Machine learning based formulation screening method for low gi white kidney bean noodles

Abstract

This invention belongs to the field of food development technology, specifically relating to a machine learning-based method for screening low-GI white kidney bean noodle recipes. The invention uses white kidney bean flour and wheat flour as noodle ingredients. The temperature, time, and enzyme activity of the in vitro digestion experiment, the mass ratio of white kidney bean flour to wheat flour in the flour ingredients, and the protein, dietary fiber, starch, and fat content of the flour ingredients are used as input features for the machine learning model. The GI value determined in the in vitro digestion experiment is used as the output label for the machine learning model. The machine learning model is then constructed and trained. The trained model is then used to perform high-throughput GI value prediction on the amount of white kidney bean flour added, screening out low-GI noodle recipes with a GI value ≤ 55. Noodles made using the selected recipes are then experimentally verified. This method can accurately capture the nonlinear dose-response relationship between the amount of white kidney bean added and the GI value of the noodles.

Description

Technical Field

[0001] This invention belongs to the field of food development technology, specifically relating to a method for screening the recipe of low-GI white kidney bean noodles based on machine learning. Background Technology

[0002] Currently, the development of low-GI flour products mainly adopts the traditional technical route of "raw material substitution + in vitro digestion experiments," including: Formulation optimization stage: This often involves single-factor or orthogonal experiments to optimize the amount of whole grains and legumes added, relying on in vitro starch digestion experiments (such as α-amylase hydrolysis) or animal and human in vivo experiments to determine the GI value and verify the formula's hypoglycemic effect; Mechanism analysis stage: This stage mainly focuses on the physical barrier effect of raw materials (such as dietary fiber encapsulating starch) or enzyme inhibition effect (such as α-amylase inhibitors), lacking systematic mechanism verification at the molecular target and signaling pathway levels; Product development stage: This stage primarily focuses on the sensory quality and physicochemical indicators of the product, failing to form a closed-loop technical system integrating formulation optimization, functional prediction, and mechanism analysis.

[0003] The existing methods for developing low-GI flour products mainly have the following problems:

[0004] GI value prediction is inefficient and costly: in vitro digestion experiments and in vivo GI determination are time-consuming and costly, and it is difficult to achieve high-throughput screening of formulations; traditional regression models have insufficient fitting accuracy for the relationship between formulations and GI, and cannot accurately capture nonlinear dose-response relationships, resulting in low efficiency of formulation optimization.

[0005] The mechanism analysis is unsystematic and superficial: existing technologies only observe digestive behavior at the macro level, lacking a systematic analysis of the target sites and signaling pathways of active ingredients in white kidney beans (such as ferulic acid, vanillic acid, and polyphenols), and thus failing to elucidate the molecular mechanism of the hypoglycemic effect of noodles.

[0006] The disconnect between formulation optimization and functional verification: In the traditional technical route, formulation optimization, GI prediction and mechanism verification are independent of each other and have not formed a system from "formulation screening → GI prediction → mechanism verification", resulting in long product development cycles and insufficient targeting.

[0007] The products have limited functionality and lack synergistic effects: existing low-GI noodles mostly focus on reducing the rate of starch digestion, without taking into account additional health benefits such as anti-oxidation and improved insulin sensitivity, and the comprehensive nutritional value of the products has not been fully explored. Summary of the Invention

[0008] To address the problems in existing low-GI noodle product development technologies, such as the inability of formula optimization to accurately capture the nonlinear dose-response relationship between the amount of white kidney beans added and the GI value of noodles, and the fact that formula optimization, GI prediction, and mechanism verification are independent and do not form a closed-loop development system, resulting in long product development cycles and insufficient targeting, this invention provides a formula screening method for low-GI white kidney bean noodles based on machine learning.

[0009] The method for screening low-GI white kidney bean noodle recipes based on machine learning proposed in this invention specifically adopts the following technical solution:

[0010] Model Construction and Training: White kidney bean flour and wheat flour were used as noodle raw materials. The temperature, time, enzyme activity of the in vitro digestion experiment of noodles, the mass ratio of white kidney bean flour to wheat flour in the flour raw materials, and the protein content, dietary fiber content, starch content, and fat content of the noodle raw materials were used as input features of the machine learning model. The GI value determined by the in vitro digestion experiment was used as the output label of the machine learning model. The protein content, dietary fiber content, starch content, and fat content refer to the total protein content, total dietary fiber content, total starch content, and total fat content in wheat flour and white kidney bean flour.

