A method and apparatus for targeted vaccine construction

Abstract

The application provides a targeted vaccine construction method and device, and relates to the technical field of biological medicine. The method comprises the following steps: obtaining three-dimensional structure data of an antibody-antigen complex; converting an antigen part of the three-dimensional structure data into an antigen residue layer level diagram; training an artificial intelligence model through the antigen residue layer level diagram to obtain an epitope recognition model; processing a target target point through the epitope recognition model to obtain a target epitope; and constructing a targeted vaccine of the target target point based on the target epitope, so as to overcome the deficiencies of the prior art in antigen recognition accuracy and vaccine development whole-process connection.

Description

Technical Field

[0001] This application relates to the field of biomedical technology, and more specifically, to a method and apparatus for constructing a targeted vaccine. Background Technology

[0002] In the development of targeted vaccines, accurately identifying antigen fragments on the target protein that can be specifically recognized by the immune system is a crucial prerequisite for determining whether a candidate vaccine can effectively induce a humoral immune response. Currently, while computational prediction methods for identifying these antigen fragments have been gradually applied to early-stage high-throughput screening, general methods still have limitations in processing protein three-dimensional conformational information, characterizing local geometric relationships between residues, and addressing class imbalances in residue hierarchical classification tasks. Even after obtaining candidate antigen fragments, efficiently constructing them into sufficiently immunogenic vaccine molecules and achieving stable in vivo delivery remains a significant challenge hindering the transition of targeted vaccines from screening to practical application.

[0003] Existing technologies have not yet formed an integrated technical solution that connects all the above-mentioned links. Summary of the Invention

[0004] The problem addressed in this application is how to provide a targeted vaccine development scheme that can effectively integrate antigen structural feature analysis with downstream formulation construction, so as to overcome the shortcomings of existing technologies in terms of antigen recognition accuracy and the connection of the entire vaccine development process.

[0005] To address the aforementioned issues, this application provides a method and apparatus for constructing a targeted vaccine.

[0006] Firstly, this application provides a method for constructing a targeted vaccine, including: Obtain three-dimensional structural data of the antibody-antigen complex; The antigen portion of the three-dimensional structural data is converted into an antigen residue layer diagram. An artificial intelligence model is trained using antigen residue layer maps to obtain an epitope recognition model. The target epitope is obtained by processing the target point through the epitope identification model; A targeted vaccine for the target epitope is constructed based on the target epitope.

[0007] Optionally, converting the antigen portion of the three-dimensional structural data into an antigen residue layer map includes: Each antigen residue corresponds to a node in the antigen residue layer diagram, and the spatial proximity between residues corresponds to an edge in the antigen residue layer diagram.

[0008] Optionally, converting the antigen portion of the three-dimensional structural data into an antigen residue layer map further includes: Node features are constructed based on the amino acid type information and local structural description information of the antigen portion; Edge features are constructed based on the Euclidean distance between residues of the antigen portion, the normalized three-dimensional direction vector, and the relative sequence position difference; The antigen residue layer map is constructed based on the node features and the edge features.

[0009] Optionally, training the artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model includes: By using the linear layer in the artificial intelligence model, node features are mapped to the hidden space to obtain the initial node representation, and edge features are geometrically encoded to obtain the edge geometric representation. The attention node representation is obtained by message passing between the initial node representation and the edge geometric representation through the graph convolution module; The neighborhood geometry encoding module obtains a geometry-enhanced node representation based on the attention node representation and the edge geometry representation; The attention node representation and the geometric enhancement node representation are adaptively weighted and fused through the gating fusion module to obtain the final node representation for epitope classification. The loss is calculated based on the final node representation and the real label, and the artificial intelligence model is trained to obtain the epitope recognition model.

[0010] Optionally, obtaining the geometrically enhanced node representation based on the attention node representation and the edge geometric representation through the neighborhood geometric encoding module includes: In the attention node representation, the geometric representation of the edge corresponding to the incoming edge of each node is averaged and aggregated to obtain the aggregation result; The aggregation result is concatenated with the node representation and then input into a multilayer perceptron to obtain the geometrically enhanced node representation.

[0011] Optionally, training the artificial intelligence model to obtain the epitope recognition model based on the loss calculated from the final node representation and the real label includes: A stratified sampling strategy was adopted to divide the training set and the validation set at the antigen residue level graph level. During training, node-based balanced batch sampling is performed on the training set, wherein positive residue samples are oversampled and negative residue samples are downsampled, and the validation set maintains the original class distribution; The loss is calculated using a pre-constructed hybrid loss function; The artificial intelligence model is trained to obtain the epitope recognition model.

