Screening method and application of sweet substances

Abstract

The invention provides a sweet substance screening method and application, and relates to the field of sweet substance development and application, and the sweet substance screening method is characterized by comprising the following steps: providing a sweet substance, analyzing the sweet substance by adopting an electronic tongue technology and a molecular docking technology to obtain first information and second information of the sweet substance, and constructing a bimodal model at the same time, obtaining a screening result; according to the technical scheme provided by the invention, taste features of various common sweet substances are analyzed in a mode of combining an electronic tongue technology and a molecular docking technology, so that taste feature differences of the sweet substances and sucrose can be effectively compared; a machine learning prediction model based on taste attribute values of nine electronic tongue sweet substances and four molecular docking characteristic values is established, and sensory evaluation is used for verification; sweet substances similar to cane sugar in taste are rapidly screened, and the screening accuracy is improved. Based on the method, a thought is provided for excavation and development of novel sweet substances.

Description

Technical Field

[0001] This invention relates to the field of sweetener technology, and in particular to a method for screening sweeteners and their applications. Background Technology

[0002] Sweeteners are characterized by high sweetness and low calories, but most sweeteners are not pure, often containing bitterness, astringency, or other off-flavors. Therefore, rapid screening of sweeteners with a taste similar to sucrose is crucial in actual production. The addition of sweeteners requires multiple screening steps, but traditional sensory evaluation-based methods are time-consuming and labor-intensive, increasing production complexity and cost. Furthermore, subjective factors can lead to significant biases in the screening results. In addition, in the development of novel sweeteners, which exhibit diverse structures and types, relying primarily on sensory evaluation is prone to bias and error, becoming a limiting factor in the development process.

[0003] Currently, methods for evaluating sweet substances using either electronic tongues or molecular docking have been established. However, electronic tongue technology struggles to explain the fundamental causes of sweetness and lacks a direct explanation of the relationship between the molecular structure and sweetness of sweet substances. Furthermore, it has limitations in detecting some sweet substances. Molecular docking, due to limitations in computational models, fails to adequately consider factors such as solvation effects and acceptor flexibility in real-world systems. Moreover, its high computational cost makes it difficult to accurately reflect the synergistic or competitive interactions between multiple molecules.

[0004] Therefore, it is urgent to develop a method that can quickly and efficiently screen sweet substances. This will not only promote the development of the sweet substance compounding industry and meet the market demand for high-quality and novel sweet substances, but also have great significance for promoting the development of food processing and other related fields. Summary of the Invention

[0005] The main objective of this invention is to propose a method and application for screening sweet substances, aiming to provide a method that can quickly and efficiently screen sweet substances.

[0006] To achieve the above objectives, this invention proposes a method for screening sweet substances and its application, comprising the following steps:

[0007] S10. Provide a sweet substance, perform electronic tongue analysis on the sweet substance, and obtain first information about the sweet substance;

[0008] S20. Molecular docking is performed on the sweet substance, and it is bound to a sweet taste receptor protein to obtain the second information of the sweet substance;

[0009] S30. Perform principal component analysis and cluster analysis on the first information respectively, and perform cluster analysis on the second information to locate the range of target sweet substances.

[0010] S40. By constructing a first model, the first information is subjected to predictive loss analysis, and the electronic tongue sweetness is obtained based on the first model;

[0011] S50. By constructing a second model, the second information is used for predictive loss analysis, and the molecular docking sweetness is obtained based on the second model.

[0012] S60. Construct a bimodal joint loss function by combining the first model and the second model, and obtain bimodal sweetness consistency based on the bimodal joint loss function;

[0013] S70. Extract the first and second information of the sweet substance and the target sweet substance respectively, perform similarity analysis, and obtain bimodal feature similarity;

[0014] S80. Compare the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency between the sweet substance and the target sweet substance to obtain the screening results.

[0015] In one embodiment, the first information includes the taste characteristics of the sweet substance; and / or,

[0016] The second information includes binding free energy, docking score, number of hydrogen bonds, and hydrophobic interactions; and / or,

[0017] The sweet taste receptor proteins include TAS1R2 and / or TAS1R3.

[0018] In one embodiment, the step of performing electronic tongue analysis on the sweet substance to obtain first information about the sweet substance includes:

[0019] A sweetener solution and a reference solution are provided. The sweetness sensor of the electronic tongue is immersed in the reference solution and the sweetener solution, respectively, and the potential V1 of the reference solution and the potential V2 of the sweetener solution are measured.

[0020] After cleaning the sweetness sensor in the sweetness substance solution, it was immersed in the reference solution, and the aftertaste potential V3 was measured.

[0021] The initial taste value R = V2 - V1, and the aftertaste value CPA = V3 - V1.

[0022] In one embodiment, the step of immersing the sweetness sensor in the reference solution after cleaning it from the sweetness substance solution is as follows:

[0023] During cleaning, deionized water is used for rinsing, with a rinsing time of 10–20 seconds and a deionized water volume of 1–3 mL / s; and / or,

[0024] The initial taste value includes at least one of sweetness, sourness, bitterness, astringency, umami, and saltiness.

