Virus structure three-dimensional reconstruction method and system

Abstract

The embodiment of the invention discloses a virus structure three-dimensional reconstruction method and system. The method comprises the following steps: acquiring stereoscopic data and point cloud data containing a target virus sample; separating virus particles from the point cloud data by using an optimized point cloud segmentation algorithm; separating a virus region from the stereoscopic data by using a stereoscopic data segmentation algorithm; processing the virus stereoscopic data by using a first neural network to obtain density characteristics of a virus structure; processing the virus point cloud by using a second neural network to obtain geometric features of the virus surface; fusing the density features and the geometric features by using a multi-modal fusion model to generate unified feature representation; reconstructing a three-dimensional model of a virus structure based on unified feature representation; according to the method, the three-dimensional structure of the virus can be accurately obtained under a complex background, and the accuracy of three-dimensional reconstruction of the virus is improved.

Description

Technical Field

[0001] The present application relates to the fields of biomedical imaging and computer vision technology, and in particular to a method and system for three-dimensional reconstruction of virus structure. Background Art

[0002] Three-dimensional reconstruction of viral structure is a key research area in virology, drug development, and vaccine design. Traditional methods for analyzing viral structure, such as cryo-electron microscopy and stereomicroscopy, can provide high-resolution electron density distribution data, but they still face challenges such as interference from complex biological backgrounds and difficulty extracting surface geometric features. Furthermore, the fine features of viral structure, such as subnanometer-scale details of surface proteins, require high-precision segmentation and reconstruction algorithms.

[0003] Therefore, a 3D reconstruction method for viral structure that can combine viral biological priors, achieve sub-nanometer resolution segmentation and multimodal fusion is needed to overcome the limitations of existing technologies. Summary of the Invention

[0004] The present application provides a method for three-dimensional reconstruction of virus structure, which improves the accuracy of three-dimensional reconstruction of viruses.

[0005] This application provides the following solutions:

[0006] According to a first aspect, a method for three-dimensional reconstruction of a virus structure comprises: obtaining volumetric data containing a target virus sample, the volumetric data comprising an electron density distribution of the target virus sample; obtaining point cloud data, the point cloud data comprising a three-dimensional point set representing surface geometric features of the target virus sample; separating virus particles from the point cloud data using an optimized point cloud segmentation algorithm, the optimized point cloud segmentation algorithm combining biological prior knowledge of viral surface protein distribution and a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles, thereby generating a separated virus point cloud; separating virus regions from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation to remove interference from cell debris and non-viral particles, thereby generating separated virus volumetric data; processing the virus volumetric data using a first neural network to obtain density features of the virus structure; processing the virus point cloud using a second neural network to obtain geometric features of the virus surface; fusing the density features and the geometric features using a multimodal fusion model to generate a unified feature representation; and reconstructing a three-dimensional model of the virus structure based on the unified feature representation to obtain a three-dimensional real-life model of the target virus sample.

[0007] According to an achievable method in an embodiment of the present application, the optimized point cloud segmentation algorithm is implemented by a point cloud neural network based on a sub-nanometer attention mechanism; wherein, the sub-nanometer attention mechanism includes: for the sub-nanometer resolution of the virus point cloud, limiting the local neighborhood range of attention calculation to a radius of 0.5 to 2 nanometers; utilizing biological prior knowledge of the distribution of virus surface proteins, generating prior weights based on virus symmetry and expected protein positions, and adjusting the attention score based on the prior weights.

[0008] According to an achievable method in an embodiment of the present application, generating a priori weights based on the virus symmetry and the expected protein position, and adjusting the attention score based on the prior weights include: calculating the local density of each point in the point cloud data, wherein the local density is determined by the inverse of the average distance between the k-nearest neighbor points of each point in the point cloud data; generating density weights based on the local density, enhancing the weights of high-density areas and attenuating the weights of low-density areas through a nonlinear function; generating prior weights in combination with the biological priors of the virus, wherein the prior weights are based on the symmetry of the virus and the distribution of surface proteins, and are determined by determining the distances between the point cloud points and the virus center and the protein center through a Gaussian kernel function; fusing the density weights and the prior weights to generate a comprehensive weight; and adjusting the score of the attention module in the point cloud segmentation algorithm based on the comprehensive weight.

[0009] According to an achievable method in an embodiment of the present application, the first neural network adopts a deep learning model based on a three-dimensional U-Net.

[0010] According to an implementable method in an embodiment of the present application, the second neural network is implemented using a PointNet model, a PointNet++ model, or a PointTransformer model.

[0011] According to an achievable method in an embodiment of the present application, the multimodal fusion model utilizes a cross-modal attention mechanism to align the density features and the geometric features, and optimizes feature extraction in combination with the biological prior knowledge of the virus's symmetry and surface protein distribution; wherein the cross-modal attention mechanism includes: projecting the density features from the volume grid to the point cloud coordinate space, and generating a density feature representation aligned with the geometric features through trilinear interpolation; generating biological prior weights based on the virus's symmetry and surface protein distribution, and the prior weights are obtained by calculating the distance between the point cloud point and the virus center or protein center in the Gaussian kernel function; constructing a cross-modal attention module, using density features as queries and geometric features as keys and values, calculating the attention score, and adjusting the score through the prior weights; using geometric features as queries and density features as keys and values, calculating the reverse attention score, fusing the prior weights, and generating enhanced geometric features; and generating fused features by fusing the enhanced density features and geometric features through a multi-layer perceptron.

[0012] According to an achievable method in an embodiment of the present application, the calculation of the attention score includes: obtaining the attention score through parallel calculation of 4 or more attention heads; wherein the attention heads include: an attention head for analyzing the correlation between the density features and geometric features of the spike protein region and an attention head for analyzing the correlation between the smooth geometry and low-density features of the viral membrane region.

[0013] According to a second aspect, a three-dimensional reconstruction system for virus structure is provided, the system comprising: a volumetric data acquisition unit configured to acquire volumetric data containing a target virus sample, the volumetric data comprising the electron density distribution of the target virus sample; a point cloud data acquisition unit configured to acquire point cloud data, the point cloud data comprising a three-dimensional point set representing the surface geometric features of the target virus sample; a point cloud data segmentation unit configured to separate virus particles from the point cloud data using an optimized point cloud segmentation algorithm, the optimized point cloud segmentation algorithm combining biological prior knowledge of the distribution of virus surface proteins and a sub-nanometer attention mechanism to remove interference from cell debris and non-virus particles, thereby generating a separated virus point cloud; the volumetric data segmentation unit configured to use the volumetric data segmentation algorithm to separate virus particles from the point cloud data; The virus region is separated from the volumetric data, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation, removes interference from cell debris and non-viral particles, and generates separated virus volumetric data; a density feature extraction unit is configured to process the virus volumetric data using a first neural network to obtain density features of the virus structure; a geometric feature extraction unit is configured to process the virus point cloud using a second neural network to obtain geometric features of the virus surface; a feature fusion unit is configured to fuse the density features and the geometric features using a multimodal fusion model to generate a unified feature representation; and a model reconstruction unit is configured to reconstruct a three-dimensional model of the virus structure based on the unified feature representation to obtain a three-dimensional real-scene model of the target virus sample.

