total polysaccharide content.

[0010] Preferably, in step S3.4, when the determination coefficient R 2 When the value is greater than 0.98, the model is considered to be well-fitted.

[0011] Preferably, the specific steps for establishing the second prediction model using a multilayer perceptron neural network in step S4 are as follows: S4.1: Collect multiple sets of process parameters and their corresponding measured values ​​of total polysaccharide content obtained by performing steps S1 and S2 to form the original dataset; the process parameters include molar ratio, water content, liquid-to-solid ratio, ultrasonic temperature and ultrasonic time; S4.2: Normalize the process parameter data of each dimension in the original dataset, and linearly transform its values ​​to the [0,1] interval; randomly divide the preprocessed dataset into training set and test set according to a preset ratio; S4.3: Construct a multilayer perceptron neural network model, the structure of which includes an input layer, at least one hidden layer and an output layer; the input layer is configured with five neurons, corresponding to the five process parameters respectively; the output layer is configured with one neuron, corresponding to the predicted value of total polysaccharide content; S4.4: The constructed multilayer perceptron neural network model is trained using the training set. During the training process, the backpropagation algorithm is used in combination with the set learning rate and momentum factor to iteratively optimize the network weights and biases in order to minimize the mean square error between the network output value and the actual total polysaccharide content. S4.5: Use the test set to evaluate the performance of the trained multilayer perceptron neural network model, and calculate the correlation coefficient R² between the model prediction and the measured value; when R² is greater than or equal to 0.84, the model prediction performance is deemed acceptable, and the trained network model is confirmed as an effective second prediction model. S4.6: Discretize and combine the five process parameters within their respective value ranges to generate several simulated process parameter combinations; input the simulated process parameter combinations into the confirmed effective second prediction model for batch prediction, and select the simulated process parameter combinations with the highest predicted total polysaccharide content as the second set of optimized process parameters and the corresponding second predicted total polysaccharide content.

[0012] Preferably, in step S4.3, the hidden layer is configured as two layers, with the first hidden layer containing 7 neurons and the second hidden layer containing 2 neurons.

[0013] Preferably, in step S4.4, the learning rate is set to 0.2 to 0.3, and the momentum factor is set to 0.1 to 0.2.

[0014] Preferably, in step S4.2, 70% to 80% of the data is divided into a training set and 20% to 30% of the data is divided into a test set.

[0015] Preferably, the specific method for generating several combinations of simulated process parameters in S4.6 is as follows: The ultrasound time was fixed at 60 minutes and the ultrasound temperature was fixed at 70℃. The molar ratio is set to range from 1:1 to 1:3, with a variation step of 0.1. The moisture content is set to range from 20% to 40%, with a variation step of 1%. The set value range for the material-liquid ratio is 1:20 to 1:40, with a variation step of 1. Using a text Cartesian product calculator, the three parameters—molar ratio, water content, and feed-liquid ratio—are combined in steps within their set ranges to generate a set of simulated process parameters containing all possible combinations. The generated set of simulated process parameters is saved as a CSV file for input into the second prediction model for batch prediction.

[0016] A method for ultrasound-assisted DES extraction of mogro fruit polysaccharides includes the following steps: S1: Mix L-proline, the hydrogen bond acceptor, and citric acid, the hydrogen bond donor, in a molar ratio of 1:1 to 1:3, and add water to make the water content reach 20% to 40%. Stir at 70°C to 90°C to form a transparent liquid and obtain a eutectic solvent. S2: Mix the monk fruit powder with the eutectic solvent at a liquid-to-solid ratio of 1:20 to 1:40, and extract under ultrasonic conditions of 40°C to 80°C and ultrasonic time of 30 to 150 minutes. After filtration, alcohol precipitation, centrifugation, washing, reconstitution and freeze-drying, monk fruit polysaccharide is obtained.

[0017] Application of monk fruit polysaccharide prepared by optimizing the extraction process based on model validation and comparison or by ultrasound-assisted DES extraction in antioxidant products.

[0018] The present invention has at least the following beneficial effects: First, this invention establishes two prediction models based on different principles, namely Response Surface Methodology (RSM) and Multilayer Perceptron Neural Network (MLP-ANN), in parallel, and conducts experimental verification and comparison of their respective optimization results. This effectively avoids the decision-making risks that may arise from relying on a single model, thereby significantly improving the reliability and robustness of the process optimization results.

[0019] Secondly, this invention utilizes the advantages of the MLP-ANN model in handling complex nonlinear relationships to discover better combinations of process parameters that may be overlooked by traditional response surface methodology. This allows for higher actual polysaccharide yields based on verification, thereby improving the effectiveness of process optimization.

[0020] Third, this invention provides a systematic screening mechanism that compares the error between the model's predicted value and the actual verified value, and prioritizes the selection of process parameters with higher measured yields, making the optimization process more objective and data-driven, reducing reliance on experience, and providing a more reliable basis for the industrial scale-up of the process.