[0025] The aftertaste value includes at least one of bitter aftertaste value and astringent aftertaste value.

[0026] In one embodiment, the formula for the first model in step S40 is as follows:

[0027] S E-tongue =β T X E-tongue

[0028] Among them, S E-tongue X represents the electronic tongue sweetness of the sweet substance. E-tongue β is the electronic tongue feature vector of sweet substances. T The weighting coefficient for the target sweetener.

[0029] In one embodiment, the formula for the second model in step S50 is as follows:

[0030] S dock =α T X dock

[0031] Among them, X dock S is the molecular docking feature vector of sweet substances. dock For the molecular docking sweetness of the sweet substance to be tested, α T This is the weight vector of the target sweetener.

[0032] In one embodiment, step S60 includes: obtaining the maximum canonical correlation coefficient ρ by performing canonical correlation analysis on the first information and the second information, and constructing a bimodal joint loss function by combining the first model and the second model. The bimodal joint loss function is as follows:

[0033]

[0034] Among them, S ref For reference sweetness, λ1, λ2, and λ3 are preset weighted hyperparameters.

[0035] In one embodiment, in step S70, the formula for the bimodal feature similarity is as follows:

[0036]

[0037] Where ω is the weighting coefficient and M is the evaluation method for bimodal feature similarity.

[0038] In one embodiment, step S80 includes: comparing the comparison result with preset screening conditions to obtain a screening result; when the comparison result meets the preset screening conditions, it means that the sweet substance can replace the target sweet substance; when the comparison result does not meet the preset screening conditions, it means that the sweet substance cannot replace the target sweet substance.

[0039] This invention provides a computer device, comprising:

[0040] Memory;

[0041] Processor; and

[0042] A computer program stored in the memory; wherein the processor executes the computer program to implement the method for screening sweet substances as described above.

[0043] In the technical solution of this invention, the taste characteristics of common sweet substances are analyzed by utilizing electronic tongue technology and molecular docking technology, quantifying the first and second information of sweet substances. This effectively reflects the taste characteristics of each sweet substance. Furthermore, based on the taste characteristics of different sweet substances, and through further optimization using a bimodal model, sweet substances with similar taste to the target sweet substance can be quickly and accurately screened from multiple sweet substances. This provides a new approach for quickly and accurately obtaining substitutes for the target sweet substance, saving time and resources consumed by random and blind trial-and-error. It also provides a reference for enterprises to screen high-quality and novel sweet substances needed for the production of target sweet products. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram showing the taste values ​​of eight sweet substances according to an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the electronic tongue measurement results of sweet substances in one embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the first information predicting electronic tongue sweetness in one embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram illustrating the prediction of molecular docking sweetness using the second information method according to an embodiment of the present invention.

[0049] Figure 5 This is a schematic diagram of the loss value of a dual-modal joint loss function according to an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram showing the feature similarity results under two modalities using an embodiment of the present invention.

[0051] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them.

[0053] It should be noted that, unless specific conditions are specified in the embodiments, conventional conditions or conditions recommended by the manufacturer should be followed. Reagents or instruments whose manufacturers are not specified are all commercially available products. Furthermore, the meaning of "and / or" throughout the text includes three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. In addition, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, such a combination should be considered non-existent and not within the scope of protection claimed by this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention.

[0054] Currently, methods for evaluating sweet substances using either electronic tongues or molecular docking have been established. However, electronic tongue technology struggles to explain the fundamental causes of sweetness and lacks a direct explanation of the relationship between the molecular structure and sweetness of sweet substances. Furthermore, it has limitations in detecting some sweet substances. Molecular docking, due to limitations in computational models, fails to adequately consider factors such as solvation effects and acceptor flexibility in real-world systems. Moreover, its high computational cost makes it difficult to accurately reflect the synergistic or competitive interactions between multiple molecules.

[0055] Therefore, it is urgent to develop a method that can quickly and efficiently screen sweet substances. This will not only promote the development of the sweet substance compounding industry and meet the market demand for high-quality and novel sweet substances, but also have great significance for promoting the development of food processing and other related fields.

[0056] In view of this, the present invention proposes a method for rapid screening of sweet substances, comprising the following steps:

[0057] S10. Provide a sweet substance, perform electronic tongue analysis on the sweet substance, and obtain first information about the sweet substance;

[0058] S20. Molecular docking is performed on the sweet substance, and it is bound to a sweet taste receptor protein to obtain the second information of the sweet substance;

[0059] S30. Perform principal component analysis and cluster analysis on the first information respectively, and perform cluster analysis on the second information to locate the range of target sweet substances.

[0060] S40. By constructing a first model, the first information is subjected to predictive loss analysis, and the electronic tongue sweetness is obtained based on the first model;

[0061] S50. By constructing a second model, the second information is used for predictive loss analysis, and the molecular docking sweetness is obtained based on the second model.