[0012] Optionally, the construction of the targeted vaccine based on the target epitope includes: The target epitope is combined with an immune-enhancing vector to construct the targeted vaccine for the target target, wherein the immune-enhancing vector includes a carrier protein, virus-like particles, and nanoparticles.

[0013] Optionally, the targeted vaccine construction method further includes: Obtain exosomes; The target epitope, the binding product of the target epitope and the immune enhancement vector, and the exosomes are mixed to obtain a mixture; The mixture is screened, and the screened mixture is used as a targeted delivery formulation.

[0014] Optionally, the target site includes the angiotensin II type 1 receptor, extracellular region, functional fragment, or mutant encoded by the AGTR1 gene; The targeted vaccine or targeted delivery formulation is used to prepare metabolic regulatory substances.

[0015] Secondly, this application provides a targeted vaccine construction apparatus, comprising: The structural data acquisition module is used to acquire the three-dimensional structural data of the antibody-antigen complex; The graph construction module is used to convert the antigen portion of the three-dimensional structural data into an antigen residue layer map. The training module is used to train an artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model. The prediction module is used to process the target point through the epitope identification model to obtain the target epitope; A construction module is used to construct a targeted vaccine for the target target based on the target epitope.

[0016] The beneficial effects of the targeted vaccine construction method in this application are: Using the antibody-antigen binding conformation as input, the spatial coordinate information of the antigen components is extracted and discretized into a graph structure representation with amino acid residues as nodes and spatial proximity or interaction relationships as edges. This representation method eliminates atomic-level redundant details while preserving key topological and geometric relationships, effectively avoiding the problem of potentially missing conformation-dependent recognition sites. An artificial intelligence model is trained using the antigen residue-level graph to obtain an epitope recognition model. The graph structure itself encodes local microenvironmental features in non-Euclidean space, enabling the model to internalize conformation-sensitive recognition rules; the modeling approach using residues as basic units matches the task granularity of epitope prediction, avoiding information distortion caused by cross-scale transformation. The residue-level graph of a specific target is input into the trained epitope recognition model to obtain the target epitope, and this is used as the core immunogen to construct a targeted vaccine. Since the prediction results can be directly mapped to specific residue positions, and the selected epitopes originate from the real complex structure and retain key recognition interfaces, it is more likely to induce neutralizing antibodies with functional regulatory activity. This achieves a connection from structural feature analysis to formulation construction, overcoming the shortcomings in antigen recognition accuracy and the coordination of the entire vaccine development process. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the targeted vaccine construction method according to an embodiment of this application; Figure 2 This is an example diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, specific embodiments of this application are described in detail below with reference to the accompanying drawings. Although some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this application. It should be understood that the accompanying drawings and embodiments of this application are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0019] It should be understood that the steps described in the method embodiments of this application may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this application is not limited in this respect.

[0020] The term "comprising" and its variations as used herein are open-ended, meaning "including but not limited to"; the term "based on" means "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; and the term "optionally" means "optional embodiments". Definitions of other terms will be given in the description below. It should be noted that the concepts of "first," "second," etc., mentioned in this application are used only to distinguish different devices, modules, or units, and are not intended to limit the order of functions performed by these devices, modules, or units or their interdependencies.

[0021] It should be noted that the terms "one" and "more" used in this application are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0022] The names of messages or information exchanged between multiple devices in the embodiments of this application are for illustrative purposes only and are not intended to limit the scope of these messages or information.

[0023] like Figure 1 As shown in the embodiment of this application, a method for constructing a targeted vaccine includes: Step S100: Obtain the three-dimensional structural data of the antibody-antigen complex.

[0024] The three-dimensional structural data of antibody-antigen complexes refers to a digital model obtained through experimental determination or computational simulation, which characterizes the spatial coordinates and relative positions of each atom of the antibody molecule and the antigen molecule in the bound state. It is used to visually present the geometric morphology of the interaction interface and the distribution of contact residues.

[0025] In one embodiment, three-dimensional structural data of antibody-antigen complexes are obtained from a structural database, and the antigenic portion is extracted as the object for subsequent graph construction. Since the formation of B-cell epitopes depends not only on amino acid composition but also on residue spatial proximity, surface exposure, and local conformation, using antigen structures as input allows the model to directly learn the real spatial information related to epitope formation. By standardizing and uniformly representing the antigen structures, a structural dataset suitable for graph neural network modeling can be obtained for subsequent training and prediction.

