Gene expression map generation method and device based on spatial omics hierarchy reconstruction, storage medium and equipment

Abstract

The invention discloses a gene expression map generation method and device based on spatial omics hierarchical reconstruction, a storage medium and equipment, relates to the technical field of biological information, and mainly aims to solve the problem of poor generation accuracy of an existing gene expression map. Comprising the following steps: acquiring original transcriptome data of a tissue gene; the gene data are predicted based on a hybrid neural network model, target slice space transcriptome data are obtained, the hybrid neural network model comprises a regulation and control niche network, a cell niche network and a cell communication prediction network, and a contrast diffusion bridge is constructed in the regulation and control niche network; a multi-modal condition diffusion bridge is constructed in the cell ecological niche network, and an optimal transport stream matching diffusion bridge is constructed in the cell communication prediction network; and taking the target slice space transcriptome data as an anchor slice to carry out space stream matching alignment to obtain an alignment result, and reconstructing the target slice space transcriptome data based on the alignment result to obtain a gene expression map of the tissue gene.

Description

Technical Field

[0001] This application relates to the field of bioinformatics technology, and in particular to a method, apparatus, storage medium, and device for generating gene expression maps based on spatial omics hierarchical reconstruction. Background Technology

[0002] Spatial resolved transcriptomics (SRT) is a technique used to understand the organizational and functional information of complex biological systems. By capturing molecular profiles within a spatial context, it enables in-situ studies of organizational architecture, developmental processes, and disease microenvironments. The spatial organization of biological systems inherently possesses a hierarchical nature, involving multiple levels from molecular regulation and cell type identity to cell-cell communication and tissue-level patterns. This hierarchical structure determines how molecular regulation, cell identity, intercellular communication, and spatial organization collectively shape tissue-level functions. Therefore, capturing these hierarchical biological interactions across multiple scales—including molecular, cellular, functional region, and tissue section scales—and establishing an integrative and dynamic understanding of space biology—i.e., generating gene expression maps—is a core challenge currently facing this field.

[0003] Currently, gene expression atlas generation typically employs spatial information neural networks for tissue functional region segmentation, deconvolution combining single-cell reference data for spatial point cellular composition inference, transcriptome data-based reconstruction of transcriptional regulatory relationships, and ligand-receptor co-expression inference of intercellular communication to achieve structural description of gene expression atlases. However, these methods often handle hierarchical descriptions in isolation, such as focusing only on clustering, deconvolution, or multi-slice geometric alignment. They lack integrated modeling of interdependencies at the molecular, cellular, and tissue scales, and cannot simulate hierarchical biological interactions in a three-dimensional context or predict systemic perturbations at the tissue / organ level. Therefore, a gene expression atlas generation method is urgently needed to address these issues. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus, storage medium, and device for generating gene expression maps based on spatial omics hierarchical reconstruction, with the main purpose of solving the problem of poor accuracy in existing gene expression map generation.

[0005] According to one aspect of this application, a method for generating gene expression maps based on spatial omics hierarchical reconstruction is provided, comprising: Obtain raw transcriptome data of tissue genes; The original transcriptome data is processed based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. A contrastive diffusion bridge is constructed in the regulatory niche network, a multimodal conditional diffusion bridge is constructed in the cellular niche network, and an optimal transport flow matching diffusion bridge is constructed in the cell communication prediction network. Using the target slice spatial transcriptome data as anchor slices, spatial flow matching and alignment are performed to obtain alignment results. Based on the alignment results, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes.

[0006] Furthermore, before processing the original transcriptome data using the hybrid neural network model that has completed model training to obtain the target slice spatial transcriptome data, the method further includes: Gene samples are obtained, and the network start point and network end point are determined based on the gene samples. The gene samples include original transcriptome samples that match transcriptome samples from different slice spaces. A regulatory niche network is constructed, and a contrastive diffusion bridge including the network start point and the network end point is constructed based on a conditional variational autoencoder. The contrastive diffusion bridge includes a first forward diffusion process and a first reverse diffusion process. A cell niche network is constructed, and a multimodal conditional diffusion bridge including the network start point and network end point is constructed based on a deep neural network. The multimodal conditional diffusion bridge includes a second forward diffusion process and a second reverse diffusion process. A cell communication prediction network is constructed, and an optimal transport flow matching diffusion bridge containing the network start point and network end point is constructed based on the diffusion bridge model. The optimal transport flow matching diffusion bridge includes a third forward diffusion process and a third reverse diffusion process. Based on the gene samples, the regulatory niche network, the cellular niche network, and the cell communication prediction network are trained together to obtain a hybrid neural network model.

[0007] Furthermore, the hybrid neural network model obtained by training the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: An enhanced transcriptome sample is obtained by fitting an objective function based on a graph neural network and determining the regulatory features of the original transcriptome sample based on the objective function. The graph neural network includes a position encoding layer, a multi-head self-attention layer, a fully connected feedforward network layer, and a normalization layer. Based on the enhanced transcriptome samples, conditional constraints are added during the first reverse diffusion process, and a first model loss function is determined based on learnable parameters to be applied to the regulatory niche network for hybrid training.

[0008] Furthermore, the hybrid neural network model obtained by training the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: A second model loss function is constructed based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss, and applied to the cellular niche network for hybrid training.

[0009] Furthermore, the hybrid neural network model obtained by training the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: A third model loss function is constructed based on single-cell flow matching loss and spatial omics flow matching loss; Perturbation parameters are configured based on ligand receptor activity samples in the gene samples, and communication strength parameters are determined based on the perturbation parameters to perform hybrid training on the cell communication prediction network.