[0014] According to a third aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods according to the first aspect are implemented.

[0015] According to a fourth aspect, an electronic device is provided, comprising: one or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method described in any one of the above-mentioned first aspects.

[0016] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0017] This application combines multimodal fusion of stereo data and point cloud data, and adopts optimized point cloud segmentation algorithms and stereo data segmentation algorithms to effectively remove the interference of cell debris and non-viral particles, and accurately separate viral particles and viral areas. By introducing biological prior knowledge and sub-nanometer attention mechanisms, the segmentation accuracy and processing efficiency are improved. The density features and geometric features of the virus structure are extracted respectively using the first neural network and the second neural network, and then these two features are integrated through the multimodal fusion model to generate a unified feature representation and accurately reconstruct the three-dimensional model of the virus. This method can accurately obtain the three-dimensional structure of the virus in a complex background, improving the accuracy of the three-dimensional reconstruction of the virus.

[0018] Of course, any product implementing the present application does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0020] Figure 1 A diagram of the system architecture applicable to the embodiments of the present application;

[0021] Figure 2 A flowchart of a method for three-dimensional reconstruction of viral structure provided in an embodiment of the present application;

[0022] Figure 3 A structural block diagram of the virus structure 3D reconstruction system provided in an embodiment of the present application;

[0023] Figure 4 A schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field are within the scope of protection of this application.

[0025] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "a", "an", "the" and "the" used in the embodiments of the present invention and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise.

[0026] It should be understood that the term "and / or" as used herein is merely a description of the relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the associated objects are in an "or" relationship.

[0027] The word "if," as used herein, may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to the determination" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)," depending on the context.

[0028] Several technologies currently exist for 3D reconstruction based on volumetric data. These methods do not use point cloud data but instead directly convert the volumetric data into surface meshes. While these methods can achieve 3D reconstruction, they suffer from complex background interference, loss of surface details, and insufficient sub-nanometer accuracy.

[0029] In view of this, the present application provides a new approach. To facilitate understanding of the present application, the system architecture on which the present application is based is first described. Figure 1 An exemplary system architecture to which the embodiments of the present application can be applied is shown. Figure 1 As shown in , the system architecture may include a user device and a 3D reconstruction system located on the server side.

[0030] The user can input stereoscopic data and point cloud data through the user device, and the user device sends it to the 3D reconstruction system on the server side. The 3D reconstruction system can use the method provided in the embodiment of the present application to perform 3D reconstruction on the target virus sample to obtain a 3D real-life model. The server side can send the 3D real-life model to the user terminal, which uses the 3D real-life model for rendering to obtain a 2D image, a 3D image, a VR (virtual reality) scene, or an AR (augmented reality) scene.

[0031] User devices may include, but are not limited to, smart mobile terminals, smart home devices, wearable devices, and personal computers (PCs). Smart mobile devices may include mobile phones, tablets, laptops, PDAs (Personal Digital Assistants), and internet-connected cars. Smart home devices may include smart TVs and smart refrigerators. Wearable devices may include smart watches, smart glasses, virtual reality devices, augmented reality devices, and mixed reality devices (i.e., devices that support both virtual reality and augmented reality).

[0032] The 3D reconstruction system can be set up as an independent server, a server group, or a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a host product in the cloud computing service system. It solves the problems of difficult management and weak service scalability in traditional physical hosts and virtual private servers (VPS). Figure 1 In addition to the shown architecture, the 3D reconstruction system can also be set up on a computer terminal with strong computing capabilities.

[0033] It should be understood that Figure 1 The user equipment and 3D reconstruction system in the figure are merely illustrative. Any number of user equipment and 3D reconstruction systems may be provided as required.

[0034] Figure 2 Flowchart of the method for three-dimensional reconstruction of virus structure provided in the embodiment of the present application. The method can be performed by Figure 1 The virus structure 3D reconstruction system is performed in the system shown. Figure 2 As shown in , the method may include the following steps:

[0035] Step 201: Acquire volumetric data containing a target virus sample, where the volumetric data includes electron density distribution of the target virus sample.

[0036] Step 202: Acquire point cloud data, where the point cloud data includes a three-dimensional point set representing surface geometric features of the target virus sample.

[0037] Step 203: Separate virus particles from the point cloud data using an optimized point cloud segmentation algorithm. The optimized point cloud segmentation algorithm combines biological prior knowledge of the distribution of virus surface proteins with a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles, thereby generating a separated virus point cloud.

[0038] Step 204: Separate the virus region from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation, removes interference from cell debris and non-viral particles, and generates separated virus volumetric data.

[0039] Step 205: Process the virus volume data using the first neural network to obtain density features of the virus structure.

[0040] Step 206: Process the virus point cloud using a second neural network to obtain geometric features of the virus surface.

[0041] Step 207: Utilize a multimodal fusion model to fuse the density feature and the geometric feature to generate a unified feature representation.

[0042] Step 208: Reconstruct a three-dimensional model of the virus structure using the unified feature representation to obtain a three-dimensional realistic model of the target virus sample.

[0043] As can be seen from the above process, this application combines the multimodal fusion of stereo data and point cloud data, and adopts an optimized point cloud segmentation algorithm and a stereo data segmentation algorithm to effectively remove the interference of cell debris and non-viral particles, and accurately separate viral particles and viral areas. By introducing biological prior knowledge and sub-nanometer attention mechanisms, the segmentation accuracy and processing efficiency are improved. The density features and geometric features of the virus structure are extracted respectively using the first neural network and the second neural network, and then these two features are integrated through the multimodal fusion model to generate a unified feature representation and accurately reconstruct the three-dimensional model of the virus. This method can accurately obtain the three-dimensional structure of the virus in a complex background, improving the accuracy of the three-dimensional reconstruction of the virus.

[0044] The following describes in detail each step of the above process and the effects that can be produced, in conjunction with the embodiments. It should be noted that the terms "first" and "second" in this disclosure do not restrict the size, order, or quantity, but are merely used to distinguish between them in terms of name. For example, "first neural network" and "second neural network" are used to distinguish between two neural networks.

[0045] First, the above step 201, namely, "obtaining volumetric data containing a target virus sample, wherein the volumetric data includes the electron density distribution of the target virus sample" is described in detail with reference to an embodiment.

