A method and system for screening oral mucin-adhesive wall materials

Abstract

The present application relates to a kind of oral mucin adhesion type wall material screening method and system, method includes the following steps: establishing candidate wall material database, database includes the name of candidate wall material, the SMILES expression of candidate wall material and wall material attribute information;Based on the SMILES expression of candidate wall material, construct wall material trimer structure, carry out molecular docking with oral mucin receptor, obtain binding energy;According to binding energy, construct adhesion level label according to pre-set uniform threshold;With the molecular descriptor of candidate wall material and wall material attribute information as input feature, with adhesion level label as output label, train machine learning classification model;The trained machine learning classification model is used to predict candidate wall material, and according to the prediction result, generate experiment priority ranking.Compared with prior art, the present application integrates multidimensional technical means, improves molecular docking accuracy, realizes multi-target point differentiation evaluation, improves screening efficiency and reduces research and development cost.

Description

Technical Field

[0001] This invention relates to the fields of oral-grade microcapsule wall material screening, oral adhesive delivery, and artificial intelligence-assisted material design technology, and in particular to a method and system for screening oral mucin-adhesive wall materials. Background Technology

[0002] Oral adhesive wall materials play a crucial role in oral microcapsules, flavor sustained-release, and functional factor delivery systems. The interaction between the wall material and oral mucins affects the residence time, release behavior, and sensory performance of the delivery system in the oral cavity. The oral mucus layer contains various mucins such as MUC1, MUC5B, and MUC7. Different mucins differ in structure, glycosylation level, and interaction sites; therefore, the adhesion ability of the same wall material to different oral mucins may vary. Currently, the screening of oral adhesive wall materials mainly relies on in vitro adhesion experiments, rheological tests, mucin adsorption experiments, and empirical judgment. While these methods can obtain relatively intuitive experimental results, they typically suffer from long experimental cycles, high sample consumption, and low candidate material screening efficiency. Screening candidate wall materials through a step-by-step experimental verification approach usually requires a large number of samples and experimental cycles, significantly increasing R&D costs and failing to meet the needs of rapid early screening. Furthermore, the adhesion ability of wall materials is influenced by a variety of factors, including molecular weight, lipid-water partition coefficient, topological polar surface area, number of hydrogen bond donors, number of hydrogen bond acceptors, number of rotatable bonds, net charge type, functional group type, hydrophilic / hydrophobic characteristics, and whether chemical modification has been performed. Traditional single-factor evaluation methods are insufficient to comprehensively assess their adhesion potential. In recent years, molecular docking and machine learning have provided new technical means for material screening, delivery system optimization, and prediction of biomolecular interactions. Molecular docking can predict the binding trend between wall materials and oral mucin at the molecular level, while machine learning classification models can establish a mapping relationship between candidate materials and adhesion levels using multidimensional features. However, existing technologies still lack a screening method for oral adhesive wall materials that combines wall material trimer-oral mucin molecular docking results, RDKit molecular descriptors, wall material property characteristics, and machine learning classification models.

[0003] Therefore, it is urgent to establish a model that integrates molecular docking results, molecular structure descriptors, and wall material properties to create a classification and prediction model for high adhesion, medium adhesion, and low adhesion. This model would output the experimental priority ranking of candidate wall materials, enabling efficient prediction and ranking of candidate wall materials before wet experiments. This would reduce invalid experiments and provide technical support for the subsequent development of food microcapsules, flavor release systems, and oral delivery systems for functional factors. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method and system for screening oral mucin-adhesive wall materials. This method integrates multi-dimensional technical means to improve molecular docking accuracy, achieve multi-target differential evaluation, improve screening efficiency, and reduce R&D costs.