[0021] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0022] Figure 1 A comparison chart of the total polysaccharide content of monk fruit polysaccharides extracted with 16 different eutectic solvents (DES); Figure 2 The correlation analysis diagram shows the relationship between pH value, viscosity, conductivity and total polysaccharide content of 16 different DES. Figure 3 A comparison of Fourier transform infrared (FT-IR) spectra of L-proline (L-Pro), citric acid (CA), and the eutectic solvent (D1) formed by the two; Figure 4 This is a diagram illustrating the molecular interaction mechanism of L-proline (L-Pro) and citric acid (CA) forming a eutectic solvent (DES) based on density functional theory (DFT) simulations; where: Figure 4 -A is a schematic diagram of the stable geometry and hydrogen bond network of L-proline and citric acid after forming a eutectic solvent (D1); Figure 4 -B is the quantitative distribution diagram of the electrostatic potential (ESP) of L-proline; Figure 4 Quantitative distribution of electrostatic potential (ESP) of -C citric acid; Figure 4 Quantitative distribution of electrostatic potential (ESP) of -D DES (D1); Figure 4 -E is the electrostatic potential equivalent surface diagram and its extreme points of L-proline, citric acid and eutectic solvent (D1); Figure 5 The reduced density gradient analysis diagram for the eutectic solvent (D1) formed by L-proline and citric acid; where: Figure 5 -A is the reduced density gradient isosurface plot of D1; Figure 5 -B is the corresponding RDG scatter plot; Figure 5 -C is a schematic diagram of local weak interactions in the structure of the D1 molecule; Figure 6 This is a comparative graph showing the effects of various process parameters on the total polysaccharide content of *Siraitia grosvenorii* polysaccharide in a single-factor experiment; where: Figure 6 -A is the molar ratio effect diagram; Figure 6 -B is the diagram showing the effect of water content; Figure 6-C is the graph showing the effect of the feed-to-liquid ratio; Figure 6 -D is the ultrasonic temperature effect diagram; Figure 6 -E is the ultrasound time effect diagram; Figure 7 The response surface methodology (RSM) was used to optimize the extraction process of monk fruit polysaccharides, and the interaction between the molar ratio, water content, and liquid-to-solid ratio was analyzed using 3D and 2D graphs. Figure 7 -a, 7-b are response surface plots and contour plots of the interaction between molar ratio and water content; Figure 7 -c, 7-d are the response surface plot and contour plot of the interaction between molar ratio and liquid-to-material ratio; Figure 7 -e, 7-f are the response surface plot and contour plot of the interaction between water content and liquid-to-solid ratio; Figure 8 A scatter plot comparing the predicted and actual values ​​of the total polysaccharide content of monk fruit polysaccharides, and a visualization of the network structure, based on a multilayer perceptron (MLP) neural network model built using Weka software. Figure 9 A comparison chart of the total polysaccharide content of monk fruit polysaccharides extracted by different extraction methods (hot water ultrasonic-assisted extraction for ethanol precipitation SGP-W, eutectic solvent ultrasonic-assisted extraction for ethanol precipitation SGP-DE, and eutectic solvent ultrasonic-assisted extraction for methanol precipitation SGP-DM). Figure 10 This is a concentration-dependent comparison of the inhibitory activities of *Siraitia grosvenorii* polysaccharides (SGP-W, SGP-DE, SGP-DM) and acarbose (AK, positive control) obtained by different extraction methods on α-amylase and α-glucosidase; among them, Figure 10 -A is a comparison chart of α-amylase inhibition rates; Figure 10 -B is a comparison chart of α-glucosidase inhibition rates; Figure 11 A concentration-dependent comparison of the ability of different extraction methods to scavenge DPPH and ABTS free radicals in *Siraitia grosvenorii* polysaccharides (SGP-W, SGP-DE, SGP-DM) and ascorbic acid (VC, positive control) obtained by different extraction methods; among them, Figure 11 -A is a comparison chart of DPPH free radical scavenging rates; Figure 11 -B is a comparison chart of ABTS free radical scavenging rates. Detailed Implementation

[0023] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0024] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0025] It should be noted that, unless otherwise specified, the experimental methods described in the following implementation plan are all conventional methods, and the reagents and materials described are all commercially available unless otherwise specified.

[0026] 1. Test Methods 1.1 Preparation and Screening of Eutectic Solvents Appropriate amounts of HBA (choline chloride, betaine, L-proline) and HBD (citric acid, oxalic acid, malic acid, lactic acid, glycerol, ethylene glycol, urea, 1,3-butanediol) were weighed and mixed to prepare a DES system with a water content of 30% (Table 1) according to a certain molar ratio. The mixture was continuously stirred on a magnetic stirrer at 80°C to form a stable transparent liquid, which was used as DES solution for later use and named D1~D16 respectively.

[0027] Table 1. Different DES synthesized from HBA and HBD 1.2 DES Synthesis Mechanism 1.2.1 Physicochemical Properties of DES DES was measured at room temperature using a pH meter, conductivity meter, and digital viscometer (rotor No. 1, rotation speed 60 r / min), with each group repeated three times. Changes in functional groups on the sample surface were measured by FTIR. The mechanism of DES synthesis and polysaccharide extraction was further discussed.

[0028] 1.2.2 Density Functional Theory Simulation (DFT) Single-crystal structures were drawn using ChemDraw 22.0.064-bit software, and structural optimization and calculations were performed using Gaussian 16 C01 software. The Becke three-parameter mixed exchange functional and the Lee-Yang-Parr correlated functional (B3LYP) were selected for molecular accuracy calculations. DFT-D3BJ was used to correct intermolecular dispersion forces neglected in DFT calculations, improving the calculation accuracy for weak interactions (such as hydrogen bonds and van der Waals interactions). To represent the water solvation effect, the mathematical function set describing atomic wavefunctions was used with the 6-311+G(d,p) basis set, and the solvation model density (SMD) was simulated to reflect the solvation effect. Electrostatic potential (ESP) and weak interaction (IRI) analyses were performed on all structures using the Multiwfn program, and all illustrations were visualized using VMD software.

[0029] 1.3 Extraction of Monk Fruit Polysaccharides Take 1.0 g of monk fruit powder into an Erlenmeyer flask, add a certain mass of DES solvent according to Table 1, and extract by sonication at a certain temperature. After extraction, filter with gauze and collect the supernatant. Slowly add 4 times the volume of anhydrous ethanol while stirring continuously, let stand in a refrigerator at 2~8℃ for 12 h, centrifuge at 6000 r / min for 10 min to collect the precipitate, wash the precipitate twice with anhydrous ethanol, centrifuge again at 6000 r / min for 10 min to collect the precipitate, add a small amount of pure water to reconstitute and mix well, then freeze dry to obtain monk fruit polysaccharide.

[0030] 1.4 Determination of total polysaccharide content Prepare 0.1 mg / mL glucose standard solutions by diluting 10 mg of glucose standard solution into a 100 mL volumetric flask and making up to volume. Take 0, 0.2, 0.4, 0.6, 0.8, and 1.0 mL of the glucose standard solution into test tubes, add distilled water to a final volume of 1 mL, then add 1 mL of 5% phenol solution and 5.0 mL of concentrated sulfuric acid. Shake well and allow to cool to room temperature. Use the corresponding reagents as blank controls. Measure the absorbance at 490 nm using a UV spectrophotometer. Plot a polysaccharide standard curve with glucose concentration (mg / mL, C) on the x-axis and absorbance (A) on the y-axis. The equation is A = 10.2931C - 0.00542, R... 2 =0.99933.

[0031] The sample solution was prepared according to the phenol-sulfuric acid method, and the total content of monk fruit polysaccharides was calculated according to the standard curve equation, expressed as an equivalent amount of glucose, as shown in the following formula.

[0032] Total polysaccharide content of monk fruit = (C×V×N) / M×100%; where: C is the mass concentration of the sample solution, mg / mL; V is the volume of the sample solution, mL; N is the dilution factor; M is the dry weight of monk fruit polysaccharides in the sample, g.

[0033] 1.5 Single-factor experiment Using the total polysaccharide content of *Siraitia grosvenorii* polysaccharide as the indicator, single-factor experiments were conducted on five factors: molar ratio, water content, liquid-to-solid ratio, ultrasonic temperature, and ultrasonic time. Fixed factor levels were set: molar ratio 1:2, water content 30%, liquid-to-solid ratio 1:30, ultrasonic temperature 70℃, and ultrasonic time 30 min. The effects of molar ratio (1:1, 1:2, 1:3, 1:4, 1:5), water content (20%, 30%, 40%, 50%, 60%), liquid-to-solid ratio (1:20, 1:30, 1:40, 1:50, 1:60), ultrasonic temperature (40℃, 50℃, 60℃, 70℃, 80℃), and ultrasonic time (30 min, 60 min, 90 min, 120 min, 150 min) on polysaccharide yield were investigated sequentially.