[0010] Furthermore, the step of using the target slice spatial transcriptome data as anchor slices for spatial flow matching and alignment to obtain the alignment result includes: Using the target slice spatial transcriptome data as anchor slices, the point similarity between the anchor slices and the target slices is determined, and a coordinate set corresponding to the anchor slices and the target slices is constructed. The point similarity is used to construct the loss function of the flow matching model. The coordinate set is aligned based on the flow matching model that has completed model training to obtain the alignment result.

[0011] Further, the reconstruction of the target slice spatial transcriptome data based on the alignment results to obtain the gene expression map of the tissue genes includes: The flow matching model is iteratively trained based on the alignment results, and the model training of the flow matching model is completed when the number of training times matches a preset iteration threshold. Based on the flow matching model that has completed model training, the spatial transcriptome data of the target slice is reconstructed to obtain the gene expression map of the tissue genes.

[0012] According to another aspect of this application, a gene expression map generation device based on spatial omics hierarchical reconstruction is provided, comprising: The acquisition module is used to acquire raw transcriptome data of tissue genes; The processing module is used to process the original transcriptome data based on the hybrid neural network model that has been trained to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network has a contrastive diffusion bridge constructed in it, the cellular niche network has a multimodal conditional diffusion bridge constructed in it, and the cell communication prediction network has an optimal transport flow matching diffusion bridge constructed in it. The reconstruction module is used to perform spatial flow matching and alignment using the target slice spatial transcriptome data as anchor slices, obtain alignment results, and reconstruct the target slice spatial transcriptome data based on the alignment results to obtain the gene expression map of the tissue genes.

[0013] Furthermore, the device also includes: A determination module is used to acquire gene samples and determine the network start point and network end point based on the gene samples. The gene samples include original transcriptome samples that match transcriptome samples from different slice spaces. The first construction module is used to construct a regulatory niche network and construct a contrastive diffusion bridge including the network start point and the network end point based on a conditional variational autoencoder. The contrastive diffusion bridge includes a first forward diffusion process and a first reverse diffusion process. The second construction module is used to construct a cell niche network and construct a multimodal conditional diffusion bridge based on a deep neural network, including the network start point and the network end point. The multimodal conditional diffusion bridge includes a second forward diffusion process and a second reverse diffusion process. The third construction module is used to construct a cell communication prediction network and construct an optimal transport flow matching diffusion bridge based on the diffusion bridge model, which includes the network start point and the network end point. The optimal transport flow matching diffusion bridge includes a third forward diffusion process and a third reverse diffusion process. The training module is used to perform hybrid training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples to obtain a hybrid neural network model.

[0014] Furthermore, the training module is specifically used to fit an objective function based on a graph neural network, and to determine the regulatory features of the original transcriptome sample based on the objective function to obtain an enhanced transcriptome sample. The graph neural network includes a position encoding layer, a multi-head self-attention layer, a fully connected feedforward network layer, and a normalization layer. Based on the enhanced transcriptome sample, conditional constraints are added during the first backdiffusion process, and a first model loss function is determined based on learnable parameters to be applied to the regulatory niche network for hybrid training.

[0015] Furthermore, the training module is specifically used to construct a second model loss function based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss, so as to apply it to the cellular niche network for hybrid training.

[0016] Furthermore, the training module is specifically used to construct a third model loss function based on single-cell flow matching loss and spatial omics flow matching loss; configure perturbation parameters based on ligand samples of ligand receptor activity samples in the gene samples; and determine communication strength parameters based on the perturbation parameters, so as to perform hybrid training on the cell communication prediction network.

[0017] Furthermore, the reconstruction module is specifically used to determine the point similarity between the target slice and the target slice using the target slice spatial transcriptome data as anchor slices, and to construct a coordinate set corresponding to the anchor slice and the target slice. The point similarity is used to construct the loss function of the flow matching model. The coordinate set is aligned based on the flow matching model that has completed model training to obtain the alignment result.

[0018] Furthermore, the reconstruction module is specifically used to iteratively train the flow matching model based on the alignment results, and complete the model training of the flow matching model when the number of training times matches a preset iteration threshold; and reconstruct the target slice spatial transcriptome data based on the flow matching model that has completed model training to obtain the gene expression map of the tissue genes.

[0019] According to another aspect of this application, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform operations corresponding to the gene expression map generation method based on spatial omics hierarchical reconstruction described above.

[0020] According to another aspect of this application, a device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the gene expression map generation method based on spatial omics hierarchical reconstruction described above.

[0021] By employing the above technical solutions, the technical solutions provided in the embodiments of this application have at least the following advantages: This application provides a method, apparatus, storage medium, and device for generating gene expression maps based on hierarchical reconstruction of spatial omics. Compared with the prior art, the embodiments of this application obtain raw transcriptome data of tissue genes; process the raw transcriptome data based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network is constructed with a contrastive diffusion bridge, the cellular niche network is constructed with a multimodal conditional diffusion bridge, and the cell communication prediction network is constructed with an optimal transport flow matching diffusion bridge. The target slice spatial transcriptome data is used as an anchor slice for spatial flow matching and alignment to obtain the alignment result. Based on the alignment result, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes. This overcomes the defects in hierarchical modeling and 3D biological reconstruction, realizes system-wide perturbation response prediction, and more accurately simulates hierarchical biological interactions in a three-dimensional background and performs system-level perturbation prediction at the tissue and organ level, thereby achieving the goal of unified characterization of spatial biological systems across molecular, cellular, functional region, and tissue slice scales.

