Prediction method for hypoglycemic activity of mulberry leaf water extract based on BP neural network

Abstract

The invention belongs to the technical field of modernized analysis of traditional Chinese medicines, and provides a method for predicting the hypoglycemic activity of a mulberry leaf aqueous extract based on a BP neural network, and the method comprises the following steps: preparing mulberry leaf aqueous extracts under different extraction conditions, and measuring the contents of various active components and the alpha-glucosidase inhibition rate; screening out a key active component combination remarkably related to the inhibition ratio from the various components by utilizing Pearson correlation analysis; constructing and training a BP (Back Propagation) neural network prediction model by taking the content of the key component as input and the alpha-glucosidase inhibition ratio as output; preparing an independent verification set by adopting mulberry leaf raw materials from different sources, substituting the independent verification set into the trained model and generating a predicted value, and comparing the predicted value with an actual value to verify that the model has good prediction precision and universality. The method overcomes the defect that the prior art depends on tedious biological experiments to measure the blood sugar lowering activity of the mulberry leaves, realizes rapid and high-throughput prediction of the blood sugar lowering activity of the mulberry leaves, and brings convenience to accurate screening and quality control of mulberry leaf raw materials.

Description

Technical Field

[0001] This invention belongs to the field of modern analytical technology of traditional Chinese medicine, specifically relating to a method for predicting the hypoglycemic activity of mulberry leaf water extract based on BP neural network. Background Technology

[0002] Mulberry leaves, as an important medicinal and edible resource, exert their hypoglycemic effect through a complex system composed of various active substances (such as alkaloids, flavonoids, polysaccharides, and proteins). These active substances may exhibit synergistic or antagonistic effects, working together to target key sites such as α-glucosidase, thereby delaying carbohydrate absorption. However, this multi-component, multi-target characteristic makes the hypoglycemic mechanism of mulberry leaves extremely complex, and the hypoglycemic activity of mulberry leaf water extracts from different sources cannot be accurately predicted.

[0003] Currently, screening for the hypoglycemic activity of mulberry leaves requires in vitro α-glucosidase inhibition experiments. However, the results of this method are directly related to the test concentration. When faced with commercially available mulberry leaf extracts of unknown concentration, the directly measured inhibition rates are incomparable due to concentration differences. To achieve fair comparison, existing techniques require determining the half-maximal inhibitory concentration (IC50) for each sample. 50 This process requires a series of dilutions and repeated experiments. Furthermore, traditional single-component (such as DNJ) content determination cannot reflect the complex synergistic effects of multiple active substances. Therefore, there is an urgent need for a method that eliminates the need for cumbersome IC50 assays. 50 This method is a novel approach that directly and rapidly predicts and compares the activity of different samples by analyzing their chemical composition.

[0004] Rapid and accurate screening of mulberry leaf raw materials is crucial for ensuring the hypoglycemic efficacy of products. Existing technologies disclose the use of artificial neural networks to analyze HPLC fingerprint peaks of mulberry leaves to classify sample types. However, this model aims to classify mulberry leaf samples rather than directly predict the bioactivity of mulberry leaf water extracts linked to efficacy, especially hypoglycemic activity. Therefore, constructing an intelligent screening mechanism for key active ingredients based on statistical correlation analysis, specifically designed for predicting hypoglycemic activity, has significant practical implications. Summary of the Invention

[0005] The technical problem to be solved by this invention is to overcome the efficiency bottleneck of existing technologies that rely on cumbersome biological experiments to determine the hypoglycemic activity of mulberry leaves, and to provide a prediction method that can use rapid chemical component detection to replace time-consuming biological assays.