[0034] 1.6 Response Surface Methodology (RSM) Experiment Based on the results of the single-factor experiment, three factors were used as response factors: molar ratio (A), water content (B), and liquid-to-solid ratio (C), with total polysaccharide content as the response value. The Box-Behnken method in the response surface methodology was used to design a three-factor, three-level system, as shown in Table 2.

[0035] Table 2 Response Surface Factor Level Design 1.7 Establishment of Multilayer Perceptron (MLP) Prediction Model 1.7.1 Construction and Preprocessing of the Original Dataset Seventy-five sets of valid data obtained through single-factor experiments were collected. Each data point contained five process parameters (molar ratio, water content, liquid-to-material ratio, ultrasonic temperature, and ultrasonic time) and their corresponding measured values ​​of total polysaccharide content, forming the original dataset.

[0036] The process parameter data for each dimension of the original dataset are normalized using a minimum-maximum normalization method, which linearly transforms the values ​​to the [0, 1] interval to eliminate the influence of dimensions and accelerate model convergence. The normalization formula is as follows: ; The preprocessed dataset was randomly divided into training and test sets in a 70%:30% ratio. The training set was used for model training, and the test set was used for independent evaluation of model performance.

[0037] 1.7.2 Construction and Training of Neural Network Models We used Weka software for analysis. We selected the Multilayer Perceptron algorithm under the "Classify" interface.

[0038] Network structure parameter settings: In the algorithm parameter settings, specify hiddenLayers = "7, 2", which means constructing an MLP model with two hidden layers, with 7 and 2 neurons respectively. The number of neurons in the input layer is 5 (corresponding to the five process parameters), and the number of neurons in the output layer is 1 (corresponding to the predicted value of total polysaccharide content).

[0039] Learning parameter settings: set learning rate = 0.2, momentum factor = 0.1, and training time = 3000.

[0040] The constructed MLP model is trained using a pre-defined training set. During training, the software employs an error backpropagation algorithm to iteratively optimize network weights and biases, minimizing the mean squared error between the network output value and the actual total polysaccharide content.

[0041] 1.7.3 Model Performance Validation and Optimized Process Prediction The trained MLP model was evaluated using a reserved test set. Weka software automatically calculated the correlation coefficient R² between the model's predicted values ​​and the measured values. In this experiment, the model's R² on the test set was 0.8486, which is greater than the preset acceptable threshold of 0.84. Therefore, the model's predictive performance was deemed acceptable, and it was confirmed as a valid predictive model.

[0042] Simulated process parameter generation: To perform global optimization, the ultrasonic time was fixed at 60 minutes and the ultrasonic temperature at 70℃. The molar ratio was set to range from 1:1 to 1:3 (step 0.1), the water content to 20% to 40% (step 1%), and the feed-liquid ratio to 1:20 to 1:40 (step 1). A text Cartesian product calculator was used to generate 9261 simulated process parameters, which were then saved as a CSV file.

[0043] The generated set of simulated process parameters was input into a validated MLP model for batch prediction. The process parameter combination with the highest predicted total polysaccharide content was selected as the second set of optimized process parameters and its corresponding second predicted total polysaccharide content. The predicted values ​​were verified through experiments.

[0044] 1.8 Comparison of different extraction methods Ethanol precipitation was obtained by hot water ultrasonic-assisted extraction (SGP-W), eutectic solvent ultrasonic-assisted extraction (SGP-DE), and eutectic solvent ultrasonic-assisted extraction (DGP-DM), and the total polysaccharide content was compared. 1.0 g of monk fruit powder was weighed and placed in an Erlenmeyer flask, and extraction was performed under optimal conditions. The precipitate was reconstituted with pure water, freeze-dried, and then the monk fruit polysaccharide was obtained and stored in a sealed container.

[0045] 1.9 Physiological activity assay 1.9.1 Assay for α-amylase inhibitory activity Pretreatment: Different concentrations (0.5, 1, 2, 4, 6, 8, 10 mg / mL) of monk fruit polysaccharide were prepared using PBS (pH=6.9), and an α-amylase solution with an enzyme activity ratio of 2 mg / mL was prepared. 1 g of soluble starch was dissolved in 100 ml of boiling water to prepare a 1% soluble starch solution.

[0046] Methods: Take 100 μL of polysaccharide solutions of different concentrations, add 100 μL of α-amylase solution, incubate at 37℃ for 20 min, then add 200 μL of 1% soluble starch, mix well, react at 37℃ for 10 min, add 500 μL of DNS, incubate at boiling water for 5 min to terminate the reaction, cool to room temperature, add 10 mL of PBS solution to each tube, and measure the absorbance at 540 nm. Acarbose was used as a control, and each experiment was performed three times.

[0047] α-Amylase inhibition rate (%) = [1 - (Sample A - Background A) / Blank A] × 100%; Sample A: α-amylase + monk fruit polysaccharide solution + 1% soluble starch + DNS; Background A: Monk fruit polysaccharide solution + 1% soluble starch + DNS; Blank A: α-amylase + 1% soluble starch + DNS.

[0048] 1.9.2 α-Glucosidase Inhibitory Activity Pretreatment: Prepare 2.5 mmol / mL PNPG solution, 2 U / mL α-glucosidase and different concentrations (0.5, 1, 2, 4, 6, 8, 10 mg / mL) of monk fruit polysaccharide using PBS (pH=6.9, 0.1 mol / mL). Dissolve 10.6 g of NaCO3 in 100 mL of pure water to prepare a 1 mol / mL solution for later use.

[0049] Methods: Mix 500 μL of different concentrations of monk fruit polysaccharide solution with 500 μL of PBS solution, add 1000 μL of α-glucosidase solution, incubate at 37℃ for 30 min, add 1000 μL of PNPG solution, incubate at 37℃ for 30 min, cool, and then add 1 mL of NaCO3 solution to terminate the reaction. Measure the absorbance at 405 nm using a UV-Vis spectrophotometer. Acarbose was used as a control. Each experiment was performed three times.

[0050] α-glucosidase inhibition rate (%) = [1 - (A sample - A background) / A blank] × 100%.

[0051] Sample A: PBS + α-amylase + monk fruit polysaccharide solution + PNPG; Background A: PBS + monk fruit polysaccharide solution + PNPG; Blank: PBS + α-glucosidase + PNPG.

[0052] 1.9.3 Determination of DPPH free radical scavenging rate 0.2 mmol / L DPPH stock solution: Take 19.9 mg of DPPH and place it in a 250 ml brown volumetric flask. Dissolve it thoroughly in anhydrous ethanol and make up to volume. Store in the dark.

[0053] DPPH working solution: Dilute an appropriate amount of DPPH stock solution to achieve an absorbance of 0.70±0.05 at 517nm before use.