[0046] The target virus sample is the target sample for 3D reconstruction in this application. Volumetric data is a digital data set stored in the form of a 3D grid, which is used to represent the electron density distribution of the target virus sample in 3D space. Using stereoscopic microscopy techniques such as cryo-electron microscopy (Cryo-EM), the 3D electron density distribution data of the target virus sample is collected to form a data set in the form of a volumetric grid.

[0047] Volumetric data is represented as a three-dimensional array, where each voxel stores a corresponding three-dimensional coordinate point and its electron density value, reflecting the internal structure and surface features of the virus, such as high-density protein regions and low-density background regions. The high-density protein region includes the spike protein, a key functional unit for viral infection. Its high charge density and high curvature make it particularly prominent in the dense protein region and a key location for 3D reconstruction.

[0048] As a feasible approach, the process of acquiring volumetric data involves using cryo-fixation to prepare the target virus sample into a thin layer suitable for cryo-EM imaging. Cryo-EM is then used to acquire two-dimensional projection images of the sample from multiple angles, and electron density information is recorded through the interaction between the electron beam and the sample. A volumetric reconstruction algorithm is then used to reconstruct the two-dimensional projections into a three-dimensional electron density map, forming a volumetric mesh data.

[0049] The above step 202, i.e., "obtaining point cloud data, where the point cloud data includes a three-dimensional point set representing the surface geometric features of the target virus sample," is described in detail below in conjunction with an embodiment.

[0050] Point cloud data is a data structure that represents the geometric features of an object's surface in three-dimensional space. It consists of a set of discrete three-dimensional coordinate points. Each point typically contains spatial location information and may have additional attributes such as a normal vector, color, or intensity. In the context of 3D reconstruction of viral structures, point cloud data specifically refers to the set of 3D points that represent the geometric features of the target virus's surface.

[0051] Point cloud data can be obtained in a variety of ways, such as using atomic force microscopy, scanning electron microscopy, transmission electron microscopy, or assisted 3D laser scanning or structured light scanning to collect 3D point cloud data of the virus sample surface. Preferably, point cloud data can be obtained from volumetric data, from which the 3D geometric information of the target virus surface is extracted to generate a set of discrete 3D coordinate points, i.e., point cloud data. Specifically, this process uses a surface extraction algorithm to analyze the density distribution of the volumetric data, identify the boundary voxels of the virus surface, and convert them into a data structure in the form of a point cloud.

[0052] The following describes in detail step 203, i.e., "using an optimized point cloud segmentation algorithm to separate virus particles from point cloud data, combining the biological prior knowledge of virus surface protein distribution and a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles, and generate a separated virus point cloud," in conjunction with an embodiment.

[0053] This application uses an optimized point cloud segmentation algorithm to separate the point cloud data corresponding to the target virus sample from the point cloud data, accurately separating the point set representing the virus particles from the point cloud data, while removing interference from the background, such as cell debris and non-viral particles. The core of this algorithm lies in optimization processing, which combines the biological characteristics of the virus, such as the distribution of viral surface proteins, with a sub-nanometer attention mechanism. Biological prior knowledge utilizes the expected position information of viral surface proteins, such as the distribution of spike proteins obtained through protein structure prediction tools, to help the algorithm identify characteristic regions of the virus particles.

[0054] As a feasible approach, a point cloud segmentation algorithm optimized with a subnanometer attention mechanism is implemented using a point cloud neural network based on a subnanometer attention mechanism. By limiting the algorithm's focus to a localized region within the point cloud, ranging from 0.5 to 2 nanometers, it enhances the ability to capture details on the virus surface, thereby improving segmentation accuracy. Simultaneously, the algorithm leverages biological priors about the distribution of viral surface proteins to generate prior weights based on viral symmetry and expected protein locations, and adjusts the attention score based on these prior weights. Specifically, biological priors about the distribution of viral surface proteins, such as the expected location of the spike protein obtained through protein structure prediction tools, and the geometric symmetry of the virus, can be used to identify key characteristic regions of the virus particle. Based on this information, the algorithm generates prior weights that reflect the degree of association of each point in the point cloud with the virus center or protein location. Higher weights indicate a higher probability that the point belongs to the surface of the virus particle. The algorithm then applies these prior weights to the attention mechanism, adjusting the attention score to increase focus on key areas of the virus surface while reducing attention to background noise.

[0055] As a feasible method, adjusting the attention score according to the prior weight of the virus symmetry and / or the expected protein position includes: calculating the local density of each point in the point cloud based on the spatial distribution of its k-nearest neighbor points, wherein the local density is determined by the inverse of the average distance between the k-nearest neighbor points; generating a density weight based on the local density, enhancing the weight of the high-density area and attenuating the weight of the low-density area through a nonlinear function; generating a priori weight in combination with the biological prior of the virus, wherein the prior weight is based on the symmetry of the virus or the surface protein distribution, and is determined by determining the distance between the point cloud point and the virus center and the protein center through a Gaussian kernel function; fusing the density weight and the prior weight to generate a comprehensive weight; and adjusting the score of the attention module in the point cloud segmentation algorithm according to the comprehensive weight.

[0056] Specifically, the algorithm first analyzes the local density of each point in the point cloud. This density is determined by calculating the inverse of the average distance between each point and its k nearest neighbors, reflecting the density of the point cloud and specifically tailored to the high-density regions of viral surface proteins. Next, a density weight is generated based on the local density. A nonlinear function is used to enhance the weights of high-density regions while weakening the weights of low-density regions. This effectively suppresses noise points in the complex biological background, such as cell debris or non-viral particles. Furthermore, the algorithm incorporates prior knowledge of viral biology to generate prior weights. These prior weights are determined based on the symmetry of the virus and the expected distribution of surface proteins. These weights emphasize the importance of high-density regions of the virus by measuring the distance of the point cloud points from the virus center and the protein center. The algorithm then fuses the density weights and prior weights to generate a composite weight. This fusion process can be achieved through a weighted combination or other methods. Finally, the composite weight is used to adjust the score of the attention module in the point cloud segmentation algorithm. The attention module calculates an attention score for each point to determine its importance in the segmentation. The composite weight acts as a multiplicative factor, directly multiplied by the attention score to adjust the point's contribution.

[0057] Among them, as a preferred implementation method, generating prior weights in combination with prior biological knowledge of viruses can be achieved by the following method:

[0058] First, the input data is point cloud data, consisting of a set of three-dimensional points representing the virus surface, each with three-dimensional coordinates. Biological prior information consists of two parts: the virus's symmetry and the distribution of surface proteins. Symmetry information is based on the geometric properties of the virus. For example, a spherical virus has central symmetry, and its center can be obtained by calculating the average coordinates of all points in the point cloud. Surface protein distribution information comes from protein structure prediction tools such as AlphaFold, which predicts the three-dimensional structure of the virus's surface spike protein and obtains the coordinates of a set of protein centers, such as the spatial positions of 10 protein centers. This prior information provides a biological basis for subsequent weight generation.