[0005] The objective of this invention can be achieved through the following technical solutions: This invention provides a method for screening oral mucin-adhesive wall materials, comprising the following steps: S1. Establish a candidate wall material database, which includes the name of the candidate wall material, the SMILES expression of the candidate wall material, and the wall material attribute information; S2. Construct a trimer structure for the wall material based on the SMILES expression of the candidate wall material, and perform molecular docking of the trimer structure with the oral mucin receptor to obtain the binding energy; S3. Based on the binding energy, construct an adhesion level label according to a preset uniform threshold. S4. Using the molecular descriptor and property information of the candidate wall material as input features and the adhesion level label as output label, train a machine learning classification model. S5. Use the trained machine learning classification model to predict candidate wall materials and generate an experimental priority ranking based on the prediction results.

[0006] Furthermore, in step S2, the oral mucin receptor is selected from at least one of MUC1, MUC5B, and MUC7; the binding energy is obtained by molecular docking the wall material trimer with MUC1, MUC5B, and MUC7 respectively.

[0007] Furthermore, in step S2, before molecular docking, the oral mucin receptor structure is preprocessed, including removing irrelevant small molecules, supplementing hydrogen atoms, setting charges, determining docking regions, and generating the format file required for docking.

[0008] Furthermore, in step S3, the unified threshold is as follows: when the binding energy is ≤ -6.0 kcal / mol, it is marked as high adhesion; when -6.0 kcal / mol < binding energy ≤ -4.0 kcal / mol, it is marked as medium adhesion; and when the binding energy is > -4.0 kcal / mol, it is marked as low adhesion.

[0009] Furthermore, in step S4, the molecular descriptor includes molecular weight (MW), lipid-water partition coefficient (LogP), topological polar surface area (TPSA), number of hydrogen bond donors (HBD), number of hydrogen bond acceptors (HBA), and number of rotatable bonds; the wall material property information includes wall material category, net charge type, functional group type, whether it has been modified, and hydrophilic / hydrophobic grade.

[0010] Furthermore, in step S4, the machine learning classification model is selected from one or more of the following: logistic regression model, support vector machine model, random forest model, and extreme random tree model; before model training, the input features are preprocessed, including missing value imputation and standardization of numerical features, and encoding of categorical features.

[0011] Furthermore, in step S4, before model training, the classification performance of different models is compared through cross-validation. Evaluation metrics include macro-F1, weighted-F1, balanced accuracy, high recall and high-F1 for high-adhesion categories, used to determine the model's ability to identify different adhesion levels, especially high-adhesion wall materials.

[0012] Furthermore, in step S5, the prediction results include the predicted adhesion level and high adhesion probability of the candidate wall material relative to different oral mucins; the experimental priority ranking is generated based on the number of candidate wall materials judged as high adhesion, the number of medium adhesion, and the average high adhesion probability.

[0013] Furthermore, step S6 is included: according to the experimental priority ranking, select the candidate wall materials with higher ranking for offline oral mucin adhesion experiments, and compare the verification results with the model prediction results. When there is a difference between the two, the experimental data is backfilled into the candidate wall material database for retraining or optimizing the machine learning classification model.

[0014] The present invention also provides a screening system for oral mucin-adhesive wall materials, comprising: The database module is used to store the names of candidate wall materials, the SMILES expressions of candidate wall materials, and wall material attribute information; The molecular docking module is used to construct a trimer structure of the wall material based on the SMILES expression of the candidate wall material, and to perform molecular docking of the trimer structure of the wall material with the oral mucin receptor to output the binding energy; A tag construction module is used to construct adhesion level tags based on the binding energy and a preset uniform threshold. The model training module is used to train a machine learning classification model with the molecular descriptor and wall material property information of the candidate wall material as input features and the adhesion level label as output label. The sorting output module is used to predict candidate wall materials using the trained machine learning classification model, and generate and output the experimental priority sorting based on the prediction results. The model iteration module is used to receive the verification results of the offline oral mucin adhesion experiment and compare the verification results with the prediction results of the sorting output module. When there is a difference between the two, the experimental data is fed back to the database module and the model training module to achieve iterative optimization of the model.