[0054] Methods: 2.0 mL of different concentrations of monk fruit polysaccharide solutions (0.5, 1, 2, 4, 6, 8, 10 mg / mL) were dissolved in 2.0 mL of DPPH (0.2 mmol / L). Ascorbic acid (VC) was used as a positive control. The solutions were shaken well at room temperature and stored in the dark for 30 min. The absorbance was then measured at 517 nm. Each sample was tested in triplicate.

[0055] DPPH free radical scavenging rate (%) = [1 - (A sample - A background) / A blank] × 100%; Sample A: DPPH + Luo Han Guo polysaccharide solution; Background A: Monk fruit polysaccharide solution + anhydrous ethanol; A. Blank: DPPH solution.

[0056] 1.9.4 Determination of ABTS free radical scavenging rate ABTS stock solution: Dissolve 38.4 mg of ABTS in 10 ml of pure water and 6.7 mg of potassium sulfate in 10 ml of pure water. Mix them in equal volumes at a ratio of 1:1 and react at room temperature in the dark for 12-16 h to generate ABTS+ free radicals.

[0057] ABTS working solution: Take an appropriate amount of ABTS stock solution and dilute it with PBS (pH 6.9) until the absorbance at 734 nm is 0.70 ± 0.05.

[0058] Methods: Mix 0.4 mL of polysaccharide solution (0.5, 1, 2, 4, 6, 8, 10 mg / mL) with 4.0 mL of ABTS working solution in a 5 mL centrifuge tube. Use ascorbic acid (VC) as a positive control. Incubate at room temperature in the dark for 30 min. Measure the absorbance of the sample at 734 nm. Perform three replicates for each sample.

[0059] ABTS free radical scavenging rate (%) = [1 - (A sample - A background) / A blank] × 100%; Sample A: ABTS + Luo Han Guo polysaccharide solution; Background: Monk fruit polysaccharide solution + PBS; Blank: ABTS solution.

[0060] 1.9.5 Data Analysis Response surface methodology was designed using Design Expert 13 software; a neural network prediction model was established and validated using Weka 3.8.6 software; SPSS software was used for statistical analysis of the experimental data; and Origin 2024 software was used for plotting. Parallel experimental data are expressed as mean ± standard deviation.

[0061] 2. Experimental Results 2.1 DES Screening Results The component structure of DES significantly affects its extraction performance, providing a diverse range of green solvent options for the efficient extraction of polysaccharides. According to... Figure 1 It was found that the L-proline & citric acid system (D1) had the highest total polysaccharide content (77.34±2.27%), while the choline chloride & citric acid system (D13) had the lowest content at only 27.56±1.82%. Overall, the polysaccharide extraction content of the eutectic solvent (DES) system using L-proline as the hydrogen bond donor (HBD) was generally higher than that of the DES system using choline chloride as the HBD. This difference may be due to the synergistic effect of the structural characteristics of the polysaccharides and the physicochemical properties of the DES (pH, viscosity, conductivity, etc.).

[0062] from Figure 2 As shown in Table 3, DES is generally acidic and has high conductivity. Its acid-base properties are related to viscosity; when the system viscosity is low, the pH tends to be neutral or alkaline, while increasing viscosity increases acidity. Under the influence of different hydrogen bond acceptors (HBAs), the total polysaccharide content of the DES system (D1–D5) constructed with L-proline as the HBA is significantly higher than that of the betaine or choline chloride system. By comparing the polysaccharide extraction effects of different HBA combinations (D2 and D6, D4 and D8, D15 and D16) under the same hydrogen bond donor (HBD) conditions, the results show that among the factors affecting the total polysaccharide content, viscosity has the most significant effect, followed by pH, while conductivity has a relatively smaller impact.

[0063] Table 3 2.2 DES Interaction Mechanism FI-IR plays a crucial role in the study of compound structure and interactions because the interactions (hydrogen bonds) between substances have a significant impact, and it is often used to verify the synthesis of substances. FI-IR was used to characterize the structures of L-Pro, CA, and D1 (L-Pro & CA), such as... Figure 3 As shown in the image. The L-Pro is at 3064cm. -1 The broad absorption peak at 1624 cm⁻¹ is the O-H stretching vibration peak. -1 The strong absorption peak at 3429 cm⁻¹ is the C=O stretching vibration peak. -1 The peak of the NH stretching vibration is located at 1300 cm⁻¹. -1 Up to 1500cm -1 (1452cm) -1 1409cm -1 1331cm -1 1289cm -1The absorption peaks within the range of ) reflect the vibration of the CN bond in the pyrrolidine ring. Meanwhile, the CA peak at 3298 cm⁻¹... -1 The broad absorption peak at 1708 cm⁻¹ is due to the OH stretching vibration. -1 The strong absorption peak is the C=O stretching vibration peak, 1432 cm⁻¹. -1 The peak of the 0-H bending vibration is located at 1385 cm⁻¹. -1 and 1242cm -1 The peak at 3500 cm⁻¹ corresponds to the CO stretching vibration. For the infrared spectrum of D1 (L-Pre&CA), apart from the change in the main groups forming hydrogen bonds, the other peaks remain essentially unchanged, with the peak at 3500 cm⁻¹ being the CO stretching vibration peak. -1 ~3000cm -1 The presence of broad absorption peaks within the range indicates that the -OH or -NH groups of L-pro and the -OH groups of CA participate in the formation of hydrogen bonds. The C=O groups of both L-pro and CA are affected by hydrogen bonding, causing changes in the positions of the absorption peaks. The shifts in the stretching and bending vibration peaks of the hydroxyl and amino groups indicate the formation of new intermolecular hydrogen bonds between L-pro and CA.

[0064] Furthermore, to explore the interaction between L-Pro and CA, a DFT model was constructed to analyze the formation process of DES. In the DFT calculation, structural optimization was performed to adjust the atomic coordinates to minimize the system's energy, thereby obtaining the stable ground-state structure. After geometric optimization of L-Pro and CA, the optimized structures were sorted from low to high energy, and the geometry with the lowest energy value was determined to be the most stable geometry for D1. Figure 4 As shown in -A, the dashed lines represent the intermolecular hydrogen bonds in D1. The H of -COOH in CA interacts with the O of -COOH in L-Pro, and the O of -OH in CA interacts with the H of -NH in L-Pro, thus establishing a hydrogen bond network and effectively extracting the polysaccharide components.

[0065] By comparing the changes in electrostatic potential (ESP) before and after adsorption, this study reveals the charge transfer and interaction mechanisms during the adsorption of L-Pro and CA. Using electron density as the fundamental variable, the study analyzes the positive and negative regions of the charge distribution at various points in space. The quantitative ESP distribution diagram shows that L-Pro is dominated by electron-deficient regions. Figure 4 -B), and CA ( Figure 4 The electron-rich region is enhanced after the -C) reaction, but the DES region is still dominated by the electron-deficient region. Figure 4 -D), indicating a potential tendency for electrostatic attraction with monk fruit polysaccharide molecules.