[0026] Step S200: Convert the antigen portion of the three-dimensional structural data into an antigen residue layer map.

[0027] Antigen residue-level graphs represent a graph structure where each amino acid residue in the antigen protein is a node, and the spatial proximity or interaction between residues is an edge. This method discretizes continuous three-dimensional coordinate information into a topological data structure that can be directly processed by computational models. It abstracts the complex spatial conformation, which originally relied on atomic-level coordinates, into a residue network that preserves key geometry and adjacency relationships, allowing subsequent analysis to focus on residue-scale features rather than redundant atomic details.

[0028] Graph structures adapt to the non-Euclidean spatial relationships between residues in protein molecules, effectively encoding local microenvironments and long-range dependencies, avoiding the information loss and computational overhead introduced by gridding or voxelization methods. Modeling with residues as the basic unit reduces data dimensionality, improves computational efficiency, and maintains consistency with biological functional units, allowing the resulting graph representation to directly serve downstream tasks such as epitope identification without additional scale alignment operations.

[0029] Step S300: Train the artificial intelligence model using the antigen residue layer map to obtain the epitope recognition model.

[0030] Epitope identification models are computational models that, after learning, can automatically determine whether each residue belongs to a B-cell epitope based on the input antigen residue map structure. They transform complex structure-function mapping relationships into reusable predictive capabilities, thus replacing traditional methods that rely on manual feature screening. The model receives a new antigen residue layer map as input and outputs the probability or label of each residue corresponding to an epitope category, used to select candidate antigen fragments.

[0031] Graph structures preserve the spatial topological and local geometric information between residues, enabling the model to autonomously learn structures related to immune recognition during training. The residue-node representation aligns with the granularity requirements of epitope prediction tasks, avoiding information distortion during transitions to the residue-level hierarchy. The graph-based training framework facilitates the integration of geometric attributes such as distance and orientation on edges, improving the model's ability to distinguish conformation-dependent epitopes and maintaining stable prediction performance even under conditions of high imbalance between positive and negative samples. This enhances the applicability and generalization potential of the resulting model in practical vaccine design scenarios.

[0032] Step S400: Process the target point through the epitope identification model to obtain the target epitope.

[0033] Targets refer to specific protein molecules used for vaccine development, such as angiotensin II type 1 receptors and other antigenic proteins related to obesity metabolic regulation; target epitopes refer to the set of antigen fragments that are predicted by the model to be recognizable. This is used to transfer the computational power gained from previous training to specific application scenarios, enabling the application of knowledge from a general model to specific targets.

[0034] The residue-level map of the target site is input into a trained epitope identification model, which automatically outputs the determination results of each residue belonging to the epitope category, and then integrates them to form a candidate target epitope list. Compared with traditional methods, the model has internalized the spatial topology and local geometry information between residues during the training phase, thus having a stronger ability to identify conformation-dependent epitopes; the processing flow does not require manual intervention in feature extraction, and can maintain robustness to scenarios with imbalanced positive and negative samples while maintaining high throughput, thereby improving screening efficiency. The obtained target epitopes directly correspond to residue-level structural units, which facilitates subsequent integration with vectors, stability assessment, and delivery formulation design, thereby supporting the development of targeted weight-loss vaccines.

[0035] Step S500: Construct a targeted vaccine for the target epitope based on the target epitope.

[0036] In one embodiment, the target epitope refers to an antigen fragment with high prediction scores, good spatial accessibility, and strong application potential, selected by the Geo-E2T model after prediction and scoring of the AGTR1 target region, and used as a core unit to induce a specific immune response. The targeted vaccine refers to a biological agent with this target epitope as a key active component, designed to elicit a directed immune response against the AGTR1 target.

[0037] In practice, the selected AGTR1 target epitope is used as the antigen core. The corresponding polypeptide sequence is obtained through chemical synthesis or genetic engineering. It is then coupled with a suitable carrier protein or embedded in a delivery vector, supplemented with adjuvants and pharmaceutically acceptable excipients. After purification, formulation and quality control steps, a targeted vaccine formulation that can be used for subsequent immunization evaluation is prepared.