[0011] High-throughput formulation screening: A trained machine learning model was used to predict the glycemic index (GI) of white kidney bean flour addition using high-throughput methods. Low-GI noodle formulations with an EGI value ≤ 55 (white kidney bean flour addition of 40%~50%) were screened, achieving efficient and accurate formulation optimization. Noodles were then made using the selected formulations for experimental verification. The white kidney bean flour addition refers to the mass ratio of white kidney bean flour to the noodle raw materials, specifically 0~50%, such as 0%, 10%, 20%, 30%, 40%, 50%. The machine learning models included Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). The XGBoost model, as the core prediction model, optimized model parameters through cross-validation, achieving accurate fitting of the non-linear relationship between white kidney bean flour addition and noodle GI value (R²). 2 >0.99), breaking through the linear fitting limitations of traditional regression models; therefore, the model adopts the core prediction model XGBoost selected from the screening.

[0012] This invention also elucidates the mechanism of action of the active ingredients based on network pharmacology, including:

[0013] Target and pathway prediction: Thirty active ingredients in white kidney bean were identified using LC-MS. Based on the TCMSP database, potential targets of key active ingredients (ferulic acid FA, vanillic acid VA, resveratrol RES, quercetin QUE) were screened. An interaction network of "ingredient-target-pathway" was constructed, and core hub targets such as AKT1 (protein kinase B), GSK3B (glycogen synthase kinase-3β), SRC (Src tyrosine protein kinase), and ALB (albumin) and key blood glucose regulation pathways such as PI3K-Akt (phosphatidylinositol 3-kinase-protein kinase B) and MAPK (mitogen-activated protein kinase) were screened.

[0014] Molecular docking and dynamics verification: Molecular docking was used to simulate the binding mode of the active ingredient and the core target protein (such as AKT1) to verify its non-covalent binding modes such as hydrogen bonding and hydrophobic interactions; further, 100 ns molecular dynamics simulation was used to verify the dynamic stability of the complex conformation, and to elucidate the mechanism of action of the active ingredient in regulating blood glucose at the molecular level.

[0015] Based on the aforementioned method for developing low-GI white kidney bean noodles, this invention provides a method for preparing low-GI white kidney bean noodles, comprising:

[0016] Raw material pretreatment: Mix white kidney bean flour and wheat flour according to the low-GI noodle formula selected by screening, and pass through a 140-mesh sieve to ensure the uniformity of flour particles;

[0017] Dough mixing and rolling: Add water at a powder-to-water ratio of 1:0.5~0.6 and mix for 10~15 minutes. Roll the dough to a thickness of 1.0~1.5 mm to form a uniform sheet.

[0018] Steaming and drying: Steam in boiling water for 3-5 minutes, drain and then dry with hot air (temperature 60-70℃, time 30-40 minutes) to obtain finished noodles with a moisture content ≤12%.

[0019] This invention optimizes the amount of white kidney bean flour added to achieve the following noodle texture:

[0020] EGI value ≤ 55;

[0021] Enhanced antioxidant activity: DPPH free radical scavenging rate ≥54%, ABTS free radical scavenging rate ≥62%;

[0022] Optimization of cooking quality: Rehydration ratio ≥150%, cooking loss rate ≤10%, ensuring the taste and processing stability of noodles.