[0022] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart of a gene expression map generation method based on spatial omics hierarchical reconstruction provided in an embodiment of this application is shown; Figure 2 This illustration shows a schematic diagram of a large-scale spatial transcriptome dataset provided in an embodiment of this application; Figure 3 A schematic diagram of a layered diffusion bridge structure provided in an embodiment of this application is shown; Figure 4 This illustration shows a schematic diagram of a method for regulating niche generation according to an embodiment of this application; Figure 5 This illustration shows a schematic diagram of a cell niche network provided in an embodiment of this application; Figure 6This illustration shows a preliminary schematic diagram of cell-to-cell communication provided in an embodiment of this application; Figure 7 This application illustrates a learnable scoring network provided by an embodiment of the present application. Structural diagram; Figure 8 This illustration shows an alignment comparison diagram provided by an embodiment of this application; Figure 9 This illustration shows a schematic diagram of inter-slice mapping provided in an embodiment of this application; Figure 10 This illustration shows a block diagram of a gene expression map generation device based on spatial omics hierarchical reconstruction, as provided in an embodiment of this application. Figure 11 A schematic diagram of the structure of a device provided in an embodiment of this application is shown. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] The embodiments of this invention can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0027] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0028] Based on this, in one embodiment, the present invention provides a method for generating gene expression maps based on spatial omics hierarchical reconstruction. Taking the application of this method to computer devices such as servers as an example, the server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0029] This application provides a method for generating gene expression maps based on spatial omics hierarchical reconstruction, such as... Figure 1 As shown, the method includes: 101. Obtain raw transcriptome data of tissue genes.

[0030] In this embodiment, the current executing entity, acting as the processing end for gene expression map generation, can be a terminal device or a cloud server, etc., to obtain the gene data required for map generation. Here, "tissue genes" refers to gene images of tissue cells from different species, segmented in different dimensions within a large-scale spatiotemporal omics image dataset (spatial transcriptomics database), or gene images of tissue cells acquired in real-time; this embodiment does not impose specific limitations. Specifically, the raw transcriptome data includes labeled single-cell transcriptome data, spatial omics data, and ligand-receptor activity data, etc. Ligand-receptor activity data refers to the ligand-receptor interaction content within each cell or gene point, and single-cell transcriptome data refers to a single-cell composition matrix.

[0031] In some embodiments, such as Figure 2The large-scale spatial transcriptome dataset shown includes the following: Spatial barcode sequencing, which refers to a platform using spatial barcode capture-based spatial transcriptome technology, abbreviated as 10x Visium spatial transcriptome technology; Stereo sequencing, which refers to a platform using high-density spatial barcode array-based spatial transcriptome technology, abbreviated as Stereo-seq spatial transcriptome technology; and Slide sequencing, which refers to a platform using high-resolution spatial transcriptome technology based on sequencing, abbreviated as Slide-seq spatial transcriptome technology. In some embodiments, the raw transcriptome data may consist of a spatial transcriptomics (SRT) database, which may include data from STOmics (a spatiotemporal omics database), SOAR, SpatialDB (a spatial transcriptome resource), CROST (Comprehensive Repository of Spatial Transcriptomics), the 10x Genomics website, and the Census database, etc. For example, it may contain 96,700,729 cells / sites from 7,367 tissue sections, covering 365 tissue cells, including lung, skin, brain, liver, kidney, spinal cord, and embryo, and covering normal tissues and various diseases, such as pancreatic ductal adenocarcinoma, amyotrophic lateral sclerosis, non-small cell lung cancer, and hepatocellular carcinoma. The embodiments of this invention are not specifically limited.

[0032] 102. The original transcriptome data is processed based on the hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data.

[0033] In this embodiment, the hybrid neural network model is a hierarchical diffusion bridge-based model, which can be improved using the HieDiff hybrid diffusion framework in the field of medical image segmentation. The aim is to achieve a unified, dynamic, and cross-hierarchical representation of spatial biological systems. The HieDiff hybrid diffusion framework refers to a biologically based system of stochastic differential equations used to control the diffusion of flow signals between layers of a hierarchical structure, including both forward and backward processes.

[0034] The forward process refers to constructing a multilayer diffusion bridge connecting different biological levels and different slices, represented by a stochastic differential equation (SDE): ; The reverse process refers to recovering the specific expressions of each layer by solving in reverse. The corresponding probability flow ordinary differential equation (ODE) is expressed as: ; in, For the standard Wiener process, For drift term, For fluctuation terms, It is the score function of the multilayer diffusion bridge obtained from the forward process. It is the score function obtained from the reverse process, and z represents the spatial representation. State variables Includes raw transcriptome data (referred to as batch) Spatial data and single-cell data Enhanced transcriptome data representing gene regulatory characteristics (by functions) Modeling based on raw transcriptome data), ligand-receptor activity data characterizing ligand-receptor activity (including...) , , , ) and cell composition data (which can encode the relational weight matrix of each cell type to the spot) middle).

[0035] It should be noted that, as Figure 3 The schematic diagram of the hierarchical diffusion bridge structure shown in this application illustrates that the improved hybrid neural network model in this embodiment specifically includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network incorporates a contrastive diffusion bridge, the cellular niche network incorporates a multimodal conditional diffusion bridge, and the cell communication prediction network incorporates an optimal transport flow matching diffusion bridge. This allows for the representation of spatial biology dynamics / systems across molecular, cellular, functional region, and tissue slice scales as a continuous random diffusion process, thereby capturing how molecular regulation, cell type, intercellular communication, and spatial organization collectively shape tissue-level functions. Specifically, for the contrastive diffusion bridge, the multimodal conditional diffusion bridge, and the optimal transport flow matching diffusion bridge, a forward SDE diffusion process connecting biological levels can be constructed, along with a reverse process that recovers hierarchical-specific characteristics by solving the reverse probability flow ODE, thereby achieving cross-level integration.