[0066] Figure 4-E visually displays the equivalent electrostatic potential surface of the molecule. Since the molecular structure is neutral, the electrostatic potential is colored red and blue, with blue representing regions with more negative electrostatic potential and red representing regions with more positive electrostatic potential. Small spheres on the equivalent surface represent extreme points within those regions. The CA molecule has a minimum of -55.98 kcal / mol and a maximum of 70.97 kcal / mol, while the L-Pro molecule has a minimum of -63.34 kcal / mol and a maximum of 58.58 kcal / mol. After the reaction of CA and L-Pro, the minimum of the generated DES becomes -47.40 kcal / mol, and the maximum becomes 72.27 kcal / mol. The charge within the surface molecules changes, possibly due to the redistribution or sharing of electrons. This also reflects the reorganization of the molecular structure, proving the formation of hydrogen bonds. L-Pro molecules exhibit a negative electrostatic potential when near nitrogen and oxygen atoms, making them more likely to undergo electrophilic reactions with hydrogen atoms in CA molecules to form DES. The binding energy is -6.797 kcal / mol. The more negative the binding energy, the more stable the complex structure and the easier it is to bind, thus resulting in better extraction of Luo Han Guo polysaccharides.

[0067] Reduced density gradient (RDG / IRI) is widely used in the study of weak interactions. It visualizes and quantifies interactions between atoms or molecules by using electron density and its gradient information. Regions with small RDG values ​​(close to 0) indicate a flat electron density distribution, corresponding to non-covalent interaction regions; regions with large RDG values ​​correspond to chemical bonds or vacuum regions with drastic changes in electron density. This can be achieved by plotting RDG isopleths (RDG values ​​are then used to map these interactions). Figure 5 -A) and RDG scatter plot ( Figure 5 -B) Quantifying the nature and strength of D1 interactions, a large negative sign(λ2)ρ value is displayed in blue, indicating strong absorption interactions, such as hydrogen bonding; a sign(λ2)ρ value close to zero is displayed in green, indicating weak non-bonded interactions, such as van der Waals forces or π-π stacking; a positive sign(λ2)ρ value is displayed in red, indicating repulsive effects, such as steric hindrance. The results show that there are blue isosurfaces between L-Pro carboxyl oxygen and CA carboxyl hydrogen, and between L-Pro amino hydrogen and CA hydroxy oxygen, which are hydrogen bonding interactions. The weak attraction between C=O and CH of L-Pro and CA carboxylic acid groups is displayed in green, which is essentially a van der Waals interaction, giving DES the characteristics of low density, gentle gradient, and wide distribution range. In addition, combined with FI-IR analysis, the change in chromatographic peak shape and the occurrence of shifts further confirm the formation of hydrogen bonds, thus forming DES with a certain viscosity. By reducing density gradient (RDG / IRI) analysis ( Figure 5-C): The blue isosurface appears between L-Pro(–COOH)O⋯H–(COOH)CA and L-Pro(–NH)H⋯O–(OH)CA, confirming strong hydrogen bonding; the green region corresponds to van der Waals interactions such as C=O⋯H–C. Combined with the FT-IR peak shifts, this further verifies hydrogen-bonded molecular assembly, endowing D1 with suitable viscosity and improving polysaccharide extraction performance.

[0068] 2.3 Results of Single-Factor Experiment After determining the DES system, the polysaccharide content first increased and then decreased with the increase of the CA ratio, from Figure 6 As shown in Figure A, the total polysaccharide content was highest (61.92 ± 2.26%) when the molar ratio of L-Pro to CA was 1:2. This is likely because CA is a tribasic acid, and the pH value of DES gradually decreases with increasing CA content. L-Pro cannot form hydrogen bonds with excess CA, leading to excessive acidity in DES, which damages the chemical structure and biological activity of the polysaccharides, resulting in reduced solubility, low extraction efficiency, and compromised functional properties. Therefore, a molar ratio of 1:2 was chosen.

[0069] Moisture content is a key factor in improving the viscosity of DES (distilled solvent). Excessively high viscosity hinders contact between the solvent and the raw material components, leading to a reduced diffusion rate and slower dissolution of polysaccharides from the raw material. This negatively impacts ultrasonication and mixing effects, ultimately decreasing polysaccharide extraction efficiency. Furthermore, excessively high water content competes with components in the DES for hydrogen bonds, disrupting the existing hydrogen bond network and affecting the stability of the DES and its polysaccharide extraction efficiency. Figure 6 -B When the DES moisture content is 30% and 40%, the total polysaccharide content is similar, but the total polysaccharide content with 40% moisture content has a larger error. Therefore, it can be seen that DES with 30% moisture content has a better extraction effect on monk fruit polysaccharides.

[0070] A suitable feed-to-liquid ratio allows the raw materials to fully expand, resulting in a loose cell structure and promoting polysaccharide diffusion. Figure 6 As indicated by -D, a solid-liquid ratio of 1:30 resulted in sufficient cell swelling and the highest total polysaccharide content. Increasing the solvent further would dilute the suspension, introduce impurities, and increase costs, thus reducing the economic viability of the process. Therefore, a solid-liquid ratio of 1:30 was determined.

[0071] Ultrasonic temperature significantly affects polysaccharide extraction by influencing factors such as cavitation, molecular diffusion, polysaccharide structural stability, and solvent solubility. Due to the limited temperature settings of ultrasonic instruments, the investigation range was set at 40–80℃. Figure 6As shown in Figure C, within the temperature range of 50–70℃, increasing temperature reduces solution viscosity, enhances the cavitation effect of ultrasound, and intensifies molecular thermal motion, promoting the migration of polysaccharides from the cell interior into the solvent, resulting in a gradual increase in total polysaccharide content. However, when the temperature exceeds the polysaccharide's thermal stability range (80℃), it may lead to polysaccharide degradation, oxidation, or glycosidic bond breakage, damaging its structure and consequently reducing extraction efficiency. Therefore, an ultrasound temperature of 70℃ was selected.

[0072] When ultrasound reaches its optimal time, the cell wall is fully ruptured, and polysaccharide release tends to reach equilibrium; insufficient time leads to incomplete extraction, while excessive time results in degradation of thermosensitive polysaccharides, a decrease in molecular weight and activity, and a decline in yield. Figure 6 As shown in Figure -E, the cell structure remains intact for a short period, and the polysaccharides have not been completely dissolved from the raw materials. When the time is 60 minutes, the cell wall breakdown and polysaccharide release reach equilibrium. As time continues, the polysaccharide structure is thermally degraded. Therefore, an ultrasound time of 60 minutes was chosen.

[0073] Plackett-Burman (PB) experiment: An experimental design based on the Plackett-Burman (PB) method was used to screen key influencing factors from 5 variables, aiming to further clarify the factors that have the most significant impact on polysaccharide yield.