[0023] The beneficial effects of this invention are as follows:

[0024] (1) Significantly improved efficiency and accuracy of formulation optimization: Based on machine learning model-based GI value prediction technology, the goodness of fit R2 >0.99 can accurately capture the non-linear dose-response relationship between the amount of white kidney bean added and the GI value of noodles, improving the efficiency of formula screening by more than 5 times, eliminating the need for a large number of in vitro and in vivo experiments, and significantly reducing development costs;

[0025] (2) Breakthrough in mechanism analysis from macroscopic to molecular level: Through network pharmacology, molecular docking and kinetic simulation, the mechanism of hypoglycemia of white kidney bean active ingredients by targeting core targets such as AKT1 and regulating the PI3K-Akt signaling pathway was systematically elucidated, providing molecular-level theoretical support for the functional verification of the product and solving the problem of insufficient mechanism analysis in existing technologies;

[0026] A closed-loop development system has been formed, making the products more targeted: a closed-loop technical route of "formula screening → GI prediction → mechanism verification → product preparation" has been constructed, with each link supporting each other, shortening the development cycle by more than 30%, and making the low GI characteristics and nutritional functions of the products highly controllable; it has solved the problem that the existing technology only stays at the observation of macroscopic digestive behavior and cannot systematically analyze the blood sugar lowering mechanism of white kidney bean active ingredients from the molecular target and signaling pathway level.

[0027] (3) Achieving multi-effect synergistic function and wider application scenarios: Based on the method of the present invention, a white kidney bean noodle with low GI characteristics and multi-effect synergistic functions such as anti-oxidation and promoting insulin resistance has been developed. The developed white kidney bean noodle not only has an EGI value ≤ 55, but also has high antioxidant activity and improves insulin sensitivity. It can meet the staple food needs of diabetic patients, obese people and ordinary healthy people at the same time, and has a wider application scenarios. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the machine learning approach used in this invention to screen low-GI white kidney bean noodle recipes.

[0029] Figure 2 This is a graph showing the predicted GI values ​​of white kidney bean noodles made from white kidney bean flour and wheat flour, based on the random forest (RF) model of machine learning in this invention.

[0030] Figure 3 This is a graph showing the predicted GI values ​​of white kidney bean noodles made from white kidney bean flour and wheat flour, based on the machine learning support vector machine (SVM) model of this invention.

[0031] Figure 4 This is a graph showing the predicted GI values ​​of white kidney bean noodles made from white kidney bean flour and wheat flour, based on the XGBoost (Gradient Boosting) machine learning model of this invention.

[0032] Figure 5This section tests the predictive performance of the three machine learning models used in this invention, and the results are presented in the form of scores.

[0033] Figure 6 The vertical axis represents the value of the reduced GI, calculated by predicting additional reductions in the GI value using the three machine learning models used in this invention (primarily considering the influence of the human digestive environment and insulin).

[0034] Figure 7 : These are the ROC curves of the three machine learning models used in this invention.

[0035] Figure 8 : This is a heatmap of the confusion matrix of the three machine learning models used in this invention. Darker colored blocks indicate that the model's predictions are accurate, while lighter colored blocks indicate that the model has made incorrect predictions.

[0036] Figure 9 This is a comparison chart of the GI values ​​predicted by three machine learning models—RF, SVM, and XGBoost—used in this invention. The horizontal axis represents the amount of white kidney bean powder added, and the vertical axis corresponds to the predicted GI values, which are used for subsequent experimental simulation and control.

[0037] Figure 10 This invention uses white kidney beans as raw material and measures the antioxidant properties of the raw material. The strength of the antioxidant properties of the raw material affects the antioxidant capacity of the finished product. The two curves correspond to DPPH and ABTS, respectively. + Sweep rate, with the X-axis representing the volume of the white kidney bean sample solution (the white kidney bean sample solution was prepared using white kidney bean powder with a concentration of 1 g / 100 mL).

[0038] Figure 11 : This is a graph showing the change in rehydration rate of white kidney bean noodles with different amounts of added ingredients produced according to the present invention.

[0039] Figure 12 : This is a graph showing the change in cooking loss rate of white kidney bean noodles with different amounts of added ingredients produced according to the present invention.

[0040] Figure 13 This is the glucose standard curve used in the in vitro digestion experiment of this invention to determine the glucose content in the sample.

[0041] Figure 14 : This is the first-order kinetic curve of starch hydrolysis rate of the sample during the in vitro digestion experiment of this invention from 0-180s.

[0042] Figure 15 : This refers to the changes in the content of rapidly digestible starch (RDS), slowly digestible starch (SDS), and resistant starch (RS) in each sample after the in vitro digestion experiment in this invention.