[0015] Compared with the prior art, the present invention has the following advantages: (1) It integrates multi-dimensional technical means. It organically combines the molecular docking results of wall material trimer-oral mucin, RDKit molecular descriptors, wall material property characteristics and machine learning classification models. This integration not only examines the binding stability of wall material and specific mucin at the molecular micro level (molecular docking), but also comprehensively evaluates it from the perspectives of molecular structural characteristics (such as molecular weight and polarity) and macroscopic physicochemical properties (such as charge type and hydrophilicity / hydrophobicity), overcoming the shortcomings of traditional single-factor evaluation methods that are difficult to comprehensively judge adhesion potential.

[0016] (2) Trimeric structure simulation is adopted to improve the accuracy of molecular docking. Unlike the conventional use of monomers for molecular docking, this scheme constructs a trimer structure of the wall material. This design is closer to the spatial conformation and functional group distribution characteristics of the actual polymer wall material in solution, so that the molecular docking results can more realistically reflect the actual interaction between the wall material and oral mucoprotein, thereby improving the scientificity and accuracy of the prediction.

[0017] (3) Achieving multi-target differential assessment. This approach involves molecular docking and model training for three different oral mucins: MUC1, MUC5B, and MUC7. Due to differences in structure and glycosylation levels among different mucins, the adhesion ability of the same wall material to different mucins may vary. This method can identify and quantify this difference, providing more refined guidance for designing delivery systems with specific site adhesion or controlled release functions.

[0018] (4) Improve screening efficiency and reduce R&D costs. This method, through computer simulation and machine learning models, can quickly pre-screen and prioritize a large number of candidate wall materials before conducting any wet experiments. This significantly shortens the experimental cycle, reduces the consumption of samples and reagents, and effectively avoids the high cost and inefficiency of conducting wet experiments one by one, thus meeting the need for rapid screening in the early stages.

[0019] (5) Constructing a closed-loop iterative mechanism to achieve model self-optimization. This scheme forms a closed-loop process of "computational prediction, candidate screening, wet experiment verification, and model iteration". When there is a difference between the offline wet experiment results and the model prediction, the experimental data will be backfilled into the database for retraining or optimizing the machine learning classification model. This allows the model's screening accuracy to continuously improve with the accumulation of data, forming a continuously evolving intelligent screening system. Attached Figure Description

[0020] Figure 1 A flowchart for screening oral mucin-adhesive wall materials; Figure 2 A flowchart for establishing a candidate wall material database; Figure 3 A flowchart illustrating the molecular docking of the wall material trimer structure with oral mucin receptors; Figure 4 A flowchart for constructing adhesion level labels; Figure 5 A flowchart for training a machine learning classification model and ranking candidate wall materials. Detailed Implementation

[0021] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Component models, material names, connection structures, control methods, algorithms, and other features not explicitly described in this technical solution are considered common technical features disclosed in the prior art.

[0022] Example 1 This embodiment provides a method for screening oral mucin-adhesive wall materials, such as... Figures 1 to 5 As shown, it includes the following steps: S1. Organize food-grade wall materials and establish a candidate wall material database. Each type of wall material in the database corresponds to one record. The database contains the name of the candidate wall material, the SMILES expression of the candidate wall material, and the wall material attribute information. S2. Construct a trimer structure for the wall material based on the SMILES expression of the candidate wall material, and perform molecular docking of the trimer structure with the oral mucin receptor to obtain the binding energy; Specifically, the SMILES expression is subjected to structural analysis, hydrogen atom completion, 3D conformation generation, and energy optimization to obtain the 3D monomer structure of the candidate wall material. Subsequently, based on the repeating unit characteristics of the wall material, the monomer structure is expanded into a wall material trimer structure to more closely approximate the spatial conformation and functional group distribution characteristics of actual wall material molecular chain segments. The generated trimer structure is then conformationally optimized to obtain the wall material ligand structure for molecular docking.