[0074] Table 4 Levels of Factors Affecting PB Note: Each round of the experiment was repeated 3 times, with a total of 12 experimental groups (N=12, corresponding to a PB design with 5 factors). Table 5. Analysis of Variance in the Plackett-Burman Experiment Table 5 shows that the significant influencing factor (P<0.05) is the liquid-to-solid ratio (C); the insignificant factors are molar ratio (A), water content (B), ultrasonic time (E), and ultrasonic temperature (D). Therefore, A (molar ratio), B (water content), and C (liquid-to-solid ratio) are selected as the three key variables for subsequent response surface optimization.

[0075] 2.4 RSM Optimization Results Based on the single-factor experiments and PB experiments, the experimental design was carried out using the Box-Benhnken method of Design-Expert-13 software with the total polysaccharide content as the indicator, and the process parameters for ultrasonic-assisted DES extraction of Luo Han Guo polysaccharides were determined. The results are shown in Table 6.

[0076] Table 6 Results of RSM extraction of polysaccharides from Luo Han Guo (Monk Fruit) Design Expert software was used to perform a quadratic multiple regression fitting on the experimental data to obtain the multiple regression equations for total polysaccharide content on molar ratio (A), water content (B), and liquid-to-solid ratio (C): Y = 80.81 + 6.28A + 6.54B - 2.36C + 0.9725AB + 1.05AC - 2.31BC - 6.74A 2 -11.44B 2 -9.03C 2 .

[0077] Table 7 Results of Analysis of Variance Note: p < 0.01 indicates extremely significant effect; p < 0.05 indicates significant effect.

[0078] Analysis of variance (ANOVA) uses F-values ​​and P-values ​​to determine the overall significance of the model. A larger F-value indicates a more significant model, while a smaller P-value indicates a more significant impact of each factor and its interaction on the response variable. According to Table 7, the model's F-value of 42.92 (P < 0.0001) is highly significant, and the lack-of-fit term P-value of 0.1254 > 0.05 is insignificant, proving that the prediction model has not missed any important terms and the prediction results are effective and reliable. ANOVA shows that A, B, A2, B2, and C2 all have highly significant effects (P < 0.01), C has a significant effect (P < 0.05), and the interaction effects between AB, AC, and BC are not significant (P > 0.05). The F-values ​​indicate that the order of influence of the three factors on the polysaccharide is: water content > molar ratio > material-to-liquid ratio. R0.05 2 =0.9822, R 2 The adj=0.9593 and coefficient of variation (CV) of 3.29% indicate that the experimental data and the predicted values ​​have a good model fit, which can be used to predict SGP extraction by DES and provide a scientific basis for process optimization.

[0079] The interactions of various factors are statistically analyzed using visualization techniques. Response surface methodology (3D) visually illustrates the interactions between multiple variables; a steeper surface indicates a more significant impact of the variable on the response, while a flatter surface indicates a smaller impact. Contour plots (2D) simplify the data and highlight the main trends; circles or ellipses represent independent effects of variables, while irregular shapes indicate interactions. Small (large) contour line spacing indicates rapid (gradual) response changes. Figure 7 It is clear from the contour map that the moisture content (7-B) and the feed-to-liquid ratio (7-C) are different. Figure 7 -E) Deviates slightly from an ellipse. Figure 7 -F ratio Figure 7 -B、 Figure 7The steep D-curve indicates that the interaction between water content and material-liquid ratio has a more significant impact than the other interactions. However, the contour plots of the three factors (molar ratio, water content, and material-liquid ratio) are all relatively circular, and the three-dimensional response surface curves are relatively flat, indicating that their interactions have no significant impact on the polysaccharide extraction rate (P>0.05), which is consistent with the results of the table variance and significance analysis.

[0080] Analyzing the fitted equation and variance analysis results, the optimal extraction conditions, determined using Design Expert 13 software, were: an L-proline to citric acid molar ratio of 1:2.48, a water content of 32.66%, a solid-liquid ratio of 1:29.1, an ultrasonic temperature of 70℃, and an ultrasonic time of 60 min. For ease of practical operation, the extraction conditions were adjusted to an L-proline to citric acid molar ratio of 1:2.5, a water content of 33%, a solid-liquid ratio of 1:30, an ultrasonic temperature of 70℃, and an ultrasonic time of 60 min. Three parallel experiments were conducted under these conditions. The actual total polysaccharide content of *Siraitia grosvenorii* was 83.98 ± 1.36%, with a relatively small error (0.6%) compared to the model's predicted value of 83.48%. This indicates that the established regression model has a high degree of fit and can effectively predict the total polysaccharide content of *Siraitia grosvenorii* during the experimental process.

[0081] 2.5 MLP Prediction Results Preliminary Random Forest (RF) Prediction: The Random Forest algorithm, a decision tree-based ensemble algorithm in Weka, improves the accuracy and robustness of the model by constructing multiple decision trees and integrating their predictions. It makes no assumptions about data distribution, is suitable for mixed-type variables, and outputs the contribution of variables to the model, making it an important indicator of variable influence. The Random Forest correlation coefficient is 0.9013, which helps explain the association between target variables, indicating that the model's predictions are highly correlated.

[0082] Multilayer Perceptron (MLP) Prediction: The Multilayer Perceptron (MLP) is a type of feedforward artificial neural network. As a fundamental model of deep learning, it improves training algorithms, solves the gradient vanishing problem, and is used to adjust network parameters. Using default parameters leads to increased correlation coefficients and error values. Therefore, by adjusting the parameters, the optimal parameter conditions were found in this experiment: the number of hidden layers (neurons) was set to 7 and 2, the learning rate to 0.2, the momentum to 0.1, and the number of training iterations to 3000. The final prediction result showed a correlation coefficient R² of 0.8486, indicating that it has a certain predictive value for the data. The visualization of the MLP neural network structure can intuitively show the connection relationships between its layers and the direction of data flow. The model uses molar ratio (MEB), water content (HSL), feed-to-liquid ratio (LYB), ultrasonic temperature (CSWD), and ultrasonic time (CSSJ) as input layers to receive raw data. Two layers of neurons handle the nonlinear transformation of the data, generating the final predicted process parameters. The error per epoch is 0.0160797, reflecting a small prediction error on the training data in each epoch during deep learning training, indicating effective model learning and good fit. The results are then visualized. Figure 8 The X-axis represents the actual value, and the Y-axis represents the predicted value. The "×" scatter points in the figure represent the same values ​​as the predicted data. The smaller the value, the smaller the error. After preprocessing, the predicted values ​​of the prediction model and the actual values ​​basically show a linear distribution relationship. The prediction model has high accuracy and can be used to establish a prediction model for the process parameters of DES extraction of Luo Han Guo polysaccharides.