[0043] Figure 16The graph shows the starch hydrolysis index (HI) and predicted GI value (EGI) results from the in vitro digestion experiment in this invention. The blue bars represent HI, and the red bars represent EGI.

[0044] Figure 17 This invention analyzes ferulic acid (FA) and vanillic acid (VA) in relation to disease targets related to type 2 diabetes, hyperglycemia, and insulin resistance.

[0045] Figure 18 The diagram shows the PI3K-AKT signaling pathway. The PI3K-AKT signaling pathway mainly affects human glucose metabolism and insulin sensitivity, and is a key pathway for regulating blood glucose levels. The red part in the diagram marks the core target location of the active ingredients in white kidney beans in this study.

[0046] Figure 19 The image shows the mulberry bubble diagrams for FA and VA, illustrating the enrichment analysis results of the corresponding target sites of the two active ingredients from white kidney beans in the PI3K-AKT signaling pathway.

[0047] Figure 20 The image shows the enrichment analysis of FA and VA, illustrating the signaling pathways related to blood glucose regulation that are significantly enriched by the two active ingredients in white kidney beans.

[0048] Figure 21 : HOMO and LUMO orbital analysis of FA (A), VA (B), RES (C), and QUE (D), and molecular docking diagrams of FA, VA, RES, and QUE with α-amylase.

[0049] Figure 22 The graph is a root mean square error plot of molecular dynamics simulations of QUE (the best small molecule in the molecular dynamics simulation results) and α-amylase. The horizontal axis represents time (0-100 ns), and the vertical axis represents the root mean square error of the simulation system, showing the stability of the system structure fluctuations during the simulation process.

[0050] Figure 23 : This is the root mean square fluctuation plot of QUE and α-amylase molecular dynamics simulation. The horizontal axis represents the amino acid number, and the vertical axis represents the root mean square fluctuation value of each amino acid residue, demonstrating the conformational flexibility of different amino acid residues in the system.

[0051] Figure 24 The graph shows the gyration radius from the QUE and α-amylase molecular dynamics simulations. The horizontal axis represents time (0-100 ns), and the vertical axis represents the gyration radius of the system, illustrating the changes in the compactness of the overall α-amylase molecular structure during the simulation.

[0052] Figure 25This is a hydrogen bonding diagram from molecular dynamics simulations of QUE and α-amylase. The horizontal axis represents time (0-100 ns), and the vertical axis represents the number of hydrogen bonds formed between QUE and α-amylase during the simulation, demonstrating the stability and binding strength of hydrogen bonding between QUE and α-amylase. Red bonds represent interactions between the ligand and acceptor at a distance ≤ 0.35 nm, encompassing all close-range interactions such as hydrogen bonds, hydrophobic interactions, and van der Waals interactions. Blue bonds represent only hydrogen bond interactions between the ligand and acceptor at a distance ≤ 0.35 nm.

[0053] Figure 26 : This represents the solvent-accessible surface area of ​​QUE and α-amylase molecular dynamics simulations. The horizontal axis represents time (0-100 ns), and the vertical axis represents the solvent-accessible surface area of ​​the system. It shows the change in the surface area of ​​the molecule exposed in the solvent after QUE binds to α-amylase during the simulation, reflecting the influence of the binding of the two on the overall hydrophobicity and surface exposure of α-amylase.

[0054] Figure 27 : This is a 2D free energy morphology diagram of QUE and α-amylase molecular dynamics simulation, with the horizontal axis representing the first principal component (PC1) and the vertical axis representing the second principal component (PC2).

[0055] Figure 28 : This is a 3D free energy morphology diagram of QUE and α-amylase molecular dynamics simulation. The X-axis represents PC1, the Y-axis represents PC2, and the Z-axis represents free energy, with units of kcal / mol.

[0056] Figure 29 : This represents the binding free energy from QUE and α-amylase molecular dynamics simulations. The horizontal axis represents the energy terms of the binding free energy decomposition and the protein dissociation constant (Kd). The left vertical axis represents the free energy in kcal / mol, and the right vertical axis represents Kd in nM.