[0038] In this embodiment, the spatial coordinate information of the antigen component is extracted from the antibody-antigen binding conformation and discretized into a graph structure with amino acid residues as nodes and spatial proximity or interaction relationships as edges. This representation method eliminates atomic-level redundant details while preserving key topological and geometric relationships, effectively avoiding the problem of missing conformation-dependent recognition sites. An artificial intelligence model is trained using the antigen residue-level graph to obtain an epitope recognition model. The graph structure itself encodes local microenvironmental features in non-Euclidean space, enabling the model to internalize conformation-sensitive recognition rules; the modeling method using residues as basic units matches the task granularity of epitope prediction, avoiding information distortion caused by cross-scale conversion. The residue-level graph of a specific target is input into the trained epitope recognition model to obtain the target epitope, and this is used as the core immunogen to construct a targeted vaccine. Since the prediction results can be directly mapped to specific residue positions, and the selected epitopes originate from the real complex structure and retain key recognition interfaces, it is more likely to induce neutralizing antibodies with functional regulatory activity. This achieves a connection from structural feature analysis to formulation construction, overcoming the shortcomings in antigen recognition accuracy and the coordination of the entire vaccine development process.

[0039] Optionally, converting the antigen portion of the three-dimensional structural data into an antigen residue layer map includes: Each antigen residue corresponds to a node in the antigen residue layer diagram, and the spatial proximity between residues corresponds to an edge in the antigen residue layer diagram.

[0040] The continuous spatial coordinates of proteins are discretized into a computable graph data structure to provide standardized input that takes into account both local details and overall topology. In practice, each amino acid residue in the antigen sequence is treated as a node in the graph, carrying the sequence attributes and local structural attributes of the corresponding residue. Simultaneously, edges are established between nodes based on the proximity of residues in three-dimensional space, describing the geometric relationships and relative positional information between different residues. Through this modeling approach, the antigen residue-level graph can simultaneously preserve the individual characteristics of each residue and the interactions between residues. This allows subsequent epitope identification models to learn the properties of the residues themselves and utilize spatial correlation information for comprehensive judgment, thus fully reflecting the structural characteristics of the antigen in its bound state.

[0041] Optionally, converting the antigen portion of the three-dimensional structural data into an antigen residue layer map further includes: Node features are constructed based on the amino acid type information and local structural description information of the antigen portion; Edge features are constructed based on the Euclidean distance between residues of the antigen portion, the normalized three-dimensional direction vector, and the relative sequence position difference; The antigen residue layer map is constructed based on the node features and the edge features.

[0042] Node features are constructed based on amino acid type information and local structural description information of the antigen moiety. Amino acid type information reflects the differences in the chemical properties of residues, while local structural description information represents the conformational state of the residue in three-dimensional space. Edge features are constructed based on Euclidean distance between residues, normalized three-dimensional direction vectors, and relative sequence position differences. Euclidean distance quantifies the spatial proximity between residues, normalized three-dimensional direction vectors describe the relative orientation between residues, and relative sequence position differences preserve the topological associations on the primary sequence. Integrating node and edge features forms a complete antigen residue-level graph. In this way, the resulting graph structure simultaneously encodes the physicochemical and conformational properties of the residues themselves, as well as the multiple spatial and sequence-level associations between residues, providing rich and structured input information for the epitope identification model.

[0043] In specific implementation, each antigen residue node is represented by a feature vector containing amino acid identity information and local structural description information. Preferably, the node feature is a 25-dimensional vector, including 20-dimensional one-hot encoding of amino acids and 5 local structural descriptors. The 5 local structural descriptors include the number of heavy atoms, the average B factor, the root mean square radius of the residue atom relative to the centroid, the local residue density within a 5 Å range, and surface exposure indication information. Through the above design, properties such as residue category, spatial compactness, local flexibility, and surface state can be characterized simultaneously at the node level.

[0044] For establishing graph edges, when the Cα–Cα distance between two residues is less than or equal to 5 Å, a bidirectional edge is established between the corresponding nodes. Edge features are used to characterize the geometric relationships between residues, preferably including Euclidean distance, normalized three-dimensional direction vector, and relative sequence position difference. Instead of explicitly introducing edge features into the model, the geometric directionality and local spatial dependence between residues are further encoded. Integrating the above node features with edge features forms a complete antigen residue layer map, providing rich and structured input information for the epitope recognition model.

[0045] Optionally, training the artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model includes: By using the linear layer in the artificial intelligence model, node features are mapped to the hidden space to obtain the initial node representation, and edge features are geometrically encoded to obtain the edge geometric representation. The attention node representation is obtained by message passing between the initial node representation and the edge geometric representation through the graph convolution module; The neighborhood geometry encoding module obtains a geometry-enhanced node representation based on the attention node representation and the edge geometry representation; The attention node representation and the geometric enhancement node representation are adaptively weighted and fused through the gating fusion module to obtain the final node representation for epitope classification. The loss is calculated based on the final node representation and the real label, and the artificial intelligence model is trained to obtain the epitope recognition model.