[0059] Next, the distance weight between each point in the point cloud and the center of the virus is calculated. The algorithm traverses each point in the point cloud and calculates its spatial distance from the center point of the virus. These distances are processed using the Gaussian kernel function to generate symmetric weights: points closer to the center of the virus are given higher weights because they are more likely to be in the core area of the virus; points farther away have lower weights and may belong to the background or edge area. Specifically, the Gaussian kernel function generates a smoothly decaying weight based on the distance value: when the distance is close to 0, that is, the point is very close to the center of the virus, the weight is close to 1; when the distance exceeds 2 nanometers, the weight is close to 0. In order to adapt to the scale of the virus point cloud, the standard deviation of the Gaussian kernel function is set to 1 nanometer to ensure that the weight changes smoothly in the range of 0.5 to 2 nanometers.

[0060] Next, the distance weight between each point in the point cloud and the center of the surface protein is calculated. The algorithm iterates through each point in the point cloud again, calculating the minimum distance between that point and all protein centers—that is, the spatial distance from that point to the nearest protein center. This minimum distance is also converted into a protein distribution weight using a Gaussian kernel function: points closer to the protein center receive higher weights because they are more likely to be located in high-density areas such as the spike protein; points farther away receive lower weights because they may correspond to the viral membrane or other non-protein areas.

[0061] The symmetry weights and protein distribution weights are then weighted and fused to generate the final prior weights. This fusion process is achieved through a weighted average, with weights assigned based on the actual situation. For example, 30% of the weight is assigned to the symmetry weight and 70% to the protein distribution weight, reflecting the importance of protein distribution in viral surface segmentation.

[0062] As an implementable method, the present application can also combine the charge distribution on the virus surface and determine the prior weights through a dynamic modulation method based on charge density. First, obtain the point cloud data and the charge distribution on the virus surface predicted by bioinformatics tools to generate the charge weights of the point cloud data. High charge density areas usually correspond to the most functional parts of the virus surface, such as the binding sites of the spike protein. These areas play a key role in the infection mechanism of the virus. The charge distribution on the virus surface is determined by the charged amino acids of the protein. High charge density areas are often rich in these amino acids, forming sites where positive or negative charges are concentrated. Specifically, point clouds close to high charge density areas obtain higher charge weights, and low charge areas obtain lower charge weights. Then, the algorithm uses the charge weight as a modulation factor to dynamically adjust the contribution of the symmetry weight and the protein distribution weight: for point cloud points with high charge weights, the protein distribution weight is preferentially used because the spike protein often has a high charge; for point cloud points with low charge weights, the symmetry of the virus membrane is more significant, and the symmetry weight is preferentially used. In specific implementations, the charge weight serves as a gating mechanism to determine the relative proportion of the two weights. The gating mechanism determines the fusion ratio of the two weights based on the value of the charge weight through a smooth interpolation function. For example, the charge weight ranges from 0 to 1. When the charge weight is greater than 0.7, 90% of the weight is assigned to the protein distribution weight and 10% to the symmetry weight. When the charge weight is less than 0.7, 20% of the weight is assigned to the protein distribution weight and 80% to the symmetry weight. When the charge weight is between 0.3 and 0.7, the charge weight is used to dynamically adjust the contribution ratio of the protein distribution weight and the symmetry weight through a smooth function. Specifically, when the charge weight value is between 0.3 and 0.7, a sigmoid function is used to map the charge weight value to between 0 and 1, outputting a smooth interpolation factor for the linear combination of the two weights.

[0063] The following describes in detail step 204, i.e., "separating the virus region from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation to remove interference from cell debris and non-viral particles, thereby generating separated virus volumetric data," in conjunction with an embodiment.

[0064] This step segments the volumetric data. By using a specialized volumetric data segmentation algorithm and combining it with the synergistic effect of point cloud segmentation results, it effectively removes interference from the complex biological background and ultimately generates volumetric data containing only the virus region. Cross-modal collaborative segmentation generates a preliminary virus region mask by projecting the point cloud segmentation results into the volumetric grid space. This mask provides additional spatial constraints for the volumetric segmentation algorithm, helping the algorithm to more accurately locate the virus region. For example, the virus surface points identified by point cloud segmentation can be mapped to the corresponding voxels of the volumetric grid, guiding the algorithm to focus on these high-density areas, thereby reducing the misjudgment of background noise.

[0065] Specifically, the virus particles are isolated using a point cloud segmentation algorithm, generating a subset of point clouds labeled as virus. Each point in this subset is assigned a classification label, indicating that it belongs to a virus particle rather than background. Next, the volumetric segmentation algorithm is initialized. A deep learning model based on a 3D U-Net can be used. Through multiple layers of convolution and upsampling operations, the deep learning model analyzes the density characteristics of each voxel in the volumetric grid and its spatial neighborhood relationships to predict whether each voxel belongs to a virus region. The network input is the raw volumetric data, and the output is a probability map of the same size as the input, indicating the probability of each voxel belonging to a virus. The point cloud segmentation results are then projected into the volumetric grid space to generate a preliminary virus region mask. The algorithm then uses the projected point cloud mask as an additional input channel, along with the density features, to feed the network, or employ an attention mechanism to adjust the network's focus. For example, voxel regions marked as 1 in the mask are assigned higher attention weights, encouraging the network to prioritize learning the features of these regions while deemphasizing regions marked as 0 in the mask.

[0066] As a preferred embodiment, the stereoscopic data separation algorithm of the present application can also be combined with biological prior knowledge and implemented using the prior weights generated by the symmetry of the virus and the distribution of surface proteins. These weights are based on the distance of the voxel to the virus center or protein center. Voxels close to the key area obtain higher weights, and voxels far away have lower weights.

[0067] The above step 205, namely "processing the virus volume data using the first neural network to obtain the density characteristics of the virus structure" is described in detail below with reference to an embodiment.

[0068] Viral volumetric data is a three-dimensional mesh processed by a volumetric segmentation algorithm, containing electron density distributions representing only the viral region. These density values reflect the internal structure and surface features of the virus, such as the high density of the spike protein region and density variations in the nucleocapsid or membrane regions. The first neural network analyzes this volumetric mesh and extracts density information that summarizes the viral structural features.

[0069] This step relies on a pre-trained first neural network, which can be obtained through supervised learning on a training set containing labeled viral volumetric data. The training goal is to teach the network to extract density features from the volumetric data that accurately reflect the viral structure, while ignoring the effects of background noise or segmentation errors.

[0070] The first neural network can be implemented by a variety of neural network types, such as 3D convolutional neural networks, 3D residual networks, and so on. Preferably, a 3D U-Net architecture can be used, which is a deep learning model designed specifically for volumetric data and can efficiently process the voxel information of 3D grids. The encoder part of U-Net gradually compresses the volumetric data and extracts deep features, while the decoder part restores the spatial resolution through upsampling, ensuring that the output density features are consistent with the original volumetric grid size. Techniques such as residual connections or batch normalization may also be added to the network to enhance training stability and feature expression capabilities.