[0023] In a specific embodiment, the oral mucin receptor is selected from at least one of MUC1, MUC5B, and MUC7; the binding energy is obtained by molecular docking the wall material trimer with MUC1, MUC5B, and MUC7 respectively. Before molecular docking, the oral mucin receptor structure is preprocessed, including removing irrelevant small molecules, adding hydrogen atoms, setting charges, determining the docking region, and generating the format file required for docking.

[0024] Subsequently, the optimized wall material trimer structure was molecularly docked with MUC1, MUC5B, and MUC7, respectively, to obtain the binding energy of each candidate wall material relative to different oral mucins. The unit of binding energy is kcal / mol. The more negative the binding energy value, the more stable the predicted binding and the stronger the potential adhesion between the candidate wall material and the corresponding oral mucin.

[0025] S3. Based on the binding energy, construct an adhesion level label according to a preset uniform threshold. Specifically, after obtaining the molecular docking binding energy, adhesion level labels are constructed based on the binding energy distribution of different oral mucins and preset thresholds. Specifically, the binding energies of candidate wall materials relative to MUC1, MUC5B, and MUC7 are classified into high adhesion, medium adhesion, and low adhesion levels, respectively. A uniform binding energy threshold is used to classify the docking results of candidate wall materials with different oral mucins. When the binding energy is less than or equal to -6.0 kcal / mol, it is labeled as high adhesion; when the binding energy is greater than -6.0 kcal / mol and less than or equal to -4.0 kcal / mol, it is labeled as medium adhesion; and when the binding energy is greater than -4.0 kcal / mol, it is labeled as low adhesion. These thresholds are used to calculate the initial label construction in the screening stage, and the threshold range and model parameters can be subsequently corrected based on the results of wet experiments with oral mucins. In this way, each candidate wall material can obtain an independent adhesion level label corresponding to MUC1, MUC5B, and MUC7, which can be used for subsequent classification model training and candidate wall material screening.

[0026] S4. Using the molecular descriptor and wall material property information of the candidate wall material as input features and the adhesion level label as the output label, train the machine learning classification model; the molecular descriptor is used to characterize the size, polarity, hydrogen bond forming ability, and hydrophilicity / hydrophobicity of the wall material molecules, providing input features for the subsequent adhesion prediction model; In specific embodiments, the molecular descriptor includes molecular weight (MW), lipid-water partition coefficient (LogP), topological polar surface area (TPSA), number of hydrogen bond donors (HBD), number of hydrogen bond acceptors (HBA), and number of rotatable bonds; the wall material property information includes wall material category, net charge type, functional group type, whether it is modified, and hydrophilic / hydrophobic grade. Wall material categories may include polysaccharides, modified starches, cyclodextrins, cellulose derivatives, and synthetic or semi-synthetic polymers, etc.

[0027] Before model training, missing values ​​were imputed and standardized for numerical features, and categorical features were encoded. Subsequently, classification models such as Logistic Regression, Support Vector Machine (SVM), Random Forest, and ExtraTrees were used for training, and the classification performance of different models was compared through cross-validation. Evaluation metrics included macro-F1 score, weighted-F1 score, balanced accuracy, high recall for high-adhesion categories, and high-F1 score for high-adhesion categories, used to assess the model's ability to identify different adhesion levels, especially high-adhesion wall materials.

[0028] After model training, the predicted adhesion levels and high adhesion probabilities of candidate wall materials relative to MUC1, MUC5B, and MUC7 are output. Further analysis is performed on the number of wall materials classified as highly adhesive and moderately adhesive among the three oral mucins. This data, combined with the model's predicted average high adhesion probability, generates an experimental priority ranking. The top-ranked candidate wall materials can be prioritized for subsequent offline oral mucin adhesion experiments.

[0029] S5. Use the trained machine learning classification model to predict candidate wall materials and generate an experimental priority ranking based on the prediction results.