[0046] Linear layers map node features to the latent space to generate initial node representations, while geometrically encoding edge features to obtain edge geometric representations, transforming the raw input into an intermediate form suitable for subsequent processing. A graph convolution module receives the initial node representations and edge geometric representations, performs message passing operations, and outputs attention node representations, used to aggregate neighborhood information and capture topological relationships between residues. A neighborhood geometry encoding module generates geometrically enhanced node representations based on the attention node representations and edge geometric representations, supplementing geometric details such as orientation and distance in the local 3D structure. A gated fusion module performs adaptive weighted fusion of the attention node representations and geometrically enhanced node representations to obtain the final node representation for epitope classification. This allows for dynamic adjustment of the contribution ratio of the two types of information according to the characteristics of different nodes, avoiding the limitations of a single representation. During the training phase, the loss value is calculated based on the final node representation and the ground truth label, and the model parameters are updated accordingly. This enables the AI ​​model to gradually learn the ability to distinguish epitopes from the antigen residue level graph, ultimately converging into an epitope recognition model with predictive capabilities.

[0047] In one embodiment, the artificial intelligence model employs an edge-aware graph Transformer architecture. This model explicitly integrates edge geometric information into the attention message passing process, enabling it to consider both node representations and the actual spatial relationships between residues when aggregating neighborhood information. Specifically, the AI ​​model maps node features to the hidden space via linear layers to obtain initial node representations, while simultaneously using a multilayer perceptron to encode edge features geometrically. The encoded node and edge representations are then input into a three-layer Transformer-based graph convolutional module for message passing. The first two layers employ a multi-head attention mechanism and feature concatenation. Normalization and dropout are then applied sequentially after each layer to enhance training stability and mitigate overfitting. The third layer outputs attention node representations with fixed hidden dimensions for subsequent classification. The graph convolutional module directly injects edge features into the attention weight calculation, combining the attention level with the distance, direction, and relative sequence position information carried by the edges. This effectively identifies residue regions that are geographically distant in the primary sequence but are adjacent in three-dimensional space and collectively constitute the epitope microenvironment.

[0048] To further enhance the representation of local geometric microenvironments, a neighborhood geometric encoding module is introduced. This module performs mean aggregation on the geometric features of the incoming edges of each node, and then concatenates the aggregation result with the current attention node representation before processing it through a multilayer perceptron to obtain a geometrically enhanced node representation. This allows for the explicit integration of spatial organization information of neighborhood edges at the node level. The gating fusion module adaptively weights and fuses the geometrically enhanced node representation and the attention node representation to generate the final node representation for epitope classification.

[0049] Preferably, the gating coefficient is jointly determined by the geometrically enhanced representation and surface exposure indication information, calculated using the sigmoid activation function. This allows the two feature streams to dynamically adjust their weights according to the local environment of the residues, balancing the contributions of geometric information and contextual semantic information. During the training phase, the loss value is calculated based on the final node representation and the true label, and the model parameters are iteratively updated. This enables the edge-aware graph Transformer to gradually learn the ability to distinguish B cell epitopes from the antigen residue level map, ultimately converging into an epitope recognition model with predictive capabilities.

[0050] Optionally, obtaining the geometrically enhanced node representation based on the attention node representation and the edge geometric representation through the neighborhood geometric encoding module includes: In the attention node representation, the geometric representation of the edge corresponding to the incoming edge of each node is averaged and aggregated to obtain the aggregation result; The aggregation result is concatenated with the node representation and then input into a multilayer perceptron to obtain the geometrically enhanced node representation.

[0051] The incoming edges represent directed connections to the current node in the antigen residue level graph, and the corresponding edge geometry representation carries the spatial orientation and distance information of adjacent residues relative to the current node; mean aggregation represents the operation of calculating the arithmetic mean of all incoming edge geometry representations, used to compress discrete neighborhood geometry cues into a unified local structure descriptor; concatenation represents linking the aggregation result with the original node representation in the feature dimension to form a joint vector that contains both semantic and geometric information; the multilayer perceptron represents a feedforward network composed of several fully connected layers and nonlinear activation functions, used to perform nonlinear transformation and feature extraction on the concatenated joint vector.