[0062] S60. Construct a bimodal joint loss function by combining the first model and the second model, and obtain bimodal sweetness consistency based on the bimodal joint loss function;

[0063] S70. Extract the first and second information of the sweet substance and the target sweet substance respectively, perform similarity analysis, and obtain bimodal feature similarity;

[0064] S80. Compare the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency between the sweet substance and the target sweet substance to obtain the screening results.

[0065] In the technical solution of this invention, the taste characteristics of common sweet substances are analyzed by utilizing electronic tongue technology and molecular docking technology, quantifying the first and second information of sweet substances. This effectively reflects the taste characteristics of each sweet substance. Furthermore, based on the taste characteristics of different sweet substances, and through further optimization using a bimodal model, sweet substances with similar taste to the target sweet substance can be quickly and accurately screened from multiple sweet substances. This provides a new approach for quickly and accurately obtaining substitutes for the target sweet substance, saving time and resources consumed by random and blind trial-and-error. It also provides a reference for enterprises to screen high-quality and novel sweet substances needed for the production of target sweet products.

[0066] In one embodiment, the first information includes the taste characteristics of the sweet substance sample solution, specifically, the taste characteristics include an initial taste value and a follow-through taste value.

[0067] The second information includes at least one of binding energy, docking score, number of hydrogen bonds, and hydrophobic interactions.

[0068] In the technical solution of this invention, by detecting the initial taste value and aftertaste value of sweet substances, the taste characteristics of each sweet substance can be analyzed from two dimensions: initial taste value and aftertaste value. Effective distinctions can be made based on sweetness, sourness, bitterness, astringency, aftertaste bitterness, aftertaste astringency, saltiness, umami, and richness. From a variety of (generally 12) common sweet substances, sweet substances with similar taste characteristics to sucrose and less affected by differences in various taste attributes can be screened. The docking stability score reflects the stability of the sweet substance after docking with the sweet receptor, thus reflecting the persistence of sweetness and the degree of aftertaste to a certain extent. Furthermore, molecular docking technology is used to reveal the sweetness perception of common sweet substances at the molecular level, the interaction mechanism between each sweet substance monomer and the sweet receptor protein TS1R2 / TS1R3, and the stability of their binding with the sweet receptor protein TS1R2 / TS1R3, thus obtaining the second information of the sweet substance.

[0069] Specifically, the binding free energy can reflect the ease with which sweet substances bind to sweet receptors, and to some extent, it can reflect the relative sweetness of the sweetness. By combining the nine parameter values ​​of taste attributes, the binding stability score, and the seven parameters of binding free energy, target sweet substances that meet the requirements of long sweetness duration, pure sweet taste, and few off-flavors and aftertastes can be screened.

[0070] In one embodiment, the sweet taste receptor protein includes TAS1R2 and / or TAS1R3.

[0071] In the technical solution of this invention, by using TAS1R2 and / or TAS1R3 as sweet taste receptor proteins, the binding of each sweet taste substance to the sweet taste receptor protein can be reflected from the chemical structure, thereby judging the sweet taste perception intensity, sweet taste persistence, and other aspects of the sweet taste substance, and providing more parameter indicators for the screening of new sweet taste substances.

[0072] In one embodiment, the step of performing electronic tongue analysis on the sweet substance to obtain first information about the sweet substance includes:

[0073] A sweetener solution and a reference solution are provided. The sweetness sensor of the electronic tongue is immersed in the reference solution and the sweetener solution, respectively, and the potential V1 of the reference solution and the potential V2 of the sweetener solution are measured.

[0074] After cleaning the sweetness sensor in the sweetness substance solution, it was immersed in the reference solution, and the aftertaste potential V3 was measured.

[0075] The initial taste value R = V2 - V1, and the aftertaste value CPA = V3 - V1.

[0076] In the technical solution of the present invention, the objective error in the detection process can be reduced by using a reference solution, thereby improving the measurement accuracy.

[0077] In one embodiment, in the step of immersing the sensor in the reference solution after cleaning the sweet substance solution: the cleaning is performed by rinsing with deionized water for 10-20 seconds, and the amount of deionized water used is 1-3 mL / s.

[0078] Specifically, during cleaning, positive and negative electrode cleaning solutions are used. The positive and negative electrodes of the probe are immersed in the positive and negative electrode cleaning solutions respectively for 90 seconds, followed by cleaning with a reference solution for 120 seconds, and then cleaning with another reference solution for 120 seconds. After the sensor returns to its equilibrium position for 30 seconds, the sweet substance solution test begins. The positive electrode cleaning solution is composed of potassium chloride, water, ethanol, and potassium hydroxide in a certain proportion, while the negative electrode cleaning solution is composed of water, ethanol, and hydrochloric acid in a certain proportion. The specific cleaning process can be referenced from the cleaning process of the Japanese Insent electronic tongue.