[0006] To solve the above problems, the technical solution adopted by the present invention is as follows: A method for predicting the hypoglycemic activity of mulberry leaf water extract based on a backpropagation neural network includes the following steps: S1, by treating mulberry leaf raw materials under different extraction conditions, a wide range of representative mulberry leaf water extract samples were prepared; S2, determine the content of multiple active ingredients and α-glucosidase inhibition rate in each mulberry leaf water extract sample; S3. Through statistical correlation analysis, the correlation strength between the content of each active ingredient and the α-glucosidase inhibition rate is quantified. Through quaternary spatial distribution law analysis, the consistency of the variation law of the content of each active ingredient and the α-glucosidase inhibition rate in the multi-dimensional process space is visualized. Key active ingredients that are significantly related to the α-glucosidase inhibition rate are screened from the multiple active ingredients. S4. Define the BP neural network structure, use the content of key active ingredients as input variables and the α-glucosidase inhibition rate as output variables, construct and train the BP neural network model, use statistical indicators to evaluate the predictive performance of the model, and finally obtain the blood sugar lowering activity prediction model of mulberry leaf water extract. S5 uses a validation sample set independent of the input set to externally validate the model's universality. Test samples are prepared, and the content of key active ingredients and α-glucosidase inhibition rate are measured. The data are then input into the mulberry leaf water extract hypoglycemic activity prediction model, and the prediction results are output.

[0007] A further improvement of the present invention is that, in S1, the material-to-liquid ratio is set to 1:40. w:v The stirring speed was set to 700 r / min, the heat treatment temperature T was continuously varied from 60 °C to 100 °C, and the extraction time t was continuously varied from 0.5 min to 120 min.

[0008] A further improvement of the present invention is that: in S2, the multiple active ingredients include at least: total flavonoids, total phenols, total sugars, DNJ and total protein.

[0009] A further improvement of the present invention is that, in S3, the correlation analysis is a Pearson correlation analysis, and the calculation method is as follows: ; In the formula, r xy The correlation coefficients between the content of each active ingredient and the α-glucosidase inhibition rate are given. x i This represents the original data value of the active substance content. x̄ This represents the average value among similar indicators of active substance content. y i This represents the raw data value of α-glucosidase inhibition rate. ȳ This represents the average value among similar indicators of α-glucosidase inhibition rate.

[0010] A further improvement of the present invention is as follows: In S3, a four-dimensional relationship diagram of "process parameters-component content-bioactivity" is drawn, in which extraction temperature is the X-axis and time is the Y-axis, forming a process parameter plane; the α-glucosidase inhibition rate is used as the Z-axis to characterize the level of bioactivity; the content of candidate active ingredients is mapped as a color gradient to visualize the consistency of the change law of each active ingredient content and α-glucosidase inhibition rate in the multi-dimensional process space, and active ingredients with high consistency are screened as key active ingredients.

[0011] A further improvement of the present invention is that: in S3, active ingredients with a correlation coefficient ≥ 0.7 with α-glucosidase inhibition rate are selected as the key active ingredients; the key active ingredients include total flavonoids, total phenols, total sugars and DNJ.

[0012] A further improvement of the present invention is that, in S3, the key active ingredients include total flavonoids, total phenols, total sugars, and DNJ.

[0013] The invention is further improved in that: in S4, the method for constructing the blood sugar lowering activity prediction model of mulberry leaf water extract includes: setting the number of layers of the BP neural network to 3, wherein the number of input layer nodes is 4, the number of output layer nodes is 1, and the hidden layer contains 10 neurons.

[0014] A further improvement of the present invention is that, in S4, the statistical indicator includes the coefficient of determination R. 2 The values ​​of root mean square error (RMSE) and mean absolute percentage error (MAPE) are calculated using the following formulas: ; In the formula, Y i,p These are the model's predicted values. Y i,e These are experimentally measured values. Y m is the average value of the experimental measurements, and n is the sample size.

[0015] A further improvement of the present invention is that, in S4, to alleviate the gradient offset problem and ensure symmetrical weight updates during gradient descent, the tansig function is used as the activation function for the hidden layer of the BP neural network. ; In the formula, x i This is the output value of the previous layer. y l This is the input value for the next layer; The output layer uses the purelin function as the activation function: y=purelin(y l)=y l ; In the formula, y l This is the output value of the previous layer. y This is the final output value.