[0083] Analysis using Weka software determined the optimal processing conditions to be a molar ratio of 1:1.1, a water content of 40%, a solid-liquid ratio of 1:40, an ultrasonic temperature of 70℃, and an overtime of 60 min. The polysaccharide extracted from *Siraitia grosvenorii* at a molar ratio of 1:1 was unstable, easily becoming fluid and difficult to freeze-dry. Therefore, a molar ratio of 1:2 was used as the screening condition, with a water content of 40% and a solid-liquid ratio of 1:32, predicting a total polysaccharide content of 88.94%. For experimental convenience, the optimal actual conditions were set as a molar ratio of 1:2, a water content of 40%, a solid-liquid ratio of 1:30, an overtime of 60 min, and an ultrasonic temperature of 70℃. Three parallel experiments were conducted under these optimized conditions. The actual extracted total polysaccharide content from *Siraitia grosvenorii* was 87.21 ± 1.58%, with an error of 1.98% compared to the model prediction, which is within an acceptable range, indicating that the prediction results are accurate and the model was successfully established.

[0084] 2.6 Comparison of RSM and MLP Comparing the correlation coefficients of the RSM and MLP models, it was found that the R-value of RSM is higher. 2 R compared to MLP 2The R-value is closer to 1.00, but the total polysaccharide content extracted by the MLP method is 3.23% higher than that obtained by the RSM method. Furthermore, the actual values ​​of the total polysaccharide content from both methods are close in error, possibly because the reduced CA content, increased water content, and decreased viscosity are beneficial to the extraction efficiency of monk fruit polysaccharides. In addition, different models require different choices. MLP needs to consider the generalization effect; a large R-value does not necessarily mean that the predicted process parameters and results are reasonable. The prediction results are very intuitive, so the actual values ​​should be used for determination. As can be seen from Table 8, MLP can provide a methodological reference for the process parameters of DES extraction of monk fruit polysaccharides.

[0085] Table 8 Comparison of optimization results between RSM and MLP models 2.7 Comparison of different extraction methods The traditional extraction method for monk fruit polysaccharides involves water extraction and alcohol precipitation. Ethanol, methanol, and acetone can be used to precipitate the polysaccharides. Considering that acetone is a limiting reagent and is more toxic than methanol, this experiment only used ethanol and methanol to precipitate the polysaccharides. Under the same conditions, three methods were used for ultrasound-assisted extraction of monk fruit polysaccharides, and the results are as follows: Figure 9 As shown, the average total polysaccharide content of SGP-W was 65.39±39%, the average total polysaccharide content of SGP-DE (88.47±0.84%) was 1.35 times that of SGP-W, and the average total polysaccharide content of SGP-DM (92.92±1.56%) was 1.42 times that of SGP-W. SGP-DM showed the best extraction effect on monk fruit polysaccharides, followed by SGP-DE. Considering the low toxicity of methanol, in order to compare the differences among the three methods and screen out the most suitable monk fruit polysaccharides, the structural characterization and in vitro activity studies of the monk fruit polysaccharides extracted by the three methods were subsequently conducted.

[0086] 2.8 Bioactivity Analysis 2.8.1 In vitro hypoglycemic activity In vitro hypoglycemic studies of monk fruit polysaccharides mainly focus on exploring their effects on the activity of enzymes related to glucose metabolism and their potential mechanisms. α-Amylase and α-glucosidase are two common glucose metabolism enzymes. Using enzyme inhibition rate as an indicator, this study investigates whether monk fruit polysaccharides extracted by DES have hypoglycemic potential. Results are as follows... Figure 10As shown, when the concentration of SGP-DE exceeded 4 mg / mL, its inhibitory effect on α-amylase significantly increased, while the inhibitory effects of SGP-W and SGP-DM on α-amylase slightly decreased with increasing concentration. Conversely, the inhibitory effect on α-glucosidase showed a concentration-dependent effect for SGP-DE; from 4 mg / mL onwards, the increase in the α-glucosidase inhibition rate of SGP-DE and SGP-DM was significantly greater than that of SGP-W. This indicates that both SGP-DE and SGP-DM can inhibit enzyme activity, and their inhibitory effect is better than that of SGP-W. This also explains their smaller molecular weight and can provide a reference for in vivo blood glucose control.

[0087] 2.8.2 Antioxidant Activity Analysis Monk fruit polysaccharides reduce the accumulation of reactive oxygen species (ROS) and alleviate oxidative stress-induced cell damage by scavenging DPPH and ABTS free radicals. Structural characteristics such as molecular weight and monosaccharide composition are also key factors influencing the antioxidant activity of polysaccharides. Figure 11 As shown, the antioxidant activity of monk fruit polysaccharides is positively correlated with their concentration; with increasing concentration, the DPPH and ABTS free radical scavenging rates also significantly increase. SGP-W may be affected by its molecular weight and small amount of monk fruit polysaccharide pigments; large-molecule polysaccharides have difficulty crossing multiple cell membrane barriers to exert their biological activity in vivo, leading to a decrease in scavenging rate with increasing concentration. Comparing the free radical scavenging rates of SGP-DE and SGP-DM revealed that the former has a stronger free radical scavenging ability, presumably due to the synergistic effect of Ara and Glc, resulting in its better antioxidant effect.

[0088] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for optimizing the extraction process of momordica grosvenori polysaccharide based on model verification comparison, characterized in that, The method comprises the following steps: S1: mixing a hydrogen bond acceptor L-proline and a hydrogen bond donor citric acid according to a molar ratio of 1:1 to 1:3, adding water to reach a water content of 20% to 40%, and stirring to form a transparent liquid at 70 DEG C to 90 DEG C to obtain a deep eutectic solvent; S2: mixing monk fruit powder and the deep eutectic solvent according to a liquid-to-material ratio of 1:20 to 1:40, performing extraction under ultrasonic temperature of 40 DEG C to 80 DEG C and ultrasonic time of 30 minutes to 150 minutes, and obtaining monk fruit polysaccharide after filtering, alcohol precipitation, centrifugation, washing, redissolution and freeze-drying of the extraction liquid; and determining the total polysaccharide content of the monk fruit polysaccharide by a phenol-sulfuric acid method; S3: taking the molar ratio, water content in step S1 and the liquid-to-material ratio in step S2 as variables, taking the total polysaccharide content determined in step S2 as a response value, establishing a first prediction model by a response surface method, and obtaining a first group of optimized process parameters and a corresponding first predicted total polysaccharide content by the model; S4: taking the molar ratio, water content in step S1 and the liquid-to-material ratio, ultrasonic temperature and ultrasonic time in step S2 as input variables, taking the total polysaccharide content determined in step S2 as an output variable, establishing a second prediction model by a multilayer perception neural network, and obtaining a second group of optimized process parameters and a corresponding second predicted total polysaccharide content by the model; S5: respectively adopting the first group of optimized process parameters and the second group of optimized process parameters, repeatedly performing steps S1 and S2 to obtain a first actual total polysaccharide content and a second actual total polysaccharide content respectively; calculating a first relative error between the first predicted total polysaccharide content and the first actual total polysaccharide content, and calculating a second relative error between the second predicted total polysaccharide content and the second actual total polysaccharide content; if the first relative error and the second relative error are within an error range, selecting the optimized process parameters with a higher measured total polysaccharide content as the final monk fruit polysaccharide extraction process.