[0036] 103. Using the target slice spatial transcriptome data as anchor slices, perform spatial flow matching and alignment to obtain alignment results, and reconstruct the target slice spatial transcriptome data based on the alignment results to obtain the gene expression map of the tissue genes.

[0037] In this embodiment, after obtaining the predicted target slice spatial transcriptome data, it is used as an anchor slice for spatial alignment. Specifically, the spatial alignment process can employ Optimal Flow Matching (OT-flowmatching) or Rectified Flow algorithms to smoothly map the target slice spatial transcriptome data to a unified aligned coordinate system; this embodiment does not impose specific limitations. Furthermore, based on the coordinate alignment results, three-dimensional tissue reconstruction is performed on the target slice spatial transcriptome data. During the reconstruction phase, a hierarchical diffusion bridge model is used to jointly model and iterate the state of spatial transcriptome data from multiple slices. This approach enables continuous feature generation and three-dimensional tissue reconstruction across slices, ultimately yielding a high-resolution gene expression map of the tissue. Here, the target slice spatial transcriptome data refers to the target slice spatial transcriptome feature matrix containing multi-scale biological constraints, expressing a precise mapping between gene expression profiles and tissue spatial structure. Through a hierarchically constructed diffusion bridge algorithm, a complex spatial heterogeneity map driven by the bottom-level gene regulatory network, constrained by the mid-level cell niche distribution, and coupled with high-level intercellular signal communication is deeply reconstructed.

[0038] In another embodiment of this application, for further definition and explanation, before the step of processing the original transcriptome data based on a hybrid neural network model that has completed model training to obtain the target slice spatial transcriptome data, the method further includes: Obtain gene samples and determine the network start point and network end point based on the gene samples; Construct a niche regulation network and build a contrastive diffusion bridge including the network start point and the network end point based on a conditional variational autoencoder; A cellular niche network was constructed, and a multimodal conditional diffusion bridge, including the network's starting point and ending point, was built based on a deep neural network. A cell communication prediction network is constructed, and an optimal transport flow matching diffusion bridge containing the network start point and network end point is constructed based on the diffusion bridge model. Based on the gene samples, the regulatory niche network, the cellular niche network, and the cell communication prediction network are trained together to obtain a hybrid neural network model.

[0039] To achieve multi-dimensional bridging based on hybrid neural network models, break the boundaries of isolated tasks, and effectively capture the dynamic processes that jointly shape tissue-level functions through molecular regulation, cellular identity, and spatial organization, a hybrid neural network including a regulatory niche network, a cellular niche network, and a cell communication prediction network is pre-constructed at the current execution end.

[0040] In some embodiments, for constructing a regulatory niche network to associate the inferred network with tissue functional regions or cell types, the regulatory niche network can be a deep neural network, such as a diffusion model, a generative adversarial network (GAN), or other generative models; this application does not specifically limit the specific model. Furthermore, the encoder in the regulatory niche network employs a conditional variational autoencoder, and a contrastive diffusion bridge including the network start point and the network end point is constructed based on the conditional variational autoencoder. The network start point can be the network input. The network endpoint can be the network output. ,in, For batch The original transcriptome sample, containing One gene and A spatial spot, the network endpoint is batch The original transcriptome sample and the enhanced transcriptome sample were exported together. To enhance transcriptome samples, it is possible to simultaneously infer gene regulatory networks and learn spatial representations. This links the control procedures to the spatial domain, such as Figure 4 The diagram illustrating the regulation of niche generation shows that the graph attention transformer can employ a Graph Transformers model. Furthermore, the constructed contrastive diffusion bridge includes a first forward diffusion process and a first backward diffusion process. Specifically, since the endpoints of the contrastive diffusion bridge can be constructed based on a conditional variational autoencoder (VAE), it is represented as follows: ; .

[0041] Furthermore, in constructing the first forward diffusion process, Doob's h-transform can be used to define a forward stochastic differential equation (SDE) for the contrastive diffusion bridge, expressed as: ; ; in, and Let represent the intermediate diffusion states originating from the original transcriptome sample and the enhanced transcriptome sample, respectively, and the corresponding functions are expressed as: ; ; Among them, through the transfer of nuclear and It satisfies the Kolmogorov inverse equation and follows a pre-specified normal distribution to accommodate the distribution characteristics of different spatial omics data (such as sparsity or continuity).

[0042] In a specific implementation scenario, two specific diffusion bridge instantiation strategies are compared: variance-preserving (VP) and variance-expanding (VE) diffusion bridges, with drift terms... diffusion coefficient The corresponding transfer kernel definitions are shown in Tables 1 and 2.

[0043] Table 1. Instantiation of VP and VE in diffusion bridges during primary transcription.

[0044] Table 2 Instantiation of VP and VE for enhancing diffusion bridges in transcription.