[0016] A further improvement of the present invention is that the prediction model for the hypoglycemic activity of the mulberry leaf water extract is as follows: ; in, Y ANN b1 represents the model's predicted value, LW represents the output layer weights, IW represents the input layer weights, b1 represents the input layer bias, and b2 represents the output layer bias.

[0017] A further improvement of the present invention is that: in S5, mulberry leaf raw materials from different sources than those used in the model are selected, and extraction parameters are randomly selected within the range of extraction conditions for extraction.

[0018] Beneficial effects 1. This invention proposes a predictive method for the hypoglycemic activity of mulberry leaf aqueous extracts. By establishing a backpropagation neural network model of mulberry leaf active components and α-glucosidase inhibition rate, it successfully eliminates the direct dependence on this complex biological process and uses stable and precisely quantifiable physical data of mulberry leaf components to predict hypoglycemic activity. This invention successfully constructs a more accurate quantitative mapping relationship for predicting hypoglycemic activity based on mulberry leaf aqueous extract components, providing a more efficient and reliable technical means for rapid screening of mulberry leaf raw materials and stable control of product quality. 2. Before constructing the prediction model, this invention introduces a key active ingredient screening mechanism based on statistical correlation analysis. Based on the correlation between the physical quantities of five active ingredients (total flavonoids, total phenols, total sugars, DNJ, and total protein) and the α-glucosidase inhibition rate, four key active ingredients with high correlation coefficients are screened to ensure that the variables input into the neural network are the optimal combination of ingredients with the strongest intrinsic driving relationship with the target biological activity. Attached Figure Description

[0019] Figure 1 This is a flowchart of the steps in an embodiment of the present invention to predict the hypoglycemic activity of mulberry leaf water extract based on a BP neural network. Figure 2 This is a Pearson correlation heatmap of the main active components of mulberry leaf water extract and α-glucosidase inhibition rate in the embodiments of the present invention; Figure 3 This is a four-dimensional spatial distribution diagram showing the relationship between the total flavonoid content and α-glucosidase inhibition rate as a function of extraction temperature and time in the embodiments of the present invention. Figure 4This is a four-dimensional spatial distribution diagram of the total phenol content and α-glucosidase inhibition rate as a function of extraction temperature and time in the embodiments of the present invention. Figure 5 This is a four-dimensional spatial distribution diagram of the total sugar content and α-glucosidase inhibition rate as a function of extraction temperature and time in the embodiments of the present invention; Figure 6 This is a four-dimensional spatial distribution diagram of DNJ content and α-glucosidase inhibition rate as a function of extraction temperature and time in the embodiments of the present invention. Figure 7 This is a four-dimensional spatial distribution diagram showing the relationship between total protein content and α-glucosidase inhibition rate as a function of extraction temperature and time in embodiments of the present invention. Detailed Implementation

[0020] refer to Figure 1 A method for predicting the hypoglycemic activity of mulberry leaf water extract based on BP neural network includes the following steps: S1, by treating mulberry leaf raw materials under different extraction conditions, a wide range of representative mulberry leaf water extract samples were prepared; S2, determine the content of multiple active ingredients and α-glucosidase inhibition rate of each mulberry leaf water extract sample; S3. Through correlation analysis, the correlation strength between the content of each active ingredient and the α-glucosidase inhibition rate is quantified, and key active ingredients that are significantly correlated with the α-glucosidase inhibition rate are screened from the various active ingredients. S4. Define the BP neural network structure, use the content of key active ingredients as input variables and the α-glucosidase inhibition rate as output variables, construct and train the BP neural network model, use statistical indicators to evaluate the predictive performance of the model, and finally obtain the blood sugar lowering activity prediction model of mulberry leaf water extract. S5 uses a validation sample set independent of the input set to externally validate the model's universality. Test samples are prepared, and the content of key active ingredients and α-glucosidase inhibition rate are measured. The data are input into the blood glucose-lowering activity prediction model of mulberry leaf water extract, the prediction results are output, and the results are compared with the actual measured values.

[0021] Experimental materials (1) Mulberry leaf sample A, produced in Jiaxing, Zhejiang.

[0022] (2) Mulberry leaf sample B, from Sichuan.