[0079] In the technical solution of this invention, by controlling the probe cleaning time, adjusting the probe position height, and the amount of cleaning solution, it is possible to ensure that most of the sweet substance solution adhering to the sensor is cleaned, and to ensure that the probe is fully immersed in the cleaning solution to remove the adhering substances and impurities on the probe, thereby ensuring the good performance of the electrode and further reducing the detection error of the aftertaste value. If the cleaning time is too short or the amount of cleaning solution is too small, the cleaning degree will be insufficient, and there will still be a large detection error. If the cleaning time is too long or the amount of cleaning solution is too large, the sweet substances or sweet substance molecules bound to the sensor surface may be washed away, and the potential V3 of the aftertaste value will not change much compared with the potential V1 of the reference solution, thus reducing the detection sensitivity.

[0080] In one embodiment, the initial taste values ​​include sweetness, sourness, bitterness, astringency, umami, and saltiness.

[0081] The aftertaste value includes at least one of bitter aftertaste value and astringent aftertaste value;

[0082] It should be noted that richness includes a comprehensive indicator of both of these data.

[0083] In the technical solution of this invention, by employing different sensors, different initial taste values ​​(including sweetness, sourness, bitterness, astringency, saltiness, and umami) and aftertaste values ​​(including aftertaste bitterness and aftertaste astringency) can be measured, and the richness of flavor is also measured. These nine taste attribute indicators greatly broaden the dimensions that can be referenced when screening target sweet substances, providing more options for the screening process. Compared with traditional screening methods, this method, with its detailed and intuitive attribute characteristic values, can quickly identify some target sweet substances that meet specific needs from a vast number of potential sweet substances, enabling further more precise screening.

[0084] In one embodiment, the formula for the first model in step S40 is as follows:

[0085] S E-tongue =β T X E-tongue

[0086] Among them, S E-tongue X represents the electronic tongue sweetness of the sweet substance. E-tongue β is the electronic tongue feature vector of sweet substances. T The weighting coefficient for the target sweetener.

[0087] In one embodiment, the formula for the second model in step S50 is as follows:

[0088] S dock =α T X dock

[0089] Among them, X dock S is the molecular docking feature vector of sweet substances. dock For the molecular docking sweetness of the sweet substance to be tested, α T This is the weight vector of the target sweetener.

[0090] In one embodiment, step S60 includes: obtaining the maximum canonical correlation coefficient ρ by performing canonical correlation analysis on the first information and the second information, and constructing a bimodal joint loss function by combining the first model and the second model. The bimodal joint loss function is as follows:

[0091]

[0092] Among them, S ref For reference sweetness, λ1, λ2, and λ3 are preset weighted hyperparameters.

[0093] In one embodiment, in step S60, the formula for the bimodal feature similarity is as follows:

[0094]

[0095] Where ω is the weighting coefficient and M is the evaluation method for bimodal feature similarity.

[0096] It should be noted that S ref For reference, sweetness generally refers to the sweetness of sucrose, and its value is usually 1.

[0097] In one embodiment, step S80 includes: comparing the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency of the sweet substance and the target sweet substance to obtain a comparison result; comparing the comparison result with preset screening conditions to obtain a screening result; when the comparison result meets the preset screening conditions, it means that the sweet substance can replace the target sweet substance; when the comparison result does not meet the preset screening conditions, it means that the sweet substance cannot replace the target sweet substance.

[0098] It should be noted that the preset filtering criteria must meet the following conditions:

[0099] Sim(T,C)≥θ1;

[0100]

[0101] Where, θ i This is a preset value to ensure bimodal sweetness matching.

[0102] Specifically, for example, let's take sucrose (sweetness = 1) as the target sweet substance T, and set a threshold:

[0103] θ1=0.9, θ2=0.1, θ3=0.1,

[0104] The present invention can recommend alternatives to sweetener C through the following steps using a model:

[0105] 1. Calculate X of sucrose. dock and X E-tongue ;

[0106] 2. Search the database for candidate sweet substances that satisfy Sim(T,C)≥0.9;

[0107] 3. Screen out sweetness errors Candidate sweeteners;

[0108] 4. Output a recommended list (such as isomaltooligosaccharides).

[0109] The present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory; wherein the processor executes the computer program to implement the method for screening sweet substances as described above.

[0110] The technical solution of the present invention will be further described in detail below with reference to specific embodiments. It should be understood that the following embodiments are only used to explain the present invention and are not intended to limit the present invention.

[0111] Example 1

[0112] A method for screening sweeteners includes the following steps:

[0113] Step S10: Use electronic tongue technology to quickly identify the taste characteristics of sample solutions of common sweet substances and obtain the first information of the sweet substances;

[0114] The sweetness sensor and basic taste sensor of the electronic tongue are used to detect various sweet substances. The sensor is sequentially immersed in a reference solution and a prepared sample solution to be tested, and the membrane potential values ​​V1 and V2 are measured respectively. After a brief cleaning with the reference solution, the sensor is immersed in a new reference solution, and V3 is measured. The calculation formula is as follows:

[0115] First taste signal value (R) = V2 - V1,

[0116] The echo signal value (CPA) = V3 - V1.