[0071] The above step 206, i.e., "processing the virus point cloud using a second neural network to obtain geometric features of the virus surface," is described in detail below with reference to an embodiment.

[0072] The virus point cloud is a three-dimensional point set processed by a point cloud segmentation algorithm. Each point represents a spatial location on the virus surface, reflecting the virus's geometric characteristics. The second neural network's task is to analyze this unordered point set and extract high-level features that summarize the geometric properties of the virus surface, such as surface curvature, normal vectors, or the topological structure of protein regions.

[0073] The second neural network typically uses a deep learning model specifically designed for point cloud data. It can be implemented using a variety of neural network types, such as dynamic graph convolutional neural networks and multi-view convolutional neural networks. Preferably, the second neural network can use PointNet, PointNet++, or PointTransformer. These networks can directly process unordered 3D point sets and generate feature representations describing the virus surface geometry through point-by-point feature extraction and global aggregation.

[0074] As an implementable method, the second neural network optimizes the processing of irregular surfaces and dynamic conformations by embedding dynamic topological constraints of viral surface proteins. The irregular surface of the virus refers to the complex geometric shapes of the viral membrane and protein regions, such as the uneven distribution of spike proteins or the irregular undulations of the membrane. The dynamic conformation reflects the conformational changes of viral surface proteins in their natural state. For example, the spike protein may present different spatial arrangements due to environmental factors. The dynamic topological constraints are based on prior biological knowledge of the virus, such as the spatial distribution and expected topological properties of the spike protein obtained through protein structure prediction tools. These constraints are embedded in the network's training or inference process as additional guidance information to help the network more accurately identify and express the complex geometric features of the virus surface.

[0075] In practice, dynamic topological constraints can be implemented by adjusting the network's loss function or attention mechanism. For example, when training the second neural network, a topological constraint loss term is added to encourage the network's generated geometric features to align with the known protein topology, for example, ensuring that points in the spike protein region have higher curvature or a specific neighborhood distribution. Specifically, the algorithm first assigns a topological label to each point in the point cloud based on prior information about the protein center. If the distance from a point to the nearest protein center is less than a threshold, the point is labeled as a spike protein region, which is expected to have higher curvature and a dense neighborhood distribution. The topological constraint loss term compares the network's predicted geometric features with these expected properties, for example, requiring that points in the spike protein region have higher curvature values or a specific neighborhood pattern in the feature space. The training goal is to enable the network to learn these topological properties and optimize feature extraction for irregular surfaces and dynamic conformations. During the inference phase, the network can prioritize points in the protein region through an attention mechanism, assigning higher weights based on the prior positions of the protein center, thereby enhancing feature extraction for these dynamic conformational regions. This optimization enables the network to better handle irregular surfaces and dynamic conformations, avoiding the limitations of traditional point cloud processing algorithms for complex geometries.

[0076] The above step 207, i.e., “using a multimodal fusion model to fuse density features and geometric features to generate a unified feature representation”, is described in detail below with reference to an embodiment.

[0077] Density features are high-level representations extracted by the first neural network from the isolated viral volume data. These are typically multi-channel feature maps of the same size as the volume mesh, describing high-density regions of the viral internal structure and their spatial distribution patterns. Geometric features are extracted by the second neural network from the isolated viral point cloud. These are typically high-dimensional feature matrices that capture the curvature, normal vectors, or topological structure of the viral surface protein regions. These two features represent different aspects of the virus: density features focus on internal electron density, while geometric features focus on surface shape. They complement each other but differ in their formats and spatial representations.

[0078] The task of the multimodal fusion model is to integrate these two heterogeneous features into a unified feature representation to eliminate differences between the modalities and enhance the expressive power of the features. This model can be implemented based on deep learning architectures such as multi-layer perceptrons or attention mechanisms. The fusion process first requires aligning the spatial representations of density features and geometric features. Because density features are based on a volume mesh, while geometric features are based on a point cloud, the model uses projection techniques to map density features to the point cloud coordinate space. For example, the algorithm uses trilinear interpolation to assign density feature values from the volume mesh to corresponding locations in the point cloud, generating a density feature representation with the same number of points as the point cloud, thereby achieving spatial alignment between the two.

[0079] The fusion process can be implemented in a variety of ways. For example, using a multi-layer perceptron to concatenate density features and geometric features into a high-dimensional input vector, and then extracting comprehensive features layer by layer through a multi-layer neural network to generate a unified feature representation. Another approach is to introduce a cross-modal attention mechanism to enhance the feature representation of key areas by calculating the relationship between density features and geometric features.

[0080] As an optimal implementation, the multimodal fusion model utilizes a cross-modal attention mechanism to align the density features and the geometric features, and optimizes feature extraction in combination with the biological prior knowledge of the virus's symmetry and surface protein distribution; wherein the cross-modal attention mechanism includes: projecting the density features from the volume grid to the point cloud coordinate space, and generating a density feature representation aligned with the geometric features through trilinear interpolation; generating biological prior weights based on the virus's symmetry and surface protein distribution, and the prior weights are obtained by calculating the distance between the point cloud point and the virus center or protein center in the Gaussian kernel function; constructing a cross-modal attention module, using density features as queries and geometric features as keys and values, calculating the attention score, and adjusting the score through the prior weights; using geometric features as queries and density features as keys and values, calculating the reverse attention score, fusing the prior weights, and generating enhanced geometric features; and generating fused features by fusing the enhanced density features and geometric features through a multi-layer perceptron.

[0081] Specifically, the cross-modal attention mechanism achieves spatial alignment of the two modalities by projecting the density features from the volumetric grid to the point cloud coordinate space. Due to the different data structures of the density features and geometric features, a unified spatial representation is required before fusion. The algorithm uses a trilinear interpolation method to map the density feature values of the volumetric grid to the three-dimensional coordinates of the point cloud points. Specifically, for each point cloud point, the algorithm searches for neighboring voxels in the volumetric grid based on its coordinates, interpolates based on the density values of these voxels, and generates a density feature representation with the same number of point cloud points. This projection process ensures that the density features and geometric features are spatially aligned, adapts to the sub-nanometer resolution of the virus, and retains the fine density information of key areas such as the spike protein.