[0023] In this embodiment, in step S1, 15.0 g of dried mulberry leaf sample A was weighed and placed in a round-bottom flask. 600 mL of deionized water was added, and water bath extraction was performed with a stirring speed of 700 r / min. Five extraction temperature gradients were set in this embodiment: 60 ℃, 70 ℃, 80 ℃, 90 ℃, and 100 ℃. The total extraction time at each temperature was 120 min. In the initial extraction stage (0 to 10 min), intensive sampling was performed: the first and second samples were taken at 0.5 min and 1 min after the start of extraction, respectively, followed by sampling every 1 min from 2.0 min to 10 min. In the middle and later stages of extraction (after 10 min to 120 min), intermittent sampling was performed: starting from 10 min, samples were taken every 10 min until the extraction ended at 120 min.

[0024] In this embodiment, the contents of total flavonoids, total phenols, total sugars, DNJ and total protein and the α-glucosidase inhibition rate in the mulberry leaf aqueous extract in S2 are shown in Table 1.

[0025] Methods for determining the content of active ingredients include: The total flavonoid content was determined using the aluminum nitrate-sodium nitrite colorimetric method. The rutin standard curve was y = 1.2539x + 0.018, and R² = 0.9997. The total phenol content was determined using the Folin-Ciocalteu colorimetric method. The standard curve for gallic acid was y = 2.3391x + 0.0696, with R² = 0.9983. The total sugar content was determined using the phenol-sulfuric acid method. The glucose standard curve was y = 5.3447x + 0.0895, and the R² was 0.9963. The content of DNJ was determined by pre-column derivatization-high performance liquid chromatography using fluorenylmethoxychloride. The standard curve for 1-deoxynojirimycin was y = 4206x - 3.5688, with R² = 0.9999. The total protein content was determined using the Coomassie brilliant blue method. The standard curve for bovine serum albumin was y = 7.5517x + 0.2766, and the R² was 0.9934.

[0026] Methods for determining α-glucosidase inhibition rate: Add PBS buffer (pH 6.80 mol / L), mulberry leaf aqueous extract, and α-glucosidase aqueous solution (1.26 U / mL) sequentially to a 96-well plate. After shaking well, incubate at 37 ℃ for 10 min. Then add [the following to be added to the assay system]... pThe reaction was initiated with NPG (10 mmol / L) and incubated at 37 ℃ for 15 min. The reaction was terminated with Na₂CO₃ solution (1 mmol / L), and the absorbance was measured at 405 nm. The relative inhibition rate of the mulberry leaf water extract was calculated as follows: ; In the formula, ω is the α-glucosidase inhibition rate; A: absorbance of the experimental group; a: absorbance of the reaction system without α-glucosidase solution; B: absorbance of the reaction system without mulberry leaf water extract; b: absorbance of the blank group.

[0027] Table 1. Main active ingredients and α-glucosidase inhibition rate of mulberry leaves

[0028]

[0029]

[0030] In this embodiment, in S3, the correlation between total flavonoids, total phenols, total sugars, DNJ, and total protein in mulberry leaf aqueous extract and α-glucosidase inhibition rate is analyzed using the Pearson correlation analysis principle. The calculation method is as follows: ; In the formula, r xy The correlation coefficients between the content of each active ingredient and the α-glucosidase inhibition rate are given. x i This represents the original data value of the active substance content. x̄ This represents the average value among similar indicators of active substance content. y i This represents the raw data value of α-glucosidase inhibition rate. ȳ This represents the average value among similar indicators of α-glucosidase inhibition rate.

[0031] A correlation heatmap was generated using Prism software, and the results are as follows: Figure 2 As shown, the darker the color of the patch, the higher the correlation; the lighter the color, the weaker the correlation. Total flavonoids, total phenols, total sugars, and DNJ with α-glucosidase inhibition rate had correlation coefficients ≥ 0.7 and were selected as key active ingredients.