[0057] Figure 30 : This represents the residue binding energy from the molecular dynamics simulation of QUE and α-amylase. The horizontal axis represents the amino acid residues corresponding to the binding site, and the vertical axis represents the free energy. The unit is kcal / mol. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly described below. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0059] Example 1: Screening of Low-GI Noodle Recipes Based on Machine Learning

[0060] 1. Raw materials and reagents

[0061] White kidney bean powder (passed through a 140-mesh sieve) was purchased from Shanxian Wanding Trading Co., Ltd., wheat flour (high gluten flour, protein content ≥12%, brand: Jinlongyu) was purchased from the local market, ABTS and DPPH were purchased from Shanghai Maclean Biochemical Technology Co., Ltd., α-amylase (derived from pig pancreas, ≥100 U / mg) and glucosidase were purchased from Shanghai Yuanye Biotechnology Co., Ltd.

[0062] 2. Experimental Methods

[0063] The machine learning algorithm was developed using the R language, calling scikit-learn functions through the `reticulate` package in RStudio, and using Conda to manage the Python runtime environment to ensure that R can locate and call the scikit-learn library within the corresponding Python environment. Based on the scikit-learn database, 10,000 datasets were generated, randomly divided into training and test sets at an 80%:20% ratio to improve the model's generalization ability. Each dataset included temperature, time, enzyme activity from in vitro digestion experiments, as well as the protein, dietary fiber, starch, and fat content and mass ratio of wheat flour and white kidney bean flour. Three models—RF, SVM, and XGBoost—were selected for EGI value simulation analysis, and the final results were output as EGI values. Model accuracy was evaluated using ROC curves and model performance metrics, and a confusion matrix heatmap was plotted.

[0064] The overall experimental approach is as follows: Figure 1 As shown, based on machine learning models to predict low-GI formulations and GI values, suitable formulations were selected for in vitro digestion, antioxidant, and cooking experiments for verification. The hypoglycemic mechanism was explored by combining network pharmacology, molecular docking, molecular dynamics, and quantum chemistry techniques.

[0065] 3. Model Selection

[0066] Figure 2-4 The predicted GI value changes with the amount of white kidney beans added, corresponding to three machine learning models: RF, SVM, and XGBoost. Figure 2 For random forest models, Figure 3 For support vector machine models, Figure 3 (This refers to the gradient boosting algorithm model). Furthermore, the performance of the three models was tested separately, with poly used as a control group. The results are as follows: Figure 5 As shown. The additional reduction in GI values ​​for the three models was predicted, with poly as the control group. The results are as follows. Figure 6 As shown. Figure 7The ROC curves for the three models are shown. XGBoost has the largest AUC area at 0.9497. Figure 8 For the confusion matrix heatmap, darker colored blocks indicate accurate model predictions, while lighter colored blocks indicate model misclassifications. The number of colored blocks indicates that all three models have high GI classification accuracy (accuracy, precision, and recall all exceed 85%).

[0067] 4. Experimental verification

[0068] Using white kidney bean flour and wheat flour as noodle ingredients, the EGI values ​​of the produced white kidney bean noodles were predicted using three trained models: RF, SVM, and XGBoost, at white kidney bean addition levels of 10%, 20%, 30%, 40%, and 50% (corresponding to high-gluten flour addition levels of 90%, 80%, 70%, 60%, and 50%). The results are as follows: Figure 9 As shown.

[0069] Noodle samples were prepared using white kidney bean flour and wheat flour as noodle ingredients, with white kidney bean flour addition levels of 0% (control group), 10%, 20%, 30%, 40%, and 50%, and the EGI values ​​of the noodle samples were measured to verify the accuracy of the machine learning simulation. The noodle preparation method is as follows:

[0070] Take white kidney bean flour and wheat flour in the above proportions and pass them through a 140-mesh sieve to ensure uniformity of flour particles;

[0071] Dough preparation and rolling: Add water at a powder-to-water ratio of 1:0.55 and knead for 12 minutes. Roll the dough to a thickness of 1.2 mm to form a uniform sheet.