[0117] Where R represents the initial taste signal value of each sweet substance sample solution, and CPA represents the aftertaste signal value of each sweet substance sample solution. The measured initial and aftertaste signal values ​​are converted into taste information, i.e., taste values, using the analysis software integrated into the electronic tongue. The value measured by the sweetness sensor is used as the sweetness value of the electronic tongue.

[0118] Using different sensors of the electronic tongue, the sweetness sensor GL1 detects the sweetness of each sweet substance, and the basic taste sensors (including CAO, COO, AE1, CTO, and AAE) detect the acidity, bitterness, astringency, afterbitterness, after astringency, saltiness, umami, and richness of each sweet substance.

[0119] Step S20: Perform molecular docking on the sweet substances and use molecular docking technology to reveal the sweetness perception of common sweet substances at the molecular level: the interaction mechanism of each sweet substance monomer with the sweet taste receptor protein TS1R2 / TS1R3 and its binding stability with the sweet taste receptor protein TS1R2 / TS1R3, and obtain the second information of the sweet substances.

[0120] First, the two sweet taste receptor proteins were pretreated: Alpha Fold3 was used to predict the crystal structures of the sweet taste receptor proteins TS1R3 and TS1R2.

[0121] Ligand pretreatment: Schrödinger was used to pretreat the 2D sdf structures of 12 sweet substances to generate all their 3D chiral conformations.

[0122] Objective: Generating all 3D chiral conformations ensures that all possible ligand binding modes are considered during molecular docking, increasing the likelihood of discovering the optimal binding mode. Furthermore, considering the potential differences in biological activity among different chiral conformations, generating and docking all chiral conformations allows for more accurate prediction of ligand biological activity, providing more comprehensive information for screening highly active sweeteners.

[0123] To obtain the active sites of two sweet taste receptor proteins;

[0124] The sweetness of the sweet substances was predicted by molecular docking of the processed sweet substance ligand compounds with the active sites of two sweet taste receptor proteins.

[0125] MM-GBSA calculations were performed, and based on the results (including binding free energy, docking score, number of hydrogen bonds, and presence or absence of hydrophobic interactions), sweet substances with ideal sweetness and flavor were screened out.

[0126] Step S30: Perform principal component analysis and cluster analysis on the obtained electronic tongue measurement data, perform cluster analysis on the molecular docking results, and locate the range of target sweet substances.

[0127] When compared with sucrose at a sweetness level of 5% mass concentration, the degree to which each sweet substance is affected by various taste attributes was observed, identifying sweet substances whose influence on taste attributes is relatively similar to that of sucrose and whose sweetness is similar to that of sucrose. Using molecular docking technology, the docking stability and binding strength of each sweet substance with sweet taste receptor proteins were observed, identifying some sweet substances with good sweetness perception, long sweetness retention time, and relatively high sweetness.

[0128] Step S40: By constructing a first model, perform predictive loss analysis on the first information, and predict the electronic tongue sweetness corresponding to each sweet substance based on machine learning according to the first model.

[0129] Machine learning algorithms are used to extract and analyze the taste characteristic data (including but not limited to sweet, sour, and astringent) of sweet substances obtained through electronic tongue technology. Specifically, partial least squares (PLS) or other suitable machine learning algorithms are used to construct a prediction model S. E-tongue The model uses sucrose sweetness S ref As a baseline sweetness, the electronic tongue sweetness of sweet substances based on machine learning is predicted using the extracted feature vector coefficient β. During model building, the model parameters are optimized by calculating the loss (such as mean squared error) between the predicted and actual sweetness to improve prediction accuracy.

[0130] S50. By constructing a second model, performing predictive loss analysis on the second information, and predicting the molecular docking sweetness of each sweet substance based on machine learning according to the second model.

[0131] Based on the binding data of sweet substances and sweet receptor proteins obtained by molecular docking technology (including but not limited to binding free energy, docking score, number of hydrogen bonds, presence or absence of hydrophobic interactions, etc.), a prediction model S is constructed. dock The model also uses sucrose sweetness S ref As a baseline sweetness, the molecular docking sweetness of sweet substances is predicted by the weighting coefficient α of each input feature. During model construction, the model parameters are optimized by calculating the loss between the predicted sweetness and the actual sweetness (such as mean squared error) to ensure the accuracy of the prediction.

[0132] S60. Construct a bimodal joint loss function by combining the first model and the second model, and evaluate the consistency of bimodal sweetness based on the bimodal joint loss function L.

[0133] Canonical correlation analysis (CCA) was used to establish the correlation between electronic tongue features and molecular docking features, and the CCA projection weight vectors ω1 and ω2 were determined. Based on this correlation, a bimodal joint loss function was constructed, which simultaneously optimizes the correlation between sweetness prediction error and modality. The loss function includes hyperparameters to balance sweetness error and modality correlation. By minimizing this joint loss function, bimodal sweetness consistency, i.e., the degree of similarity between predicted sweetness under the two modalities, is obtained.