[0030] In a specific implementation, candidate wall materials ranked high were selected for offline validation. These candidate wall materials included α-cyclodextrin, γ-cyclodextrin, sodium alginate, and polystyrene sulfonate. Oral mucin adhesion experiments were conducted on the selected wall materials to evaluate their actual interaction capabilities with oral mucins such as MUC1, MUC5B, and MUC7.

[0031] During wet adhesion testing, methods such as mucin adsorption measurement, fluorescently labeled mucin adsorption experiments, rheological analysis, QCM-D, or other in vitro adhesion evaluation methods can be used to obtain the actual adhesion results of candidate wall materials. Multiple parallel samples are set up for each wall material, and the mean and standard deviation of the experimental results are calculated to improve the reliability of the results.

[0032] In a specific implementation, step S6 is also included: according to the experimental priority ranking, select the candidate wall materials with higher ranking for offline oral mucin adhesion experiments, and compare the verification results with the model prediction results. When there is a difference between the two, the experimental data is backfilled into the candidate wall material database for retraining or optimizing the machine learning classification model.

[0033] The adhesion results obtained from offline wet experiments are compared with the high, medium, or low adhesion levels predicted by the model. When the experimental results are consistent with the model predictions, the results are used as the basis for validating the model's effectiveness; when the experimental results differ from the model predictions, the experimental data are backfilled into the wall material database and used to retrain or optimize the machine learning classification model. Through this method, a closed-loop process of computational prediction, candidate screening, wet experimental verification, and model iteration is formed, thereby gradually improving the model's accuracy in screening oral adhesive wall materials.

[0034] The core mechanism lies in predicting the potential adhesion ability of wall materials by utilizing their molecular structure characteristics and the interaction between wall material and oral mucins. Hydroxyl, carboxyl, amino, sulfate, and hydrophobic groups in the wall material can interact with oral mucins such as MUC1, MUC5B, and MUC7 through hydrogen bonding, electrostatic interactions, and hydrophobic interactions, thereby affecting their oral adhesion ability. Based on the SMILES expression of candidate wall materials, a trimer structure of the wall material is constructed, and molecular docking is performed with MUC1, MUC5B, and MUC7 respectively to obtain the binding energy. The more negative the binding energy, the more stable the predicted binding between the wall material and the corresponding mucin. Furthermore, by combining RDKit molecular descriptors and wall material property characteristics, high-adhesion, medium-adhesion, and low-adhesion classification models are established to learn the relationship between wall material structural characteristics and adhesion levels. Through this method, candidate wall materials can be pre-screened before wet experiments, and experimental priority ranking can be generated based on the predicted adhesion levels of different mucins and the model's predicted probabilities. Subsequent wet experiment results can be used for model iterative optimization, thereby gradually improving the accuracy of screening oral adhesive wall materials.

[0035] Example 2 This embodiment provides a screening system for oral mucin-adhesive wall materials, including: The database module is used to store the names of candidate wall materials, the SMILES expressions of candidate wall materials, and wall material attribute information; The molecular docking module is used to construct a trimer structure of the wall material based on the SMILES expression of the candidate wall material, and to perform molecular docking of the trimer structure of the wall material with the oral mucin receptor to output the binding energy; A tag construction module is used to construct adhesion level tags based on the binding energy and a preset uniform threshold. The model training module is used to train a machine learning classification model with the molecular descriptor and wall material property information of the candidate wall material as input features and the adhesion level label as output label. The sorting output module is used to predict candidate wall materials using the trained machine learning classification model, and generate and output the experimental priority sorting based on the prediction results. The model iteration module is used to receive the verification results of the offline oral mucin adhesion experiment and compare the verification results with the prediction results of the sorting output module. When there is a difference between the two, the experimental data is fed back to the database module and the model training module to achieve iterative optimization of the model.

[0036] Components not described in detail in this embodiment are all existing components that can be purchased through public channels.