[0052] This approach integrates the representation of each node with the true 3D arrangement of surrounding residues, thereby enhancing the model's ability to perceive the microenvironment of conformation-dependent epitopes. Specifically, for each node in the attention node representation, the geometric representations of all its incoming edges are extracted and their mean values ​​are calculated to obtain the aggregation result for that node. This aggregation result is then concatenated with the node's own attention node representation along the feature axis and input into a pre-defined multilayer perceptron for forward propagation. The output is the geometrically enhanced node representation.

[0053] Optionally, training the artificial intelligence model to obtain the epitope recognition model based on the loss calculated from the final node representation and the real label includes: A stratified sampling strategy was adopted to divide the training set and the validation set at the antigen residue level graph level. During training, node-based balanced batch sampling is performed on the training set, wherein positive residue samples are oversampled and negative residue samples are downsampled, and the validation set maintains the original class distribution; The loss is calculated using a pre-constructed hybrid loss function; The artificial intelligence model is trained to obtain the epitope recognition model.

[0054] Stratified sampling strategy means that when splitting the dataset, the samples are grouped according to the antigen residue layer map as a whole, so that the ratio of positive to negative samples in the training set and validation set is as close as possible to the original distribution, in order to avoid class bias caused by data splitting.

[0055] Node-based balanced batch sampling dynamically adjusts the number of positive and negative samples within each training batch. Positive residue samples are oversampled to increase their frequency of occurrence, while negative residue samples are downsampled to reduce redundant information. The validation set always maintains the original class distribution, which is used to alleviate the inherent class imbalance problem in B-cell epitope prediction tasks and prevent the model from being biased towards the majority class.

[0056] The hybrid loss function is composed of focus loss and class balance loss. The former reduces the weight of easily classified samples and focuses the learning of difficult-to-classify samples by modulating the factor, while the latter inversely weights the samples according to the class frequency to compensate for the difference in the number of samples. The two work together to further weaken the learning bias caused by class imbalance.

[0057] In practice, the antigen residue layer map dataset is first divided into training and validation sets using a stratified sampling strategy. During the training phase, node-based balanced batch sampling is applied to the training set at the time of each batch generation, while the validation set does not participate in any resampling operations. After each round of forward propagation, the loss value of the current batch is calculated using a hybrid loss function, and the model parameters are updated accordingly. Training continues until the preset number of rounds is completed or the early stopping condition is met, during which the parameters with the best validation performance are retained as the weights of the final epitope recognition model. The entire process improves the stability and generalization ability of the model in highly imbalanced scenarios through three methods: data partitioning, sampling control, and loss design.

[0058] Optionally, the construction of the targeted vaccine based on the target epitope includes: The target epitope is combined with an immune-enhancing vector to construct the targeted vaccine for the target target, wherein the immune-enhancing vector includes a carrier protein, virus-like particles, and nanoparticles.

[0059] Immunostimulators are auxiliary structures that can enhance antigen immunogenicity, promote antigen presentation, or prolong in vivo retention time. They are used to compensate for the lack of self-immune stimulation ability of small molecule peptides, enhance the humoral immune response induced by vaccines, and improve the ability of AGTR1 to target immune recognition.

[0060] In one embodiment, the immune-enhancing vector includes a carrier protein, virus-like particles, nanoparticles, or other antigen-presenting vectors capable of improving antigen immunogenicity. The carrier protein is a large protein molecule that can be linked to a target epitope via chemical coupling or gene fusion, utilizing its T-cell helper epitope to activate adaptive immunity.

[0061] In one specific embodiment, a carrier protein is used as the immunomodulatory carrier, preferably keyhole hemocyanin (KLH), and binding to the target epitope is achieved based on EDC-mediated coupling chemistry. Virus-like particles refer to non-infectious hollow particles formed by the self-assembly of viral structural proteins, whose surfaces can display the target epitope at high density, mimicking the arrangement of natural pathogens to enhance B-cell receptor cross-linking; nanoparticles are artificially synthesized micron or submicron-sized delivery matrices, possessing both antigen-loading and adjuvant functions, capable of regulating release kinetics and promoting dendritic cell uptake. Specifically, after identifying the AGTR1-targeted epitope, a suitable immunomodulatory carrier is selected based on the physicochemical properties of the target epitope and the expected immunization strategy. The target epitope is then bound to it through covalent coupling, adsorption / embedding, or genetic engineering expression. The epitope-carrier binding product is then subjected to buffer replacement, purification, or combination with suitable excipients to obtain the AGTR1-targeted vaccine formulation.

[0062] Optionally, the targeted vaccine construction method further includes: Obtain exosomes; The target epitope, the binding product of the target epitope and the immune enhancement vector, and the exosomes are mixed to obtain a mixture; The mixture is screened, and the screened mixture is used as a targeted delivery formulation.