[0134] S70. Extract the first and second information of the sweet substance and the target sweet substance respectively, perform similarity analysis, and obtain bimodal feature similarity;

[0135] Electronic tongue features and molecular docking features of the sweetener and the target sweetener are extracted separately. Various similarity evaluation methods (e.g., cosine similarity, Mahalanobis distance, Euclidean distance, etc.) or hybrid similarity evaluation methods (i.e., a balancing strategy of multiple similarity evaluations) are used to calculate the feature similarity between the two modalities. Furthermore, an attention mechanism can be introduced to dynamically adjust the weights of the two modalities, or a hybrid weighting strategy can be adopted to more accurately assess the bimodal feature similarity.

[0136] S80. Compare the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency between the sweet substance and the target sweet substance to obtain the screening results.

[0137] Specifically, based on the established screening rules for alternative sweeteners, bimodal sweetness matching is ensured. The electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency of the sweetener and the target sweetener (specifically sucrose) are compared to obtain the comparison results. The electronic tongue sweetness and molecular docking sweetness of the sweetener are compared with the corresponding values ​​of the target sweetener (sucrose). Simultaneously, considering bimodal feature similarity and bimodal sweetness consistency, the degree of similarity between the sweetener and the target sweetener is comprehensively evaluated. The comparison results can be quantitatively represented using a comprehensive score or ranking method to more intuitively reflect the degree of matching between the sweetener and the target sweetener.

[0138] Finally, by combining electronic tongue technology, molecular docking technology, and a machine learning method based on dual-modal assisted screening of sweet substances, two sweet substances, maltitol and isomaltooligosaccharide, were screened out.

[0139] To further illustrate that the screened maltitol and isomaltooligosaccharide are similar to sucrose in terms of sweetness and taste characteristics, a traditional screening method, namely sensory method, will be used for supplementary explanation.

[0140] Example 1: Screening for alternative sweeteners to sucrose

[0141] (1) Rapid identification of the taste characteristics of solutions containing eight sweet substances, including sucrose, using electronic tongue technology:

[0142] The above eight sweet substances were detected using an electronic tongue's sweetness sensor and a basic taste sensor. The sensor was sequentially immersed in a reference solution and a prepared sample solution to measure the membrane potential values ​​V1 and V2, respectively. After a brief cleaning with the reference solution, the sensor was immersed again in a new reference solution, and V3 was measured. The calculation formula is as follows:

[0143] First taste signal value (R) = V2 - V1,

[0144] The echo signal value (CPA) = V3 - V1.

[0145] Where R represents the initial taste signal value of each sweet substance solution, and CPA represents the aftertaste signal value of the sweet substance solution. The measured initial and aftertaste signal values ​​are converted into specific taste information, i.e., taste values, using the analysis software built into the electronic tongue. The value measured by the sweetness sensor is used as the electronic tongue's sweetness value. Different sensors on the electronic tongue are employed: the sweetness sensor GL1 detects the sweetness of sucrose, and the basic taste sensors (including CAO, COO, AE1, CTO, and AAE) detect the acidity, bitterness, astringency, aftertaste bitterness, aftertaste astringency, saltiness, umami, and richness of these eight sweet substances, as shown in the attached figure. Figure 1 As shown, and the first information is obtained as follows. Figure 2 As shown.

[0146] (2) Using molecular docking technology to reveal the perception of sucrose sweetness at the molecular level: the interaction mechanism between sucrose and the sweet taste receptor protein TS1R2 / TS1R3 and the stability of its binding with the sweet taste receptor protein TS1R2 / TS1R3.

[0147] First, the two sweet taste receptor proteins were pretreated: Alpha Fold3 was used to predict the crystal structures of the sweet taste receptor proteins TS1R3 and TS1R2.

[0148] Ligand pretreatment: Schrödinger was used to pretreat the 2D sdf structures of eight sweet substances to generate their 3D chiral conformations.

[0149] To obtain the active sites of two sweet taste receptor proteins;

[0150] The processed sucrose ligand compound was molecularly docked with the active sites of two sweet taste receptor proteins to predict its sweetness.

[0151] MM-GBSA calculations were performed to obtain the docking fraction, binding free energy, number of hydrogen bonds, and hydrophobic interactions of the aforementioned sweet substances, and to obtain secondary information. See Table 1 for details (using sucrose, maltitol, and isomaltooligosaccharides as examples):

[0152] Table 1. Molecular docking results of sucrose, maltitol, and isomaltooligosaccharide

[0153]

[0154] (3) Principal component analysis and cluster analysis were performed on the electronic tongue measurement data, and cluster analysis was performed on the molecular docking results:

[0155] When compared to sucrose at a sweetness level of 5% by mass, the influence of nine taste attributes on eight sweet substances was observed. Maltitol, isomaltooligosaccharide, and glucose showed relatively small differences in their effects on various taste attributes compared to sucrose, and their sweetness was similar to that of sucrose. Using molecular docking technology, the stability and binding strength of the eight sweet substances to sweet taste receptor proteins were observed. Isomaltooligosaccharide and lactitol showed similarities to sucrose.