[0037] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for screening oral mucin-adhesive wall materials, characterized in that, Includes the following steps: S1. Establish a candidate wall material database, which includes the name of the candidate wall material, the SMILES expression of the candidate wall material, and the wall material attribute information; S2. Construct a trimer structure for the wall material based on the SMILES expression of the candidate wall material, and perform molecular docking of the trimer structure with the oral mucin receptor to obtain the binding energy; S3. Based on the binding energy, construct an adhesion level label according to a preset uniform threshold. S4. Using the molecular descriptor and property information of the candidate wall material as input features and the adhesion level label as output label, train a machine learning classification model. S5. Use the trained machine learning classification model to predict candidate wall materials and generate an experimental priority ranking based on the prediction results.

2. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S2, the oral mucin receptor is selected from at least one of MUC1, MUC5B and MUC7; the binding energy is obtained by molecular docking the wall material trimer with MUC1, MUC5B and MUC7 respectively.

3. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S2, before molecular docking, the oral mucin receptor structure is preprocessed, including removing irrelevant small molecules, adding hydrogen atoms, setting charges, determining the docking region, and generating the format file required for docking.

4. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S3, the unified threshold is as follows: when the binding energy is ≤ -6.0 kcal / mol, it is marked as high adhesion; when -6.0 kcal / mol < binding energy ≤ -4.0 kcal / mol, it is marked as medium adhesion; when the binding energy is > -4.0 kcal / mol, it is marked as low adhesion.

5. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S4, the molecular descriptor includes molecular weight (MW), lipid-water partition coefficient (LogP), topological polar surface area (TPSA), number of hydrogen bond donors (HBD), number of hydrogen bond acceptors (HBA), and number of rotatable bonds; the wall material property information includes wall material category, net charge type, functional group type, whether it has been modified, and hydrophilic / hydrophobic grade.

6. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S4, the machine learning classification model is selected from one or more of the following: logistic regression model, support vector machine model, random forest model, and extreme random tree model. Before model training, the input features are preprocessed, including imputing missing values ​​and standardizing numerical features, and encoding categorical features.

7. The method for screening oral mucin-adhesive wall materials according to claim 6, characterized in that, In step S4, before model training, the classification performance of different models is compared through cross-validation. The evaluation metrics include macro-F1, weighted-F1, balanced accuracy, high recall and high-F1 of the high-adhesion categories, which are used to judge the model's performance for different levels of adhesion.

8. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, In step S5, the prediction results include the predicted adhesion level and high adhesion probability of the candidate wall material relative to different oral mucins; the experimental priority ranking is generated based on the number of candidate wall materials judged to be highly adhesive, the number of medium adhesive, and the average high adhesion probability.

9. The method for screening oral mucin-adhesive wall materials according to claim 1, characterized in that, The method also includes step S6: according to the experimental priority ranking, select the candidate wall materials with higher ranking for offline oral mucin adhesion experiments, and compare the verification results with the model prediction results. When there is a difference between the two, the experimental data is backfilled into the candidate wall material database for retraining or optimizing the machine learning classification model.

10. A screening system for oral mucin-adhesive wall materials, used to implement the screening method for oral mucin-adhesive wall materials according to any one of claims 1-9, characterized in that, include: The database module is used to store the names of candidate wall materials, the SMILES expressions of candidate wall materials, and wall material attribute information; The molecular docking module is used to construct a trimer structure of the wall material based on the SMILES expression of the candidate wall material, and to perform molecular docking of the trimer structure of the wall material with the oral mucin receptor to output the binding energy; A tag construction module is used to construct adhesion level tags based on the binding energy and a preset uniform threshold. The model training module is used to train a machine learning classification model with the molecular descriptor and wall material property information of the candidate wall material as input features and the adhesion level label as output label. The sorting output module is used to predict candidate wall materials using the trained machine learning classification model, and generate and output the experimental priority sorting based on the prediction results. The model iteration module is used to receive the verification results of the offline oral mucin adhesion experiment and compare the verification results with the prediction results of the sorting output module. When there is a difference between the two, the experimental data is fed back to the database module and the model training module to achieve iterative optimization of the model.

Images