[0072] Steaming and drying: Steam in boiling water for 4 minutes, drain and then dry with hot air (temperature 65℃, time 35 minutes) to obtain finished noodles with a moisture content ≤12%.

[0073] Comparative validation results show that the XGBoost model is optimal, as it performs high-throughput GI value prediction for white kidney bean powder addition amounts (0-50%), with a model R... 2 =0.9925, prediction error <1%. Screening revealed that when white kidney bean flour was added at a concentration of 40%–50%, the highest EGI value for noodles was 55.2, and the lowest was 50.1, closest to the actual measured value (lowest EGI was 50.83). Furthermore, most noodles within this range met the low-GI food standard (GI≤55), with the lowest GI value at 50% addition, thus identified as the optimal formula (lowest GI). Subsequent steps will involve quality optimization of noodle products made with white kidney bean flour added in the 40%–50% range to select the optimal quality formula.

[0074] Example 2: Preparation and performance verification of low-GI white kidney bean noodles

[0075] Preparation of low-GI white kidney bean noodles:

[0076] The formula used was 0% (control group), 10%, 20%, 30%, 40%, 50% white kidney bean flour (passed through a 140-mesh sieve) + 100%, 90%, 80%, 70%, 60%, 50% wheat flour (high-gluten flour, protein content ≥12%), with a flour-to-water ratio of 1:0.55, a kneading time of 12 min, a rolling thickness of 1.2 mm, boiling water cooking for 4 min, and hot air drying at 65℃ for 35 min. The antioxidant activity of the white kidney bean flour was as follows: Figure 10 As shown.

[0077] Performance verification results for low-GI white kidney bean noodles show that the optimal formula is 50% white kidney bean flour + 50% wheat flour. The performance of white kidney bean noodles with this formula is as follows:

[0078] Starch digestibility characteristics: SDS ratio = 31.2%, RDS ratio = 19.8%, GI value = 50.1, which meets the low GI standard;

[0079] Antioxidant activity: DPPH free radical scavenging rate = 13.5%, ABTS free radical scavenging rate = 71.2%, significantly higher than the control group.

[0080] Cooking quality: Rehydration rate = 152.3% ( Figure 11 Cooking loss rate = 9.2% ( Figure 12 The noodles had a breakage rate of less than 5%, and their texture and taste were not significantly different from ordinary wheat noodles.

[0081] In addition, in vitro digestion experiments were conducted on white kidney bean noodles prepared with different formulations. The specific experimental method for in vitro digestion was as follows: 10 mL of pH buffer was added to the sample, followed by 10 mL of a solution containing a 1:3 mixture of 290 U / mL α-amylase and 15 U / mL glucosidase. The mixture was magnetically stirred at 37°C. At 0, 20, 60, 90, 120, 150, and 180 minutes, 0.5 mL of the digest was collected, and 2 mL of anhydrous ethanol was added to terminate the reaction. The mixture was then centrifuged at 5000 rpm for 10 minutes. The contents of RDS, SDS, and RS, as well as the starch hydrolysis rate during digestion, were calculated using the formulas in Englyst's method (Englyst et al., 1992). The digestion results were fitted using a first-order kinetic equation. A hydrolysis kinetic curve was plotted with digestion time on the x-axis and starch hydrolysis rate on the y-axis. The area under the curve was obtained by integration, and the starch hydrolysis index was calculated using the formula HI = (Area under the sample hydrolysis curve / Area under the white bread reference curve) × 100. The predicted glycemic index was calculated using the linear correlation model EGI = 8.198 + 0.862HI, and the proportions of rapidly digestible starch (RDS), slowly digestible starch (SDS), and resistant starch (RS) were simultaneously determined. Finally, the hydrolysis index (HI) and glycemic index (EGI) were calculated using the method described by Goni (Goni et al., 1997). The digestibility of the noodle formulation samples optimized by machine learning was verified, with standard white bread as the reference sample, and three parallel samples were set for each group.