[0063] Exosomes represent nanoscale vesicle structures secreted by cells, preferably milk-derived exosomes, used as natural biological carriers to mediate antigen delivery and regulate the immune microenvironment. Their good biocompatibility and tissue targeting enhance the stability and delivery efficiency of target antigen components, adapting to the needs of immune intervention under different administration routes such as oral administration and gavage. The binding product of the target epitope and the immune-enhancing carrier, or its fragments, is the component to be loaded and is the core active substance for inducing specific immune responses. In one specific embodiment, the KLH-epitope binding product can be pre-digested with trypsin to form a fragment suitable for exosome loading.

[0064] A mixture or AGTR1 targeted delivery formulation refers to a composition formed by co-treating a free target epitope, the aforementioned binding product or its fragments with exosomes. This composition retains the biological characteristics of each component and enhances delivery efficiency. Specifically, milk-derived exosomes are obtained as the basis of the delivery system; the component to be loaded is mixed with the exosomes and incubated at room temperature to achieve initial binding, followed by sonication to promote loading; subsequently, the sample is incubated at a suitable temperature to promote exosome membrane recovery, and unloaded components are removed by ultrafiltration to obtain a stable AGTR1 targeted delivery formulation. For formulation characterization, size exclusion chromatography can be used to detect changes in the high molecular weight distribution of the sample before and after loading to evaluate whether the target component has successfully entered the relevant exosome components; simultaneously, changes in ELISA reactivity before and after exosome lysis can be used to evaluate loading accessibility and encapsulation status.

[0065] Optionally, the target site includes the angiotensin II type 1 receptor, extracellular region, functional fragment, or mutant encoded by the AGTR1 gene; The targeted vaccine or targeted delivery formulation is used to prepare metabolic regulatory substances.

[0066] In one embodiment, the target includes the full-length angiotensin II type 1 receptor protein encoded by the AGTR1 gene, the extracellular domain of the receptor, the functional fragment retaining biological activity, and artificially modified mutants. These targets provide specific recognition objects for vaccine design, covering the receptor's functional region to induce targeted immune responses and adapt to different research and development needs. Metabolic regulators refer to products capable of intervening in physiological processes such as energy balance, glucose and lipid metabolism, or blood pressure homeostasis. In practice, suitable target points are selected from the full-length AGTR1 protein, its extracellular domain, functional fragments, or mutants, and targeted vaccines or targeted delivery formulations are constructed accordingly.

[0067] This application provides a targeted vaccine construction apparatus, comprising: The structural data acquisition module is used to acquire the three-dimensional structural data of the antibody-antigen complex; The graph construction module is used to convert the antigen portion of the three-dimensional structural data into an antigen residue layer map. The training module is used to train an artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model. The prediction module is used to process the target point through the epitope identification model to obtain the target epitope; A construction module is used to construct a targeted vaccine for the target target based on the target epitope.

[0068] like Figure 2As shown in the embodiment of this application, an electronic device 200 includes a memory 210 and a processor 220; the memory 210 is used to store a computer program; the processor 220 is used to implement the targeted vaccine construction method as described above when the computer program is executed.

[0069] Alternatively, an electronic device 200 includes a memory 210 and a processor 220 coupled to the memory 210; the memory 210 is configured to store a computer program; and the processor 220 is configured to perform the following operations when the computer program is executed: Obtain three-dimensional structural data of the antibody-antigen complex; The antigen portion of the three-dimensional structural data is converted into an antigen residue layer diagram. An artificial intelligence model is trained using antigen residue layer maps to obtain an epitope recognition model. The target epitope is obtained by processing the target point through the epitope identification model; A targeted vaccine for the target epitope is constructed based on the target epitope.

[0070] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the targeted vaccine construction method described above.

[0071] Alternatively, a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the following operations: Obtain three-dimensional structural data of the antibody-antigen complex; The antigen portion of the three-dimensional structural data is converted into an antigen residue layer diagram. An artificial intelligence model is trained using antigen residue layer maps to obtain an epitope recognition model. The target epitope is obtained by processing the target point through the epitope identification model; A targeted vaccine for the target epitope is constructed based on the target epitope.

[0072] Electronic device 200, which can serve as a server or client in this application, is described below as an example of hardware devices that can be applied to various aspects of this application. Electronic device 200 is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 200 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0073] Electronic device 200 includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) or a computer program loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0074] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc. In this application, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application according to actual needs. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units can be implemented in hardware or as software functional units.