[0045] In some embodiments, in order to characterize a multi-level spatial biological system including molecular regulatory networks, cell types, and tissue functional structures, embodiments of this application construct as follows: Figure 5 The illustrated cell niche network can be implemented using deep neural networks, including but not limited to generative adversarial networks (GANs) or conditional GANs (cGANs), which are not specifically limited in this embodiment. Furthermore, a multimodal conditional diffusion bridge, including the network start point and network end point, is constructed based on the deep neural network. The network start point is the model input, and the network end point is the model output. The model input may include single-cell transcriptome samples. and space omics samples , This indicates a batch. The network endpoints are composed of reconstructed representations of single-cell and spatial omics data, respectively, denoted as... and The multimodal conditional diffusion bridge includes a second forward diffusion process and a second reverse diffusion process. The second forward diffusion process is constructed based on the forward stochastic differential equation (SDE) within the comparative diffusion bridge framework, and is expressed as: ; .

[0046] Furthermore, by conditionally linking the diffusion bridge between single-cell and spatial omics data and imposing constraints on intermediate signals, an ordinary differential equation (ODE) can be derived, expressed as: ; At this point, the scoring model can be parameterized as follows: as well as Therefore, the parameterized generation process is as follows: ; .

[0047] In some embodiments, due to different generative tasks, in learning deterministic mappings for inferring cell-cell communication, a cell communication prediction network is constructed, such as... Figure 6 The schematic diagram of cell-to-cell communication shown can employ a diffusion bridge model, including but not limited to an OT-flow matching model, to integrate single-cell and spatial transcriptome data by utilizing ligand-receptor interactions within each cell or point. In this case, an optimal transport flow matching diffusion bridge is constructed based on this diffusion bridge model, containing the network initiation point and network endpoint. The network initiation point is the model input, and the network endpoint is the model output. The optimal transport flow matching diffusion bridge includes a third forward diffusion process and a third reverse diffusion process. Specifically, through parameters... The variance of the cell communication prediction network is scaled to make the conditional distribution as follows: ,in, , For conditional input, in Under the limit, and given a fixed value and (and ), making and The condition is expressed as: ; in, Indicates that in a given and Under the given conditions, the drift term of the probabilistic flow ODE indicates that, in the noise-free limit, the model learns to match the drift of the bridge probabilistic flow ODE under variance explosion scheduling. In this case, the loss function can be expressed as: .

[0048] It should be noted that after the hybrid neural network model is constructed, the hybrid neural network is trained using gene samples. At this time, the gene samples include original transcriptome samples matched with spatial transcriptome samples from different slices, ligand receptor activity samples, single-cell transcriptome samples, and spatial transcriptome samples.

[0049] In another embodiment of this application, for further definition and explanation, the step of performing hybrid training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene sample to obtain a hybrid neural network model includes: An enhanced transcriptome sample is obtained by fitting an objective function using a graph neural network and determining the regulatory features of the original transcriptome sample based on the objective function. Based on the enhanced transcriptome samples, conditional constraints are added during the first reverse diffusion process, and a first model loss function is determined based on learnable parameters to be applied to the regulatory niche network for hybrid training.

[0050] To improve the accuracy of hybrid neural network models in describing gene expression, the current execution end can use a graph Transformer neural network to fit the target function during model training. Input from the original transcriptome sample The graph Transformer neural network consists of a priori gene regulatory network A. In this case, the graph Transformer neural network includes a position encoding layer, a multi-head self-attention layer, a fully connected feedforward network layer, and a normalization layer. The position encoding is obtained by performing singular value decomposition (SVD) on network A. Then, after passing through multi-head self-attention, a fully connected feedforward network layer, and layer normalization, the enhanced signal is output, which is the enhanced transcriptome sample.

[0051] In some embodiments, the control network can be decomposed into: ; in, For including the previous A matrix with singular left and right vectors corresponds to a diagonal matrix. The largest singular value in, sign This indicates concatenation by column. It is a learnable projection matrix used to map the concatenated singular vectors to positional codes. .

[0052] In some embodiments, the graph Transformer neural network includes Layers. For each layer And attention head The calculation formula is expressed as: ; ; ; in, , , They represent the first The first in the layer A query, key, and value matrix of size. For query / key feature dimensions, It is the first Layer Attention weight matrix for each size. Function Perform line concatenation; the initial input is... The final output is Furthermore, to summarize the gene-gene relationships across all layers and all attention heads, the average attention weight can be calculated as follows: ; in, This represents the total number of attention heads. This matrix G provides a global measure of the regulatory interactions obtained from model inference.

[0053] In some embodiments, conditional constraints are added to the first backdiffusion process based on the enhanced transcriptome sample; that is, the first backdiffusion process may further include a decoding step containing an autoencoder. and Furthermore, conditional constraints are introduced during the reverse process. For the original transcriptome sample, the reverse ordinary differential equation (ODE) is expressed as: ; For enhancing transcriptome samples Constraints can be imposed on the intermediate diffusion signal to obtain , represented as: .

[0054] Because of the direct calculation of conditional probability as well as At this time, due to It is not tractable; the scoring model can be parameterized as follows: ; .

[0055] Finally, the obtained parameterized generation process can be expressed by the ordinary differential equation: ; .

[0056] In addition, regarding the loss function during model training, based on learnable parameters... A first model loss function is determined and applied to the said niche regulation network for hybrid training; specifically, a learnable parameter is introduced. ,like Figure 7 As shown, and implemented through an autoencoder, the overall loss function is used to simultaneously optimize the autoencoder parameters and the graph Transformer parameters, and can be expressed as: ;in, , This represents the latent variables obtained through the reparameterization technique. and Used for constraints Align with batch-specific normal distributions as much as possible to mitigate batch effects during representation learning.