[0082] Next, the cross-modal attention mechanism leverages prior knowledge of viral biology to generate prior weights to enhance focus on key viral regions. These prior weights are determined based on the symmetry of the virus and the distribution of surface proteins. The distance between a point cloud point and the virus center or protein center is calculated using a Gaussian kernel function. The virus center is calculated using the point cloud centroid or the density centroid of the volumetric data, reflecting the overall symmetry of the virus, such as the central nature of a spherical virus. The protein center is obtained using protein structure prediction tools. The algorithm traverses each point in the point cloud, calculates its distance to the virus center or the nearest protein center, and converts the distance into a weight value using a Gaussian kernel function. Points close to the virus center or protein center receive higher weights because they are more likely to belong to key regions, such as high-density protein regions; points further away receive lower weights and may correspond to background or smooth membrane regions. These prior weights provide biological guidance for the subsequent attention mechanism, highlighting the structural characteristics of the virus.

[0083] The mechanism then constructs a cross-modal attention module, using density features as queries and geometric features as keys and values, to calculate multi-head attention scores and achieve feature interaction between modalities. In this process, the projected density features serve as queries, interrogating relevant information from the geometric features; the geometric features serve as keys and values, providing details of the surface shape. Using a multi-head attention mechanism, the attention module calculates the similarity between density features and geometric features, generates an attention score, and determines which geometric features are most important for enhancing the density features. For example, the spike protein region may receive a higher attention score due to its high density and high curvature. The algorithm further uses prior weights to adjust these scores to enhance feature interaction in high-density protein regions. For example, points near the center of the protein receive higher attention scores due to their high weights, allowing the network to focus more on the geometric details of these regions, thereby optimizing the effect of feature fusion.

[0084] To ensure symmetrical information exchange between modalities, the mechanism symmetrically calculates reverse attention scores using geometric features as queries and density features as keys and values. This process is similar to the previous steps, but the roles are reversed: the geometric features query the density features for relevant information, and the density features provide details of the internal structure. The algorithm also uses a multi-head attention mechanism to calculate the reverse attention score and adjusts the score using prior weights to enhance the contribution of density information in key regions such as the spike protein to the geometric features. For example, the density features in the spike protein region have a stronger enhancement effect on the geometric features due to their high weight. This bidirectional attention mechanism ensures that the density features and geometric features complement each other during the fusion process, generating enhanced geometric features and preserving the complementary information of the two modalities.

[0085] Finally, the mechanism fuses the enhanced density features and enhanced geometric features through a multi-layer perceptron to generate the final fused features. Specifically, the algorithm concatenates the enhanced density features and geometric features into a high-dimensional input vector, which is then fed into the multi-layer perceptron. The multi-layer neural network then extracts comprehensive features layer by layer. The output fused features are a unified feature representation, such as a feature matrix, that combines the internal density distribution of the virus, such as the high-density characteristics of protein regions, with surface geometric information, such as the curvature and topology of the spike protein. This feature representation not only preserves the details of both modalities but also enhances the robustness and integrity of the features through cross-modal interaction.

[0086] As an implementable approach, the cross-modal attention mechanism of the present application can adopt a multi-head attention mechanism. The multi-head attention mechanism is a deep learning technology that enhances feature interaction by computing multiple attention "heads" in parallel. Each attention head is responsible for analyzing the relationship between input features from a different perspective, and the calculation of the attention score is based on the concepts of query, key, and value. Taking cross-modal fusion as an example, the density feature may serve as a query to inquire about the relevant information of the geometric feature; the geometric feature serves as the key and value, providing details of the surface shape. The attention score measures the similarity between the query and the key. The higher the score, the greater the contribution of the corresponding key-value pair to the query. In the present application, the multi-head attention mechanism divides the input density features and geometric features into multiple subspaces, and independently calculates the attention score in each subspace in parallel processing units. Each attention head focuses on a specific aspect of the feature, such as the high-density protein region of the density feature or the curvature characteristics of the spike protein of the geometric feature. Through the parallel computing of multiple heads, the mechanism can analyze the relationship between features from different perspectives and generate a more comprehensive attention score. The number of attention heads can be set according to the characteristics of the virus. For example, the cross-modal attention module uses 8 attention heads, with density features as queries and geometric features as keys and values. The algorithm first divides the density features and geometric features into 8 subspaces, with each head processing 8-dimensional features. The first head focuses on the high correlation between the density features and geometric features in the spike protein region, generating a high attention score, indicating that these points are critical for fusion. The second head focuses on the association between the smooth geometry and low-density features in the viral membrane region, generating a lower score. Other heads may capture feature interactions related to protein topology or viral symmetry. The algorithm adjusts the score based on the prior weight of the protein center, for example, giving higher weights to points close to the spike protein, so that the scores of these points are further improved in all heads.

[0087] The above step 208, namely "reconstructing a three-dimensional model of the virus structure based on the unified feature representation to obtain a three-dimensional realistic model of the target virus sample", is described in detail below with reference to an embodiment.

[0088] This application reconstructs a 3D model of the virus using a unified feature representation. The 3D model reconstruction process can utilize deep learning models or traditional geometric algorithms to convert the unified feature representation into a volumetric model or surface mesh model. For example, a 3D convolutional neural network can be used to process the unified feature representation to generate an initial volumetric model.

[0089] As an implementable method, the three-dimensional real-scene model of the present application can be realized in the following way: first, based on a unified feature representation, an initial three-dimensional stereoscopic model is generated through a three-dimensional convolutional neural network, and the initial three-dimensional stereoscopic model represents the voxel density distribution of the virus structure.

[0090] Next, the prior weights are used to generate three-dimensional prior constraints. These constraints use a Gaussian kernel function to calculate the distance weights between the volume mesh voxels and the virus center or protein center, adapting to the subnanometer resolution of the virus voxel model and enhancing the voxel representation of high-density protein regions. The virus center is typically calculated from the center of mass of the volumetric data or point cloud, reflecting the overall symmetry of the virus, such as the central nature of spherical viruses. The protein center is obtained using protein structure prediction tools, such as the expected location of the SARS-CoV-2 spike protein. The algorithm assigns a weight to each voxel. Voxels closer to the virus center or protein center receive higher weights, as these regions are more likely to correspond to high-density protein structures; voxels farther away receive lower weights, likely representing background or low-density regions. These weights form a three-dimensional prior constraint that adapts to subnanometer resolution.

[0091] Next, the initial 3D volumetric model is optimized based on 3D prior constraints. The voxel density distribution is adjusted using weighted voxel smoothing. Weighted voxel smoothing enhances surface detail in high-density protein regions based on prior weights. Weighted voxel smoothing is a smoothing algorithm that comprehensively considers the density value of a voxel and its spatial neighborhood, adjusting the smoothing strength based on prior weights. Specifically, voxels with high prior weights receive weaker smoothing, preserving their surface detail and fine structure, while voxels with low prior weights receive stronger smoothing, suppressing noise contributions.

[0092] The optimized 3D volumetric model is then subjected to surface extraction to generate a triangular mesh model of the virus surface. This triangular mesh model is adapted to subnanometer resolution and preserves the geometric features of the viral surface proteins. This step uses a surface extraction algorithm to generate a triangular mesh model of the virus surface from the volumetric mesh. Surface extraction typically uses the Marching Cubes algorithm, which analyzes the density of the volumetric mesh, extracts an isosurface, and converts it into a 3D mesh composed of triangular facets.