[0156] (4) Construct a first model, perform predictive loss analysis on the first information, and predict the electronic tongue sweetness of each sweet substance based on machine learning according to the first model:

[0157] Feature extraction and analysis were performed on taste characteristic data (including but not limited to sweetness, sourness, and astringency) of sweet substances obtained through electronic tongue technology. Partial least squares (PLS) was used to construct a prediction model, SE-tongue. Using sucrose sweetness (Sref) as the benchmark sweetness, the extracted feature vector coefficients β were used to predict...

[0158] The electronic tongue assay for sweetness of sweet substances, based on machine learning, yielded the following results: Figure 3 As shown.

[0159] (5) Construct a second model, perform predictive loss analysis on the second information, and predict the molecular docking sweetness of each sweet substance based on machine learning according to the second model:

[0160] This model also uses sucrose sweetness (Sref) as the baseline sweetness and predicts the molecular docking sweetness of sweet substances based on machine learning by using the weight coefficients α of each input feature. During model construction, the model parameters are optimized by calculating the loss between the predicted and actual sweetness (such as mean squared error) to ensure prediction accuracy. The results are as follows: Figure 4 As shown.

[0161] (6) The canonical correlation analysis (CCA) method was used to establish the correlation between the electronic tongue characteristics and the molecular docking characteristics, with the CCA projection weight vectors ω1 = 0.5 and ω2 = 0.5.

[0162] Constructing a dual-modal joint loss function The correlation between sweetness prediction error and modality is optimized synchronously, and the bimodal sweetness consistency, i.e., the degree of similarity between the predicted sweetness in the two modalities, is obtained by minimizing the joint loss function. Here, λ1, λ2, and λ3 are preset weight hyperparameters, selected according to the actual situation; in this case, λ1 = 1, λ2 = 1, and λ3 = 0.5. The joint function loss value is 0.1391818. The results are as follows... Figure 5 As shown.

[0163] (7) Extract the electronic tongue features and molecular docking features of the sweet substances (sucrose) and the target sweet substances (the other 7 sweet substances) respectively, and calculate the feature similarity between the two modes using cosine similarity. Preset sweetness matching thresholds: θ1 = 0.9, θ2 = 0.3, θ3 = 0.3. After model screening, candidate sweeteners with an overall similarity greater than 0.9 to sucrose characteristics are: maltitol, glucose, and isomaltooligosaccharide. The screening results are as follows: Figure 6 As shown.

[0164] (8) The electronic tongue sweetness and molecular docking sweetness of the sweet substances (7 kinds of sweet substances) were compared with the corresponding values ​​of the target sweet substance (sucrose). Considering the bimodal feature similarity and bimodal sweetness consistency, the similarity between the sweet substances and the target sweet substance was comprehensively evaluated. The results showed that maltitol, glucose and isomaltooligosaccharide had high comprehensive similarity with sucrose.

[0165] In summary, by combining electronic tongue technology, molecular docking technology, and machine learning methods for dual-modal assisted screening of sweet substances, maltitol, glucose, and isomaltooligosaccharide can be selected as sweet substances similar to sucrose.

[0166] (9) The taste characteristics of the selected sweet substances (such as maltitol and isomaltooligosaccharide) were further described using traditional screening methods, namely sensory methods. The test results are shown in Table 2.

[0167] Table 2 Sensory characteristics of sucrose, maltitol and isomaltooligosaccharide

[0168] name Sweetness intensity Sweetness duration / s taste Aftertaste Overall similarity sucrose 6 17.02 5 5 10 Maltitol 7 21.90 4.6 4 8 Isomaltooligosaccharide 5.8 22.18 4.4 3.5 7.6

[0169] Through the detailed sensory evaluation data statistical analysis above, the similarity of the sweet taste of maltitol and isomaltooligosaccharide to sucrose can be verified relatively objectively and accurately, providing strong data support for the screening of sweet substances.

[0170] The specific scoring rules are as follows:

[0171] Sweetness intensity (1-10 points): 1 point indicates that the sweetness is almost imperceptible; 5 points indicates that the sweetness intensity is comparable to the sweetness of a very small amount of sugar in daily drinking water; 10 points indicates that the sweetness intensity is extremely high, far exceeding the sweetness of normal sucrose.

[0172] Sweetness rate (sweetness duration (unit / s): The duration of sweetness in the oral cavity is recorded in real time using a stopwatch;

[0173] Taste (1-5 points): 1 point indicates a rough texture with obvious graininess or other unpleasant taste; 3 points indicates an average taste with no obvious advantages or disadvantages; 5 points indicates a pure and delicate taste with no off-flavors or unpleasant sensations, similar to the taste of sucrose solution.

[0174] Aftertaste (1-5 points): 1 point indicates that there is a noticeable unpleasant taste in the aftertaste, such as bitterness or astringency; 3 points indicates that there is no noticeable unpleasant taste in the aftertaste, but also no noticeable special aftertaste; 5 points indicates that there is a faint sweetness similar to sucrose in the aftertaste, and no other unpleasant taste.