2. The method for optimizing the extraction process of mogroside based on model verification comparison according to claim 1, characterized in that, The hydrogen bond acceptor is selected from L-proline, betaine and choline chloride, and the hydrogen bond donor is selected from citric acid, malic acid, oxalic acid, ethylene glycol, urea, glycerol, 1,3-butanediol and lactic acid.

3. The method for optimizing the extraction process of mogroside based on model verification comparison according to claim 1, characterized in that, The specific steps of establishing the first prediction model by the response surface method in S3 are as follows: S3.1: taking the molar ratio A, water content B and liquid-to-material ratio C as three factors, setting low, medium and high levels for each factor, and corresponding values of low, medium and high levels of each factor are respectively 1:1, 1:2 and 1:3 for the molar ratio, 20%, 30% and 40% for the water content, and 1:20, 1:30 and 1:40 for the liquid-to-material ratio; generating an experimental scheme containing 12 factor points and 5 center points, a total of 17 experimental points based on the Box-Behnken design principle; S3.2: repeatedly performing steps S1 and S2 according to the experimental scheme in step S3.1 to obtain 17 different combinations of process parameters and corresponding first actual total polysaccharide contents; S3.3: Using the molar ratio A, water content B and liquid-to-material ratio C as independent variables, and the first actual total polysaccharide content as dependent variable Y, a second-order polynomial regression model containing first-order terms, second-order terms and two-way interaction terms of the independent variables is established by using a multivariate quadratic regression method, and the form is: Y = β0+ β1A + β2B + β3C + β 12 AB+β 13 AC +β 23 BC +β 11 A 2 + β 22 B 2 + β 33 C 2 , where β is the regression coefficient of each term in the model; S3.4: performing variance analysis on the second order polynomial regression model established in step S3.3 to calculate F value, P value, determination coefficient R 2 and adjusting determination coefficient R 2 adj ; when P value is less than 0.01 and P value of the misfit term is greater than 0.05, it is determined that the model is significant and well fitted, and at this time the second order polynomial regression model is confirmed as an effective first prediction model; S3.5: Based on the confirmed effective first prediction model, a numerical optimization algorithm is used to optimize the value range of the factors, with the maximum total polysaccharide content as the objective function, to obtain the first set of optimized process parameters and the corresponding first predicted total polysaccharide content.

4. The method according to claim 3, wherein, In step S3.4, when the decision coefficient R 2 is greater than 0.98, it is determined that the model is well fitted.

5. The method for optimizing the extraction process of mogroside based on model verification comparison according to claim 1, characterized in that, The specific steps of establishing the second prediction model by using a multilayer perception neural network in step S4 are as follows: S4.1: Collecting multiple sets of process parameters and their corresponding total polysaccharide content measured values obtained by executing steps S1 and S2 to form an original data set; the process parameters include molar ratio, water content, liquid-to-material ratio, ultrasonic temperature and ultrasonic time; S4.2: Normalizing each dimension of process parameter data in the original data set to linearly transform its value to the interval [0, 1]; dividing the preprocessed data set into a training set and a test set according to a predetermined proportion at random; S4.3: Constructing a multilayer perception neural network model, which includes an input layer, at least one hidden layer and an output layer; the input layer is provided with five neurons corresponding to the five process parameters; the output layer is provided with one neuron corresponding to the predicted value of the total polysaccharide content; S4.4: Training the constructed multilayer perception neural network model using the training set; During the training process, the error back propagation algorithm is used in combination with the set learning rate and momentum factor to optimize the network weight and bias through iteration to minimize the mean square error between the network output value and the actual total polysaccharide content; S4.5: Evaluating the performance of the trained multilayer perception neural network model using the test set, calculating the correlation coefficient R² between the model predicted value and the measured value; when R² is greater than or equal to 0.84, the model prediction performance is acceptable, and the trained network model is confirmed as an effective second prediction model; S4.6: Discretizing and combining the five process parameters within their respective value ranges to generate a number of simulated process parameter combinations; inputting the simulated process parameter combinations into the confirmed effective second prediction model for batch prediction, and selecting the simulated process parameter combination with the highest predicted total polysaccharide content as the second set of optimized process parameters and the corresponding second predicted total polysaccharide content.

6. The method for optimizing the extraction process of mogroside based on model verification comparison according to claim 5, characterized in that, In S4.3, the hidden layer is set to two layers, the first hidden layer contains 7 neurons, and the second hidden layer contains 2 neurons.

7. The method according to claim 5, wherein, In S4.4, the learning rate is set to 0.2 to 0.3, and the momentum factor is set to 0.1 to 0.

2.

8. The method for optimizing the extraction process of mogroside based on model verification comparison according to claim 5, characterized in that, In S4.6, the specific method for generating a number of simulated process parameter combinations is as follows: Fixing the ultrasonic time as 60 minutes and the ultrasonic temperature as 70°C; Setting the value range of the molar ratio as 1:1 to 1:3 with a step size of 0.1; The value range of the water content is set to 20% to 40%, with a step size of 1%; The value range of the liquid-solid ratio is set to 1:20 to 1:40, with a step size of 1; Using a text Cartesian product calculator, the mole ratio, water content, and liquid-solid ratio are combined in steps within their set ranges to generate a set of simulation process parameters containing all possible combinations; The generated set of simulation process parameters is saved as a CSV format file for input into the second prediction model for batch prediction.

9. A method for ultrasonic-assisted DES extraction of momordica polysaccharide, characterized in that, The method comprises the following steps: S1: Mix the hydrogen bond acceptor L-proline with the hydrogen bond donor citric acid according to a mole ratio of 1:1 to 1:3, and add water to achieve a water content of 20% to 40%, and stir to form a transparent liquid at 70°C to 90°C to obtain a deep eutectic solvent; S2: Mix the momordica grosvenori powder with the deep eutectic solvent according to a liquid-solid ratio of 1:20 to 1:40, and perform extraction under the conditions of an ultrasonic temperature of 40°C to 80°C and an ultrasonic time of 30 minutes to 150 minutes, and after the extraction liquid is filtered, alcohol precipitated, centrifuged, washed, redissolved, and freeze-dried, momordica grosvenori polysaccharide is obtained.

10. The application of momordica grosvenori polysaccharide prepared by the model verification comparison-based momordica grosvenori polysaccharide extraction process optimization method of any one of claims 1-8 or the method of claim 9 in an antioxidant product.

Images