[0032] In this embodiment, in step S3, to further verify the optimality of the selected key active ingredient combination, this invention introduces a verification method based on multi-dimensional collaborative visualization analysis. A four-dimensional relationship diagram of "process parameters - ingredient content - bioactivity" is plotted, where extraction temperature is the X-axis, time is the Y-axis, forming a process parameter plane; the α-glucosidase inhibition rate is used as the Z-axis to characterize the level of bioactivity; and the content of candidate active ingredients is mapped as a color gradient. The results are as follows... Figure 3-6 As shown, the four-dimensional plots of total flavonoids, total phenols, total sugars, and DNJ exhibit highly regular and synergistic changes. Regions with high inhibition rates (high points on the Z-axis) almost completely overlap with regions of high component content (yellow areas) on the XY process plane. The release patterns of these four components are highly synchronized and stable with the enhancement patterns of hypoglycemic activity. This synchronicity indicates that they are not merely simple statistical correlators, but rather core functional substances directly involved in the expression of activity. Figure 7 As shown, the color change of total protein content is not very obvious, indicating that its content is not greatly affected by the process, and there is no visible and systematic correspondence between it and the distribution of inhibition rate. This excludes the possibility of total protein as a key active ingredient from a visualization perspective.

[0033] In this embodiment, in S4, the number of layers in the BP neural network is set to 3, with 4 nodes in the input layer, representing the contents of total flavonoids, total phenols, total sugars, and DNJ; 1 node in the output layer, representing the α-glucosidase inhibition rate; and 10 neurons in the hidden layer.

[0034] To mitigate the gradient offset problem and ensure symmetrical weight updates during gradient descent, the tansig function is used as the activation function for the hidden layers of the BP neural network. ; In the formula, x i This is the output value of the previous layer. y l This is the input value for the next layer; The output layer uses the purelin function as the activation function: y=purelin(y l )=y l ; In the formula, y l This is the output value of the previous layer. y This is the final output value.

[0035] The Levenberg-Marquardt learning algorithm was used for network training. This algorithm was implemented using the trainlm function in MATLAB software. Its core is to improve the convergence speed of the network by adjusting the damping coefficient to balance the gradient descent and the iterative stability of the Gauss-Newton method. The sample set was randomly divided into training set, prediction set and validation set, with the division ratios being 70%, 15% and 15%, respectively.

[0036] Statistical indicators include the coefficient of determination R. 2 The values ​​of root mean square error (RMSE) and mean absolute percentage error (MAPE) are calculated using the following formulas: ; In the formula, Y i,p These are the model's predicted values. Y i,e These are experimentally measured values. Y m is the average value of the experimental measurements, and n is the sample size.

[0037] R for training a BP neural network model 2 The RMSE and MAPE values ​​were 0.9309, 0.0008, and 0.8682%, respectively, indicating that the model has good fitting ability.

[0038] The weights and biases of the constructed blood glucose lowering prediction model are shown in Table 2. The constructed model can be expressed as: ; in, Y ANN b1 represents the model's predicted value, LW represents the output layer weights, IW represents the input layer weights, b1 represents the input layer bias, and b2 represents the output layer bias.

[0039]

[0040] The model prediction results are shown in Table 3.

[0041]

[0042] In this embodiment, in step S5, a mulberry leaf sample B from a different source than the input sample is selected. The material-liquid ratio and stirring speed are kept constant. Extraction parameters are randomly selected within the range of extraction temperature 60-120 ℃ and extraction time 0.5-120 min for extraction. A total of 3 independent verification samples with dual variability in source and extraction parameters are prepared.

[0043] For Sample 1, the extraction temperature was set at 60℃ and the extraction time was 10 min.

[0044] For Sample 2, the extraction temperature was set at 80℃ and the extraction time was set at 100 min.

[0045] For sample 3, the extraction temperature was set at 95℃ and the extraction time was 55 min.

[0046] Key active substances were measured and substituted into the trained model to output the predicted α-glucosidase inhibition rate, which was then compared with the actual value. The results are shown in Table 4. The predicted values ​​of the verification samples are close to the actual values, and the relative prediction error is less than 1%, indicating that the model can achieve good prediction results.