[0057] In another embodiment of this application, for further definition and explanation, the step of performing hybrid training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene sample to obtain a hybrid neural network model includes: A second model loss function is constructed based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss, and applied to the cellular niche network for hybrid training.

[0058] To improve the accuracy of hybrid neural network models in describing gene expression, a second model loss function is constructed during model training based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss. This function can be specifically expressed as: ; in, and The corresponding OT-flowmatching losses are for single-cell transcriptome samples and spatial omics transcriptome samples, respectively. The single-cell vector is the aggregation of ligands in all cells or points. For the corresponding spatial data, all receptors are aggregated into a vector. and . This represents the ligand-receptor interaction matrix. This is the inverse of the cell composition matrix, used to recover single-cell expression signals in spatial omics.

[0059] In another embodiment of this application, for further definition and explanation, the step of performing hybrid training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene sample to obtain a hybrid neural network model includes: A third model loss function is constructed based on single-cell flow matching loss and spatial omics flow matching loss; Perturbation parameters are configured based on ligand receptor activity samples in the gene samples, and communication strength parameters are determined based on the perturbation parameters to perform hybrid training on the cell communication prediction network.

[0060] To improve the accuracy of spatial structure learning between single cells and spatial omics, and to enhance the training effect of model parameters, the loss function of the third model can be expressed as: ;in, and The conditional diffusion bridge loss corresponding to single-cell signaling and This corresponds to the conditional diffusion bridge loss of spatial omics signals. Additionally, when spatial data can be generated by multiplying single-cell data by a matrix... When to recover, Finally, to integrate regulatory network inference with cellular niche analysis and elucidate the relationship between molecular regulatory networks, cell types, and tissue functional structures, thereby achieving a comprehensive characterization and mapping of multi-level spatial biological systems, a loss function for contrastive diffusion bridges is introduced. This yields the overall loss of the multimodal conditional diffusion bridge, expressed as: .

[0061] In another embodiment of this application, for further definition and explanation, the step of using the target slice spatial transcriptome data as an anchor slice for spatial flow matching and alignment, and obtaining the alignment result includes: Using the target slice spatial transcriptome data as anchor slices, the point similarity between the anchor slice and the target slice is determined, and a coordinate set corresponding to the anchor slice and the target slice is constructed. The coordinate set is aligned based on the flow matching model that has completed model training to obtain the alignment result.

[0062] To improve the accuracy of prediction using single-slice spatial transcriptome data by performing communication inference at multiple biological scales, the target slice spatial transcriptome data is used as an anchor slice during spatial flow matching alignment. The point similarity between the target slice and the anchor slice is determined. Specifically, OT-flow matching or rectifiedflow can be used to match the target slice spatial transcriptome data (spatial coordinates are...). Mapped to the aligned coordinate system This allows for spatial correspondence with the anchor point slice.

[0063] In some embodiments, a corrected spatial representation can be used. Calculate the point-to-point similarity matrix between two slices (anchor slice and target slice). Point similarity is used to construct the loss function of the flow matching model. Among them, Indicates the midpoint of an anchor slice or cell. Midpoint or cell of the target slice The point similarity is given by the similarity between the points. A higher similarity indicates stronger similarity and a closer expected alignment relationship, which can be expressed as: ; in, and These represent the expression vectors of the midpoint or cell i in the anchor slice and the midpoint or cell j in the target slice, respectively. Finally, the coordinate set is aligned based on the flow matching model that has completed model training, and the alignment result is obtained, as shown below. Figure 8 As shown, that is, in the coordinate set A rectified flow model is trained between the two flows, with the objective function expressed as: ; in, A model for predicting flow direction. After the model is trained, new coordinates are generated based on this model. ,in, remain unchanged. It is determined by the target coordinates The alignment result obtained by inference.

[0064] In another embodiment of this application, for further definition and explanation, the step of reconstructing the spatial transcriptome data of the target slice based on the alignment result to obtain the gene expression map of the tissue genes includes: The flow matching model is iteratively trained based on the alignment results, and the model training of the flow matching model is completed when the number of training times matches a preset iteration threshold. Based on the flow matching model that has completed model training, the spatial transcriptome data of the target slice is reconstructed to obtain the gene expression map of the tissue genes.

[0065] To further enhance the correspondence between the two distributions and thus improve the accuracy of gene characterization, when reconstructing the target slice spatial transcriptome data based on the alignment results, the flow matching model is iteratively trained based on the alignment results. When the number of training iterations matches the preset iteration threshold, preferably 5 times, the flow matching model training is completed. Finally, the target slice spatial transcriptome data is reconstructed based on the flow matching model that has completed model training to obtain the gene expression map of the tissue genes.

[0066] In some embodiments, the newly generated coordinates are used As input, the rectified flow model is retrained. Each iteration, after five iterations, results in... Figure 3 As shown, the final output It can provide robust spatial alignment results, such as Figure 9 As shown. Simultaneously, by integrating rigid transformations with constraints derived from expression similarity, a three-dimensional organizational structure reflecting the functional organization across slices can be reconstructed, yielding a gene expression map of the tissue genes.

[0067] This application provides a method for generating gene expression maps based on hierarchical reconstruction of spatial omics. Compared with the prior art, this application obtains raw transcriptome data of tissue genes; processes the raw transcriptome data based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network is constructed with a contrastive diffusion bridge, the cellular niche network is constructed with a multimodal conditional diffusion bridge, and the cell communication prediction network is constructed with an optimal transport flow matching diffusion bridge. The target slice spatial transcriptome data is used as an anchor slice for spatial flow matching and alignment to obtain the alignment result. Based on the alignment result, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes. This method overcomes the defects in hierarchical modeling and 3D biological reconstruction, achieves system-wide perturbation response prediction, and more accurately simulates hierarchical biological interactions in a three-dimensional background and performs system-level perturbation prediction at the tissue and organ level. This achieves the goal of unified characterization of spatial biological systems across molecular, cellular, functional region, and tissue slice scales.