[0082] Tests showed that the starch hydrolysis kinetics curve of the optimized formula (50% white kidney bean flour + 50% wheat flour) changed gradually after 120 seconds (e.g., ...). Figure 14 As shown), the sample RDS shows a decreasing trend, while RS shows an increasing trend (e.g. Figure 15 As shown), the lowest measured HI was 49.57 and EGI was 50.83 (as shown). Figure 16 As shown in the figure, the levels were significantly lower than those of the white bread reference sample, indicating stable digestibility and meeting the pre-set standards for low-glycemic index applications. Figure 13 This is the glucose standard curve used in in vitro digestion experiments.

[0083] Example 3: Mechanism analysis and verification using network pharmacology and molecular simulation.

[0084] Table 1. Identification of 30 bioactive substances in white kidney bean seeds by LC-MS

[0085]

[0086] like Figure 17-30 As shown, the analysis and verification will be conducted from the following aspects:

[0087] (1) Target and pathway prediction: Based on Table 1, four key active substances in white kidney bean were screened: ferulic acid (FA), vanillic acid (VA), resveratrol (RES), and quercetin (QUE). Potential targets were screened using the TCMSP (https: / / www.91tcmsp.com / # / database) database and the Pharmmapper database (https: / / www.lilab-ecust.cn / pharmmapper / ). Human gene loci related to "type 2 diabetes", "hyperglycemia", and "insulin resistance" were screened using the Genecards database (https: / / www.genecards.org / ). A protein-protein interaction (PPI) network was constructed, and core targets such as AKT1, GSK3B, SRC, and ALB, as well as the PI3K-Akt signaling pathway, were screened. Figure 17 The relevant gene loci for FA and VA and the key targets after screening (RES and QUE were analyzed and screened using the same methods). Figure 18 This is a diagram of the PI3K-AKT signaling pathway. The PI3K-AKT signaling pathway mainly affects human glucose metabolism and insulin sensitivity, and is a key pathway for regulating blood glucose levels. The red part in the diagram marks the core target location of the active ingredients in white kidney beans in this study. Figure 19 Sankey bubble chart (RES and QUE use the same analysis and screening methods). Figure 20 Pathway analysis diagrams for FA and VA (RES and QUE were analyzed and screened using the same methods).

[0088] (2) Verification of HOMO-LUMO orbital theory: Quantum chemical calculation methods were used to perform frontier molecular orbital analysis on four key active components of white kidney bean, calculating the energy levels and band gaps of the highest occupied molecular orbitals (HOMO) and lowest vacant molecular orbitals (LUMO) of each component, as well as the orbital electron cloud distribution. Figure 21 On the left, HOMO orbitals reflect a molecule's electron-donating ability, while LUMO orbitals reflect its electron-accepting ability. The band gap directly characterizes a molecule's chemical reactivity, charge transfer capability, and biological activity potential. The results show that smaller QUE and RES band gaps indicate stronger molecular reactivity, while FA and VA possess excellent charge transfer characteristics. The frontier orbitals of each active ingredient are mainly distributed in the active functional group regions such as the benzene ring and phenolic hydroxyl groups, highly matching the active binding sites of subsequent targets. This verifies, from the perspective of molecular electronic structure, that the four active substances possess excellent pharmacological reactivity and target binding potential.

[0089] (3) Molecular docking verification: such as Figure 21As shown on the right, FA, VA, RES, and QUE bind to α-amylase and the aforementioned key screening targets (such as AKT1, GSK3B, and SRC). This demonstrates that small molecules, after entering the human body, affect key proteins in blood glucose control, thereby lowering blood glucose levels. This further explains the functional mechanism of low-GI noodles at the molecular level. The results show that the four active ingredients can spontaneously form stable complexes with the target proteins, interacting through hydrogen bonds, hydrophobic interactions, and other mechanisms within the protein's active pocket. The binding energies are all less than -5.0 kcal / mol. Quercetin (QUE) exhibits the most significant binding activity with AKT1, and ferulic acid (FA) with ALB, indicating that the active ingredients can directly target and regulate the function of core proteins.