[0093] Finally, the final 3D reality model is generated based on the triangulated model. A triangulated mesh model is a very common model representation in the field of 3D reconstruction. It consists of a set of vertices, edges, and triangular patches and is used to represent the surface geometry of objects, such as the spike protein and membrane structure on the surface of a virus.

[0094] Furthermore, the present application can also refine the triangular model before generating a three-dimensional real-life model. This includes mesh smoothing and hole filling based on the topological constraints of the virus surface protein to generate the final three-dimensional real-life model. Mesh smoothing removes jagged edges or irregular shapes caused by noise by adjusting the position of triangular facets, while using protein topological constraints to ensure that the smoothing process retains geometric details in key areas, such as the protruding structure of the spike protein. Hole filling repairs mesh defects caused by stereoscopic segmentation or surface extraction errors, such as missing facets in the viral membrane area. The refined triangular mesh model is smoother and more complete, with both biological authenticity and visual quality, which improves the quality of the three-dimensional real-life model of the target virus sample.

[0095] The above-mentioned method provided in the embodiments of the present application can be applied to a variety of application scenarios, including but not limited to: application in virus structure analysis to provide high-precision structural data for antiviral drug development; application in vaccine design to analyze the structural characteristics of virus surface antigens through high-precision three-dimensional models; application in virus mutation monitoring to analyze the impact of structural changes on infectivity or pathogenicity by comparing three-dimensional models of different virus strains, etc.

[0096] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0097] According to another embodiment, a system for three-dimensional reconstruction of virus structure is provided. Figure 3 FIG. 5 is a schematic block diagram of a virus structure three-dimensional reconstruction system according to an embodiment. Figure 3 As shown, the system 300 includes:

[0098] The volumetric data acquisition unit 301 is configured to acquire volumetric data containing a target virus sample, where the volumetric data includes an electron density distribution of the target virus sample.

[0099] The point cloud data acquisition unit 302 is configured to acquire point cloud data, where the point cloud data includes a three-dimensional point set representing surface geometric features of the target virus sample.

[0100] The point cloud data segmentation unit 303 is configured to separate virus particles from the point cloud data using an optimized point cloud segmentation algorithm. The optimized point cloud segmentation algorithm combines biological prior knowledge of the distribution of virus surface proteins and a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles to generate a separated virus point cloud.

[0101] The volumetric data segmentation unit 304 is configured to separate the virus region from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation, removes interference from cell debris and non-viral particles, and generates separated virus volumetric data.

[0102] The density feature extraction unit 305 is configured to process the virus volume data using a first neural network to obtain density features of the virus structure.

[0103] The geometric feature extraction unit 306 is configured to process the virus point cloud using the second neural network to obtain geometric features of the virus surface.

[0104] The feature fusion unit 307 is configured to fuse density features and geometric features using a multimodal fusion model to generate a unified feature representation.

[0105] The model reconstruction unit 308 is configured to reconstruct a three-dimensional model of the virus structure based on the unified feature representation to obtain a three-dimensional realistic model of the target virus sample.

[0106] As an implementable manner, the point cloud data segmentation unit 303 can be configured as follows: the optimized point cloud segmentation algorithm is implemented by a point cloud neural network based on a sub-nanometer attention mechanism; the sub-nanometer attention mechanism includes: for the sub-nanometer resolution of the virus point cloud, limiting the local neighborhood range of attention calculation to a radius of 0.5 to 2 nanometers; utilizing the biological prior knowledge of the distribution of virus surface proteins, generating prior weights according to the virus symmetry and the expected protein position, and adjusting the attention score according to the prior weights.

[0107] As an implementable manner, the point cloud data segmentation unit 303 can be configured to: calculate the local density of each point in the point cloud data, wherein the local density is determined by the inverse of the average distance between the k-nearest neighbor points of each point in the point cloud data; generate density weights based on the local density, enhance the weights of high-density areas and attenuate the weights of low-density areas through a nonlinear function; generate prior weights in combination with the biological priors of the virus, wherein the prior weights are based on the symmetry of the virus and the distribution of surface proteins, and are determined by determining the distances between the point cloud points and the virus center and the protein center through a Gaussian kernel function; fuse the density weights and the prior weights to generate a comprehensive weight; and adjust the score of the attention module in the point cloud segmentation algorithm according to the comprehensive weight.

[0108] As an implementable approach, the first neural network adopts a deep learning model based on a three-dimensional U-Net.

[0109] As an implementable approach, the second neural network is implemented using a PointNet model, a PointNet++ model, or a PointTransformer model.

[0110] As an implementable manner, the feature fusion unit 307 can be configured as follows: a multimodal fusion model uses a cross-modal attention mechanism to align the density features and the geometric features, and optimizes feature extraction by combining the biological prior knowledge of the virus's symmetry and surface protein distribution; wherein the cross-modal attention mechanism includes: projecting the density features from the volume grid to the point cloud coordinate space, and generating a density feature representation aligned with the geometric features through trilinear interpolation; generating biological prior weights based on the virus's symmetry and surface protein distribution, and the prior weights are obtained by calculating the distance between the point cloud point and the virus center or protein center in the Gaussian kernel function; constructing a cross-modal attention module, using density features as queries and geometric features as keys and values, calculating the attention score, and adjusting the score through the prior weights; using geometric features as queries and density features as keys and values, calculating the reverse attention score, fusing the prior weights, and generating enhanced geometric features; and generating fused features by fusing the enhanced density features and geometric features through a multi-layer perceptron.

[0111] As an implementable method, the feature fusion unit 307 can be configured to obtain an attention score through parallel calculation of 4 or more attention heads; wherein the attention heads include: an attention head for analyzing the correlation between the density features and geometric features of the spike protein region and an attention head for analyzing the correlation between the smooth geometry and low-density features of the viral membrane region.

[0112] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0113] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0114] In addition, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of any one of the methods in the aforementioned method embodiments are implemented.

[0115] And an electronic device comprising:

[0116] One or more processors; and a memory associated with the one or more processors, the memory being used to store program instructions, which, when read and executed by the one or more processors, execute the steps of the method described in any one of the aforementioned method embodiments.

[0117] in, Figure 4 The electronic device architecture is shown as an example, and may include a processor 410, a video display adapter 411, a disk drive 412, an input / output interface 413, a network interface 414, and a memory 420. The processor 410, the video display adapter 411, the disk drive 412, the input / output interface 413, the network interface 414, and the memory 420 may be communicatively connected via a communication bus 430.

[0118] The processor 410 may be implemented as a general-purpose CPU, a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and may be used to execute relevant programs to implement the technical solutions provided in this application.