[0175] Overall similarity (1-10 points): 1 point indicates that the taste is completely different from that of sucrose; 5 points indicates that the taste is slightly similar to that of sucrose, but there are still obvious differences; 10 points indicates that the taste is almost completely identical to that of sucrose and is difficult to distinguish.

[0176] Based on the relevant data feature values ​​from the above embodiments, a bimodal model is established between the electronic tongue dataset and the molecular docking dataset to link the two datasets reflecting the characteristics of sweet substances. Finally, by combining electronic tongue technology, molecular docking technology, and a machine learning method based on bimodal assisted screening of sweet substances, sweet substances that can replace the target sweet substance can be accurately screened.

[0177] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A method for screening sweet substances, characterized in that, Includes the following steps: S10. Provide a sweet substance, perform electronic tongue analysis on the sweet substance, and obtain first information about the sweet substance; S20. Molecular docking is performed on the sweet substance, and it is bound to a sweet taste receptor protein to obtain the second information of the sweet substance; S30. Perform principal component analysis and cluster analysis on the first information respectively, and perform cluster analysis on the second information to locate the range of target sweet substances. S40. By constructing a first model, the first information is subjected to predictive loss analysis, and the electronic tongue sweetness is obtained based on the first model; S50. By constructing a second model, the second information is used for predictive loss analysis, and the molecular docking sweetness is obtained based on the second model. S60. Construct a bimodal joint loss function by combining the first model and the second model, and obtain bimodal sweetness consistency based on the bimodal joint loss function; S70. Extract the first and second information of the sweet substance and the target sweet substance respectively, perform similarity analysis, and obtain bimodal feature similarity; S80. Compare the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency between the sweet substance and the target sweet substance to obtain the screening results.

2. The method for screening sweet substances as described in claim 1, characterized in that, The first information includes the taste characteristics of the sweet substance; and / or, The second information includes at least one of binding free energy, docking score, number of hydrogen bonds, and hydrophobic interactions; and / or, The sweet taste receptor proteins include TAS1R2 and / or TAS1R3.

3. The method for screening sweet substances as described in claim 1, characterized in that, The step of performing electronic tongue analysis on the sweet substance to obtain the first information of the sweet substance includes: A solution of a sweetener and a reference solution are provided. The sweetness sensor of the electronic tongue is immersed in the reference solution and the sweetener solution, respectively. The potential V1 of the reference solution and the potential V2 of the sweetener solution are measured. Potential V2; After cleaning the sweetness sensor in the sweetness substance solution, it was immersed in the reference solution, and the aftertaste potential V3 was measured. The initial taste value R = V2 - V1, and the aftertaste value CPA = V3 - V1.

4. The method for screening sweet substances as described in claim 3, characterized in that, The step of immersing the sweetness sensor in the reference solution after cleaning it from the sweetness substance solution: During cleaning, deionized water is used for rinsing, with a rinsing time of 10–20 seconds and a deionized water volume of 1–3 mL / s; and / or, The initial taste value includes at least one of sweetness, sourness, bitterness, astringency, umami, and saltiness, and the aftertaste value includes at least one of bitterness and astringency.

5. The method for screening sweet substances as described in claim 1, characterized in that, The formula for the first model in step S40 is as follows: S E-tongue =b T X E-tongue Among them, S E-tongue X represents the electronic tongue sweetness of the sweet substance. E-tongue β is the electronic tongue feature vector of sweet substances. T The weighting coefficient for the target sweetener.

6. The method for screening sweet substances as described in claim 1, characterized in that, The formula for the second model in step S50 is as follows: S dock =a T X dock Among them, X dock S is the molecular docking feature vector of sweet substances. dock For the molecular docking sweetness of the sweet substance to be tested, α T The weighting coefficient for the target sweetener.

7. The method for screening sweet substances as described in claim 1, characterized in that, Step S60 includes: obtaining the maximum canonical correlation coefficient ρ by performing canonical correlation analysis on the first information and the second information, and constructing a bimodal joint loss function by combining the first model and the second model. The obtained bimodal joint loss function is shown below: Among them, S ref For reference sweetness, λ1, λ2, and λ3 are preset weighted hyperparameters.

8. The method for screening sweet substances as described in claim 1, characterized in that, In step S70, the formula for the bimodal feature similarity is as follows: Where ω is the weighting coefficient and M is the evaluation method for bimodal feature similarity.

9. The method for screening sweet substances as described in claim 1, characterized in that, Step S80 includes: comparing the electronic tongue sweetness, molecular docking sweetness, bimodal feature similarity, and bimodal sweetness consistency of the sweet substance and the target sweet substance to obtain a comparison result; comparing the comparison result with preset screening conditions to obtain a screening result; when the comparison result meets the preset screening conditions, it means that the sweet substance can replace the target sweet substance; when the comparison result does not meet the preset screening conditions, it means that the sweet substance cannot replace the target sweet substance.

10. A computer device, characterized in that, include: Memory; processor; as well as A computer program stored in the memory; wherein the processor executes the computer program to implement the method for screening sweet substances as described in any one of claims 1 to 9.

Images