[0047]

Claims

1. A method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network, characterized in that, Includes the following steps: S1, by treating mulberry leaf raw materials under different extraction conditions, a wide range of representative mulberry leaf water extract samples were prepared; S2, determine the content of multiple active ingredients and α-glucosidase inhibition rate of each mulberry leaf water extract sample; S3. Through correlation analysis, the correlation strength between the content of each active ingredient and the α-glucosidase inhibition rate is quantified, and key active ingredients that are significantly correlated with the α-glucosidase inhibition rate are screened from the various active ingredients. S4. Define the BP neural network structure, use the content of key active ingredients as input variables and the α-glucosidase inhibition rate as output variables, construct and train the BP neural network model, use statistical indicators to evaluate the predictive performance of the model, and finally obtain the blood sugar lowering activity prediction model of mulberry leaf water extract. S5 uses a validation sample set independent of the input set to externally validate the model's universality. Test samples are prepared, and the content of key active ingredients and α-glucosidase inhibition rate are measured. The data are then input into the mulberry leaf water extract hypoglycemic activity prediction model, and the prediction results are output.

2. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: In S1, the feed-to-liquid ratio is set to 1:

40. w:v The stirring speed was set to 700 r / min, the heat treatment temperature T was continuously varied from 60°C to 100°C, and the extraction time t was continuously varied from 0.5 min to 120 min.

3. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: S2 contains at least the following active ingredients: total flavonoids, total phenols, total sugars, DNJ, and total protein.

4. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: In S3, the correlation analysis is a Pearson correlation analysis, and the calculation method is as follows: ； In the formula, r xy The correlation coefficients between the content of each active ingredient and the α-glucosidase inhibition rate are given. x i This represents the original data value of the active substance content. x̄ This represents the average value among similar indicators of active substance content. y i This represents the raw data value of α-glucosidase inhibition rate. ȳ This represents the average value among similar indicators of α-glucosidase inhibition rate.

5. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 4, characterized in that: In S3, active ingredients with a correlation coefficient ≥ 0.7 with α-glucosidase inhibition rate are selected as the key active ingredients; the key active ingredients include total flavonoids, total phenols, total sugars and DNJ.

6. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 5, characterized in that: In S3, a four-dimensional relationship diagram of "process parameters-component content-bioactivity" is drawn, with extraction temperature as the X-axis and time as the Y-axis, forming a process parameter plane; the α-glucosidase inhibition rate is used as the Z-axis to characterize the level of bioactivity; the content of candidate active ingredients is mapped as a color gradient to visualize the consistency of the changes in the content of each active ingredient and the α-glucosidase inhibition rate in the multi-dimensional process space, and active ingredients with high consistency are selected as key active ingredients.

7. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: In S4, the method for constructing the blood sugar-lowering activity prediction model of mulberry leaf water extract includes: setting the number of layers of the BP neural network to 3, with 4 nodes in the input layer, 1 node in the output layer, and 10 neurons in the hidden layer.

8. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: In S4, the statistical indicators include the coefficient of determination R. 2 The values ​​of root mean square error (RMSE) and mean absolute percentage error (MAPE) are calculated using the following formulas: ； In the formula, Y i,p These are the model's predicted values. Y i,e These are experimentally measured values. Y m is the average value of the experimental measurements, and n is the sample size.

9. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 7, characterized in that: In S4, to alleviate the gradient offset problem and ensure symmetrical weight updates during gradient descent, the tansig function is used as the activation function for the hidden layers of the BP neural network. ； In the formula, x i This is the output value of the previous layer. y l This is the input value for the next layer; The output layer uses the purelin function as the activation function: y=purelin(y l )=y l ； In the formula, y l This is the output value of the previous layer. y This is the final output value.

10. The method for predicting the hypoglycemic activity of mulberry leaf water extract based on a BP neural network according to claim 1, characterized in that: The predictive model for the hypoglycemic activity of mulberry leaf water extract is as follows: ； in, Y ANN b1 represents the model's predicted value, LW represents the output layer weights, IW represents the input layer weights, b1 represents the input layer bias, and b2 represents the output layer bias.

Images