[0068] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this application provides a gene expression map generation device based on spatial omics hierarchical reconstruction, such as... Figure 10 As shown, the device includes: Acquisition module 21 is used to acquire raw transcriptome data of tissue genes; Prediction module 22 is used to process the original transcriptome data based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network has a contrastive diffusion bridge constructed in it, the cellular niche network has a multimodal conditional diffusion bridge constructed in it, and the cell communication prediction network has an optimal transport flow matching diffusion bridge constructed in it. The reconstruction module 23 is used to perform spatial flow matching and alignment using the target slice spatial transcriptome data as anchor slices, obtain alignment results, and reconstruct the target slice spatial transcriptome data based on the alignment results to obtain the gene expression map of the tissue genes.

[0069] Furthermore, the device also includes: A determination module is used to acquire gene samples and determine the network start point and network end point based on the gene samples. The gene samples include original transcriptome samples that match transcriptome samples from different slice spaces. The first construction module is used to construct a regulatory niche network and construct a contrastive diffusion bridge including the network start point and the network end point based on a conditional variational autoencoder. The contrastive diffusion bridge includes a first forward diffusion process and a first reverse diffusion process. The second construction module is used to construct a cell niche network and construct a multimodal conditional diffusion bridge based on a deep neural network, including the network start point and the network end point. The multimodal conditional diffusion bridge includes a second forward diffusion process and a second reverse diffusion process. The third construction module is used to construct a cell communication prediction network and construct an optimal transport flow matching diffusion bridge based on the diffusion bridge model, which includes the network start point and the network end point. The optimal transport flow matching diffusion bridge includes a third forward diffusion process and a third reverse diffusion process. The training module is used to perform hybrid training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples to obtain a hybrid neural network model.

[0070] Furthermore, the training module is specifically used to fit an objective function based on a graph neural network, and to determine the regulatory features of the original transcriptome sample based on the objective function to obtain an enhanced transcriptome sample. The graph neural network includes a position encoding layer, a multi-head self-attention layer, a fully connected feedforward network layer, and a normalization layer. Based on the enhanced transcriptome sample, conditional constraints are added during the first backdiffusion process, and a first model loss function is determined based on learnable parameters to be applied to the regulatory niche network for hybrid training.

[0071] Furthermore, the training module is specifically used to construct a second model loss function based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss, so as to apply it to the cellular niche network for hybrid training.

[0072] Furthermore, the training module is specifically used to construct a third model loss function based on single-cell flow matching loss and spatial omics flow matching loss; configure perturbation parameters based on ligand samples of ligand receptor activity samples in the gene samples; and determine communication strength parameters based on the perturbation parameters, so as to perform hybrid training on the cell communication prediction network.

[0073] Furthermore, the reconstruction module is specifically used to determine the point similarity between the target slice and the target slice using the target slice spatial transcriptome data as anchor slices, and to construct a coordinate set corresponding to the anchor slice and the target slice. The point similarity is used to construct the loss function of the flow matching model. The coordinate set is aligned based on the flow matching model that has completed model training to obtain the alignment result.

[0074] Furthermore, the reconstruction module is specifically used to iteratively train the flow matching model based on the alignment results, and complete the model training of the flow matching model when the number of training times matches a preset iteration threshold; and reconstruct the target slice spatial transcriptome data based on the flow matching model that has completed model training to obtain the gene expression map of the tissue genes.

[0075] This application provides a gene expression map generation device based on hierarchical reconstruction of spatial omics. Compared with the prior art, this application obtains the original transcriptome data of tissue genes; processes the original transcriptome data based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network is constructed with a contrastive diffusion bridge, the cellular niche network is constructed with a multimodal conditional diffusion bridge, and the cell communication prediction network is constructed with an optimal transport flow matching diffusion bridge. The target slice spatial transcriptome data is used as an anchor slice for spatial flow matching and alignment to obtain the alignment result. Based on the alignment result, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes. This overcomes the defects in hierarchical modeling and 3D biological reconstruction, realizes system-wide perturbation response prediction, and more accurately simulates hierarchical biological interactions in a three-dimensional background and performs system-level perturbation prediction at the tissue and organ level. This achieves the goal of unified characterization of spatial biological systems across molecular, cellular, functional region, and tissue slice scales.

[0076] According to one embodiment of this application, a storage medium is provided, the storage medium storing at least one executable instruction, the computer-executable instruction being able to execute the gene expression map generation method based on spatial omics hierarchical reconstruction in any of the above method embodiments.

[0077] Figure 11 The diagram shows a structural schematic of a device according to one embodiment of the present application. The specific embodiments of the present application do not limit the specific implementation of the device.

[0078] like Figure 11 As shown, the device may include: a processor 302, a communications interface 304, a memory 306, and a communications bus 308.

[0079] The processor 302, communication interface 304, and memory 306 communicate with each other via communication bus 308.

[0080] Communication interface 304 is used to communicate with other network elements such as clients or other servers.

[0081] The processor 302 is used to execute program 310, specifically to execute the relevant steps in the above-described embodiment of the gene expression map generation method based on spatial omics hierarchical reconstruction.

[0082] Specifically, program 310 may include program code that includes computer operation instructions.