[0119] The memory 420 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 420 can store an operating system 421 for controlling the operation of the electronic device 400 and a basic input and output system (BIOS) 422 for controlling the low-level operations of the electronic device 400. In addition, a web browser 423, a data storage management system 424, and a virus structure three-dimensional reconstruction system 425, etc. can also be stored. The above-mentioned virus structure three-dimensional reconstruction system 425 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present application. In short, when the technical solution provided by the present application is implemented by software or firmware, the relevant program code is stored in the memory 420 and is called and executed by the processor 410.

[0120] The input / output interface 413 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components in the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0121] The network interface 414 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0122] The bus 430 comprises a pathway for transmitting information between the various components of the device (eg, the processor 410 , the video display adapter 411 , the disk drive 412 , the input / output interface 413 , the network interface 414 , and the memory 420 ).

[0123] It should be noted that although the above device only shows a processor 410, a video display adapter 411, a disk drive 412, an input / output interface 413, a network interface 414, a memory 420, a bus 430, etc., in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may also include only the components necessary to implement the solution of the present application, and does not necessarily include all the components shown in the figure.

[0124] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer program product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0125] The above is a detailed introduction to the technical solutions provided by this application. Specific examples are used herein to illustrate the principles and implementation methods of this application. The description of the above embodiments is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the contents of this specification should not be understood as limiting this application.

Claims

1. A method for three-dimensional reconstruction of viral structure, characterized in that: The method comprises: Acquiring volumetric data containing a target virus sample, wherein the volumetric data includes an electron density distribution of the target virus sample; Acquiring point cloud data, the point cloud data comprising a three-dimensional point set representing surface geometric features of the target virus sample; separating viral particles from the point cloud data using an optimized point cloud segmentation algorithm, wherein the optimized point cloud segmentation algorithm combines biological prior knowledge of viral surface protein distribution with a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles, thereby generating a separated viral point cloud; Separating the virus region from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation to remove interference from cell debris and non-viral particles, thereby generating separated virus volumetric data; Processing the viral volume data using a first neural network to obtain density characteristics of the viral structure; Processing the virus point cloud using a second neural network to obtain geometric features of the virus surface; Using a multimodal fusion model to fuse the density feature and the geometric feature to generate a unified feature representation; A three-dimensional model of the virus structure is reconstructed based on the unified feature representation to obtain a three-dimensional realistic model of the target virus sample.

2. The method according to claim 1, characterized in that The optimized point cloud segmentation algorithm is implemented by a point cloud neural network based on a sub-nanometer attention mechanism; The sub-nanoscale attention mechanism includes: Aiming at the sub-nanometer resolution of the virus point cloud, the local neighborhood for attention calculation is limited to a radius of 0.5 to 2 nanometers; The biological prior knowledge of the distribution of viral surface proteins is used to generate prior weights based on viral symmetry and expected protein positions, and the attention score is adjusted according to the prior weights.

3. The method according to claim 2, characterized in that Generating a priori weights according to the virus symmetry and the expected protein position, and adjusting the attention score according to the prior weights include: Calculating a local density of each point in the point cloud data, wherein the local density is determined by an inverse of an average distance between k-nearest neighboring points of each point in the point cloud data; Generate density weights based on the local density, increase the weights of high-density areas and attenuate the weights of low-density areas through a nonlinear function; Generate prior weights based on the biological priors of the virus. The prior weights are determined based on the symmetry of the virus and the distribution of surface proteins, and the distances between the point cloud points and the virus center and the protein center are determined by using a Gaussian kernel function. Fusing the density weight and the prior weight to generate a comprehensive weight; The score of the attention module in the point cloud segmentation algorithm is adjusted according to the comprehensive weight.

4. The method according to claim 1, wherein The first neural network adopts a deep learning model based on three-dimensional U-Net.

5. The method according to claim 1, wherein The second neural network is implemented using a PointNet model, a PointNet++ model, or a PointTransformer model.

6. The method according to claim 1, characterized in that The multimodal fusion model utilizes a cross-modal attention mechanism to align the density features and the geometric features, and optimizes feature extraction by combining biological prior knowledge of virus symmetry and surface protein distribution; The cross-modal attention mechanism includes: Projecting the density features from the volume mesh to the point cloud coordinate space, and generating a density feature representation aligned with the geometric features by trilinear interpolation; Generate biological prior weights based on the symmetry of the virus and the distribution of surface proteins. The prior weights are obtained by calculating the distance between the point cloud points and the virus center or protein center in the Gaussian kernel function. Construct a cross-modal attention module, use density features as queries and geometry features as keys and values, calculate attention scores, and adjust the scores by the prior weights; Taking the geometric feature as the query and the density feature as the key and value, the reverse attention score is calculated and the prior weights are integrated to generate the enhanced geometric feature; The enhanced density features and geometric features are fused through a multi-layer perceptron to generate fused features.

7. The method according to claim 6, characterized in that The calculation of the attention score includes: obtaining the attention score through parallel calculation of 4 or more attention heads; wherein the attention heads include: an attention head for analyzing the correlation between the density features and geometric features of the spike protein region and an attention head for analyzing the correlation between the smooth geometry and low-density features of the viral membrane region.

8. A three-dimensional reconstruction system for virus structure, characterized in that: The system comprises: a volumetric data acquisition unit configured to acquire volumetric data containing a target virus sample, wherein the volumetric data includes an electron density distribution of the target virus sample; a point cloud data acquisition unit configured to acquire point cloud data, wherein the point cloud data includes a three-dimensional point set representing surface geometric features of the target virus sample; a point cloud data segmentation unit configured to separate virus particles from the point cloud data using an optimized point cloud segmentation algorithm, wherein the optimized point cloud segmentation algorithm combines biological prior knowledge of virus surface protein distribution with a sub-nanometer attention mechanism to remove interference from cell debris and non-viral particles, thereby generating a separated virus point cloud; a volumetric data segmentation unit configured to separate the virus region from the volumetric data using a volumetric data segmentation algorithm, wherein the volumetric data segmentation algorithm aligns the virus point cloud through cross-modal collaborative segmentation to remove interference from cell debris and non-viral particles, thereby generating separated virus volumetric data; a density feature extraction unit configured to process the virus volume data using a first neural network to obtain a density feature of the virus structure; a geometric feature extraction unit configured to process the virus point cloud using a second neural network to obtain geometric features of the virus surface; a feature fusion unit configured to fuse the density feature and the geometric feature using a multimodal fusion model to generate a unified feature representation; The model reconstruction unit is configured to reconstruct a three-dimensional model of the virus structure based on the unified feature representation to obtain a three-dimensional real-scene model of the target virus sample.

9. An electronic device, characterized in that: include: one or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions, wherein the program instructions, when read and executed by the one or more processors, perform the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images