[0090] (4) Molecular dynamics simulation: 100 ns molecular dynamics simulations were performed on the active ingredient-target protein complex. The stability of the system was evaluated using parameters such as root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (RG), hydrogen bonding, and solvent accessible surface area (SASA). Figure 22-26 As shown in the figure. The results indicate that the complex system tends to be structurally stable during the simulation, with the RMSD curve fluctuating smoothly and no significant conformational changes in the protein backbone; the RMSF fluctuation of key amino acid residues is low, and the conformation of the active site remains stable; the radius of gyration (Rg) and solvent accessible surface area (SASA) curves are stable, and hydrogen bonding continues to exist, proving that the complex formed by the active ingredient and the target protein has good conformational stability.

[0091] (5) Free energy analysis and combined model analysis: Through principal component analysis (PCA) and free energy geomorphological analysis, a free energy variation spectrum of the system is plotted, such as... Figure 27-30 As shown, the results indicate that the complex contains a low-energy stable conformation cluster, and the binding conformation of the active ingredient and the target protein is in a thermodynamically dominant state. The binding free energy and the contribution of each component were calculated by the MMPBSA method. Van der Waals forces and electrostatic interactions are the main driving forces. Among them, key residues such as VAL49 and ALA107 make significant contributions to the binding energy, further elucidating the interaction mechanism between the active ingredient and the target protein and verifying the molecular basis of its hypoglycemic activity by targeting the core pathway.

[0092] The above results indicate that the active ingredients of white kidney beans can improve insulin sensitivity and reduce postprandial blood glucose response by stably binding to the AKT1 target and regulating the PI3K-AKT signaling pathway.

[0093] The above results indicate that the noodles prepared in this embodiment have low GI characteristics and can be promoted and applied as a low GI staple food product.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for screening recipes for low-GI white kidney bean noodles based on machine learning, characterized in that, include: Using white kidney bean flour and wheat flour as noodle raw materials, the temperature, time, enzyme activity of the in vitro digestion experiment of noodles, the mass ratio of white kidney bean flour to wheat flour in the flour raw materials, and the protein content, dietary fiber content, starch content and fat content of the flour raw materials were used as input features of the machine learning model. The GI value measured in the in vitro digestion experiment was used as the output label of the machine learning model. The machine learning model was constructed and trained. A trained machine learning model was used to predict the glycemic index (GI) of white kidney bean powder addition using high-throughput methods. Low-GI noodle recipes with an EGI value ≤ 55 were then selected and used to make noodles for experimental verification.

2. The method for screening the recipe of low-GI white kidney bean noodles based on machine learning according to claim 1, characterized in that, The amount of white kidney bean powder added refers to the mass ratio of white kidney bean powder to noodle raw materials, and the amount of white kidney bean powder added is 0~50%.

3. The method for screening the recipe of low-GI white kidney bean noodles based on machine learning according to claim 1, characterized in that, The machine learning models include random forest, support vector machine, and extreme gradient boosting.

4. The method for screening the recipe of low-GI white kidney bean noodles based on machine learning according to claim 1, characterized in that, The machine learning model described is extreme gradient boosting.

5. The method for screening the recipe of low-GI white kidney bean noodles based on machine learning according to claim 2, characterized in that, Low-GI noodle formulas with a GI value ≤ 55 selected: white kidney bean flour added at a rate of 40%~50%.

6. A method for preparing low-GI white kidney bean noodles, characterized in that, include: (1) Mix white kidney bean flour and wheat flour according to the method described in any one of claims 1-5 to obtain a low-GI noodle formula, and then sieve it; (2) Kneading and rolling; (3) Steaming and drying.

7. The method for preparing low-GI white kidney bean noodles according to claim 6, characterized in that, Step (2) includes: Add water to the flour at a ratio of 1:0.5~0.6 and knead for 10~15 minutes. Roll out the dough to a thickness of 1.0~1.5 mm to form a uniform sheet.

8. The method for preparing low-GI white kidney bean noodles according to claim 6, characterized in that, Step (3) includes: boiling in boiling water for 3-5 minutes, draining and then drying with hot air to obtain finished noodles with a moisture content of ≤12%.

9. The method for preparing low-GI white kidney bean noodles according to claim 8, characterized in that, The temperature for hot air drying is 60~70℃, and the time is 30~40 minutes.

Images