[0083] Processor 302 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0084] Memory 306 is used to store program 310. Memory 306 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0085] Specifically, program 310 can be used to cause processor 302 to perform the following operations: Obtain raw transcriptome data of tissue genes; The original transcriptome data is processed based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. A contrastive diffusion bridge is constructed in the regulatory niche network, a multimodal conditional diffusion bridge is constructed in the cellular niche network, and an optimal transport flow matching diffusion bridge is constructed in the cell communication prediction network. Using the target slice spatial transcriptome data as anchor slices, spatial flow matching and alignment are performed to obtain alignment results. Based on the alignment results, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes.

[0086] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0087] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for generating gene expression maps based on spatial omics hierarchical reconstruction, characterized in that, include: Obtain raw transcriptome data of tissue genes; The original transcriptome data is processed based on a hybrid neural network model that has completed model training to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. A contrastive diffusion bridge is constructed in the regulatory niche network, a multimodal conditional diffusion bridge is constructed in the cellular niche network, and an optimal transport flow matching diffusion bridge is constructed in the cell communication prediction network. Using the target slice spatial transcriptome data as anchor slices, spatial flow matching and alignment are performed to obtain alignment results. Based on the alignment results, the target slice spatial transcriptome data is reconstructed to obtain the gene expression map of the tissue genes.

2. The method according to claim 1, characterized in that, Before processing the raw transcriptome data using the hybrid neural network model that has completed model training to obtain the target slice spatial transcriptome data, the method further includes: Gene samples are obtained, and the network start point and network end point are determined based on the gene samples. The gene samples include original transcriptome samples that match transcriptome samples from different slice spaces. A regulatory niche network is constructed, and a contrastive diffusion bridge including the network start point and the network end point is constructed based on a conditional variational autoencoder. The contrastive diffusion bridge includes a first forward diffusion process and a first reverse diffusion process. A cell niche network is constructed, and a multimodal conditional diffusion bridge including the network start point and network end point is constructed based on a deep neural network. The multimodal conditional diffusion bridge includes a second forward diffusion process and a second reverse diffusion process. A cell communication prediction network is constructed, and an optimal transport flow matching diffusion bridge containing the network start point and network end point is constructed based on the diffusion bridge model. The optimal transport flow matching diffusion bridge includes a third forward diffusion process and a third reverse diffusion process. Based on the gene samples, the regulatory niche network, the cellular niche network, and the cell communication prediction network are trained together to obtain a hybrid neural network model.

3. The method according to claim 2, characterized in that, The hybrid neural network model obtained by performing mixed training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: An enhanced transcriptome sample is obtained by fitting an objective function based on a graph neural network and determining the regulatory features of the original transcriptome sample based on the objective function. The graph neural network includes a position encoding layer, a multi-head self-attention layer, a fully connected feedforward network layer, and a normalization layer. Based on the enhanced transcriptome samples, conditional constraints are added during the first reverse diffusion process, and a first model loss function is determined based on learnable parameters to be applied to the regulatory niche network for hybrid training.

4. The method according to claim 2, characterized in that, The hybrid neural network model obtained by performing mixed training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: A second model loss function is constructed based on single-cell conditional diffusion bridge loss, spatial omics conditional diffusion bridge loss, and contrastive diffusion bridge loss, and applied to the cellular niche network for hybrid training.

5. The method according to claim 3, characterized in that, The hybrid neural network model obtained by performing mixed training on the regulatory niche network, the cellular niche network, and the cell communication prediction network based on the gene samples includes: A third model loss function is constructed based on single-cell flow matching loss and spatial omics flow matching loss; Perturbation parameters are configured based on ligand receptor activity samples in the gene samples, and communication strength parameters are determined based on the perturbation parameters to perform hybrid training on the cell communication prediction network.

6. The method according to claim 1, characterized in that, The spatial flow matching and alignment using the target slice spatial transcriptome data as anchor slices yields the following alignment results: Using the target slice spatial transcriptome data as anchor slices, the point similarity between the anchor slices and the target slices is determined, and a coordinate set corresponding to the anchor slices and the target slices is constructed. The point similarity is used to construct the loss function of the flow matching model. The coordinate set is aligned based on the flow matching model that has completed model training to obtain the alignment result.

7. The method according to claim 6, characterized in that, The process of reconstructing the spatial transcriptome data of the target slice based on the alignment results to obtain the gene expression map of the tissue includes: The flow matching model is iteratively trained based on the alignment results, and the model training of the flow matching model is completed when the number of training times matches a preset iteration threshold. Based on the flow matching model that has completed model training, the spatial transcriptome data of the target slice is reconstructed to obtain the gene expression map of the tissue genes.

8. A gene expression map generation device based on spatial omics hierarchical reconstruction, characterized in that, include: The acquisition module is used to acquire raw transcriptome data of tissue genes; The processing module is used to process the original transcriptome data based on the hybrid neural network model that has been trained to obtain target slice spatial transcriptome data. The hybrid neural network model includes a regulatory niche network, a cellular niche network, and a cell communication prediction network. The regulatory niche network has a contrastive diffusion bridge constructed in it, the cellular niche network has a multimodal conditional diffusion bridge constructed in it, and the cell communication prediction network has an optimal transport flow matching diffusion bridge constructed in it. The reconstruction module is used to perform spatial flow matching and alignment using the target slice spatial transcriptome data as anchor slices, obtain alignment results, and reconstruct the target slice spatial transcriptome data based on the alignment results to obtain the gene expression map of the tissue genes.

9. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method of claim 1.

Images