[0075] Although the above disclosure is provided, the scope of protection of this application is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this application, and all such changes and modifications will fall within the scope of protection of this application.

Claims

1. A method for constructing a targeted vaccine, characterized in that, include: Obtain three-dimensional structural data of the antibody-antigen complex; The antigen portion of the three-dimensional structural data is converted into an antigen residue layer diagram. An artificial intelligence model is trained using antigen residue layer maps to obtain an epitope recognition model. The target epitope is obtained by processing the target point through the epitope identification model; A targeted vaccine for the target epitope is constructed based on the target epitope.

2. The targeted vaccine construction method according to claim 1, characterized in that, The step of converting the antigen portion of the three-dimensional structural data into an antigen residue layer map includes: Each antigen residue corresponds to a node in the antigen residue layer diagram, and the spatial proximity between residues corresponds to an edge in the antigen residue layer diagram.

3. The targeted vaccine construction method according to claim 2, characterized in that, The step of converting the antigen portion of the three-dimensional structural data into an antigen residue layer map further includes: Node features are constructed based on the amino acid type information and local structural description information of the antigen portion; Edge features are constructed based on the Euclidean distance between residues of the antigen portion, the normalized three-dimensional direction vector, and the relative sequence position difference; The antigen residue layer map is constructed based on the node features and the edge features.

4. The targeted vaccine construction method according to claim 1, characterized in that, The step of training an artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model includes: By using the linear layer in the artificial intelligence model, node features are mapped to the hidden space to obtain the initial node representation, and edge features are geometrically encoded to obtain the edge geometric representation. The attention node representation is obtained by message passing between the initial node representation and the edge geometric representation through the graph convolution module; The neighborhood geometry encoding module obtains a geometry-enhanced node representation based on the attention node representation and the edge geometry representation; The attention node representation and the geometric enhancement node representation are adaptively weighted and fused through the gating fusion module to obtain the final node representation for epitope classification. The loss is calculated based on the final node representation and the real label, and the artificial intelligence model is trained to obtain the epitope recognition model.

5. The targeted vaccine construction method according to claim 4, characterized in that, The step of obtaining geometrically enhanced node representations based on the attention node representations and the edge geometric representations through the neighborhood geometric encoding module includes: In the attention node representation, the geometric representation of the edge corresponding to the incoming edge of each node is averaged and aggregated to obtain the aggregation result; The aggregation result is concatenated with the node representation and then input into a multilayer perceptron to obtain the geometrically enhanced node representation.

6. The targeted vaccine construction method according to claim 4, characterized in that, The step of calculating the loss based on the final node representation and the real label, and training the artificial intelligence model to obtain the epitope recognition model includes: A stratified sampling strategy was adopted to divide the training set and the validation set at the antigen residue level graph level. During training, node-based balanced batch sampling is performed on the training set, wherein positive residue samples are oversampled and negative residue samples are downsampled, and the validation set maintains the original class distribution; The loss is calculated using a pre-constructed hybrid loss function; The artificial intelligence model is trained to obtain the epitope recognition model.

7. The targeted vaccine construction method according to claim 1, characterized in that, The targeted vaccine based on the target epitope to construct the target target includes: The target epitope is combined with an immune-enhancing vector to construct the targeted vaccine for the target target, wherein the immune-enhancing vector includes a carrier protein, virus-like particles, and nanoparticles.

8. The targeted vaccine construction method according to claim 1, characterized in that, Also includes: Obtain exosomes; The target epitope, the binding product of the target epitope and the immune enhancement vector, and the exosomes are mixed to obtain a mixture; The mixture is screened, and the screened mixture is used as a targeted delivery formulation.

9. The method for constructing a targeted vaccine according to claim 1, characterized in that, The target sites include the angiotensin II type 1 receptor encoded by the AGTR1 gene, its extracellular region, functional fragments, or mutants; The targeted vaccine or targeted delivery formulation is used to prepare metabolic regulatory substances.

10. A targeted vaccine construction device, characterized in that, include: The structural data acquisition module is used to acquire the three-dimensional structural data of the antibody-antigen complex; The graph construction module is used to convert the antigen portion of the three-dimensional structural data into an antigen residue layer map. The training module is used to train an artificial intelligence model using an antigen residue layer map to obtain an epitope recognition model. The prediction module is used to process the target point through the epitope identification model to obtain the target epitope; A construction module is used to construct a targeted vaccine for the target target based on the target epitope.

Images