Virtual organ simulation method and system based on three-dimensional stacking space transcriptome data

Abstract

The invention provides a virtual organ simulation method and system based on three-dimensional stacked space transcriptome data. The method comprises the steps that a space adjacency network is constructed according to coordinate information of multiple pieces of space transcriptome slice data; performing expression embedding expression learning on each point location in the spatial adjacency network by adopting a graph attention auto-encoder; voxelization processing is carried out on the three-dimensional coordinates of all the slices, a continuum is constructed through Gaussian smoothing, and a three-dimensional morphological model is generated; training the potential spatial diffusion model to generate a potential expression of the spatial transcriptome data; and for given coordinates and conditions, sampling from the trained potential spatial diffusion model to generate an embedded vector, and inputting the embedded vector into a graph attention auto-encoder to obtain a gene expression profile of a corresponding point location. The method has the advantages that unified modeling and expression generation are carried out on the space transcriptome of the whole organ or even a cross-tissue region, and the small-scale modeling capability that an existing method is only limited to a local region or a single slice is broken through.

Description

Technical Field

[0001] This application belongs to the fields of bioinformatics and spatial omics, specifically relating to a virtual organ simulation method and system based on three-dimensional stacked spatial transcriptome data, used to construct tissue and organ models with continuous three-dimensional structures. This technology can be widely applied to scenarios such as the generation, completion, super-resolution reconstruction, and virtual experiments of spatial transcriptome data, and has the potential to contribute to disease research, drug development, and the elucidation of spatial molecular mechanisms. Background Technology

[0002] In recent years, spatial transcriptomics (ST) technology has developed rapidly, enabling the in-situ detection of expression levels of thousands of genes in a spatial coordinate system. Typical techniques such as 10x Genomics Visium, Slide-seq, Stereo-seq, and Slide-tags have advanced spatial resolution from sub-tissue scales to subcellular and single-nucleus resolutions. These technologies have driven the development of space biology, supporting researchers in analyzing tissue structure, intercellular communication, spatial heterogeneity, and other issues, and have spurred the development of numerous computational tools for spatially variable gene detection, spatial partitioning, cell atlas construction, and three-dimensional slice integration.

[0003] Nevertheless, current 3D tissue-level modeling based on real ST data still faces significant challenges, mainly in the following aspects:

[0004] High-resolution 3D reconstruction capabilities are limited. Current methods attempt to reconstruct 3D spatial structures from 2D slice data. For example, stitch3D uses graph matching strategies for inter-slice registration, but lacks generative capabilities, making it difficult to handle missing slices or reconstruct complete organ structures. STAGE, on the other hand, uses a position-supervised autoencoder to interpolate and generate high-density expression data between slices, possessing some expression completion capabilities, but lacks controllable modeling of tissue structures and scalability for large-scale 3D expression generation.

[0005] There is a lack of generative simulators for 3D tissue modeling. Some studies have explored the simulated generation of spatial transcriptome data. For example, ZINB-WaVE incorporates spatial coordinates as covariates into a zero-inflated negative binomial distribution for modeling; scDesign3 uses a copula model to jointly simulate single-cell and spatial omics data; and SRTsim rearranges gene expression order to preserve spatial structure. While these methods support spatial data simulation to some extent, they generally have several limitations: they rely on strong distribution assumptions (such as Poisson or negative binomial distributions), making it difficult to fit complex real spatial structures; they only support 2D single-slice modeling and lack 3D simulation capabilities; they lack user controllability, making it difficult to support specific tissue structure settings and label control; and they lack scalability in handling large-scale data, making it difficult to apply to modeling tasks at the scale of millions of spatial points or organs.

[0006] There is a lack of controllable generative platforms to support virtual experiments. Most existing simulators are designed for the development and evaluation of computational methods, lacking support for application scenarios, particularly in areas such as virtual tissue generation and structural perturbation simulation. As spatial transcriptomics technology evolves towards "digital organs" or "tissue digital twins," researchers urgently need a simulation platform with capabilities for expression completion, structure generation, conditional control, and interactive operation. Currently, no publicly available tools meet these needs, especially in tasks such as cross-slice 3D integration and tissue-scale expression generation, where a general and scalable generative solution is lacking. Summary of the Invention

[0007] The purpose of this application is to overcome the shortcomings of existing spatial transcriptome data in constructing structurally continuous, fully expressed, and controllably generated three-dimensional organ-level models.

[0008] To achieve the above objectives, this application proposes a virtual organ simulation method based on three-dimensional stacked spatial transcriptome data, comprising:

[0009] Step 1: Construct a spatial adjacency network based on the coordinate information of several continuous or non-continuous spatial transcriptome slice data;

[0010] Step 2: Use a graph attention autoencoder to learn the embedding representation of each point in the spatial adjacency network;

[0011] Step 3: Voxelize the three-dimensional coordinates of each slice, construct a continuum using Gaussian smoothing, and generate a complete three-dimensional morphological model;

[0012] Step 4: Train the potential spatial diffusion model using a three-dimensional morphological model to generate the potential expression of spatial transcriptome data;

[0013] Step 5: For given coordinates and conditions, sample and generate embedding vectors from the trained latent spatial diffusion model, input the embedding vectors into the graph attention autoencoder, and obtain the gene expression profile of the corresponding location.

[0014] As an improvement to the above method, step 2 further includes:

[0015] For data from multiple slices, a triplet contrastive learning mechanism is introduced, which sets positive and negative samples to train the network to align the embedding representations across slice points.

[0016] As an improvement to the above method, step 3 further includes:

[0017] The slides are aligned and registered manually or automatically to mark the structural cavities inside the organ.

[0018] As an improvement to the above method, the potential spatial diffusion model is based on the U-Net architecture and includes a forward process and a backward process;

[0019] The forward process proceeds by progressively moving towards the potential expression Adding noise creates the following, represented as:

[0020] ;

[0021] in, This represents the potential expression at step t; Indicates noise control parameters; Indicates standard Gaussian noise;

[0022] The backward process is passed through a noise reduction network. The inverse reconstruction of noise is expressed as:

[0023] ;

[0024] in, For control items; Encoding spatial location;

[0025] The location encoding of the potential spatial diffusion model is a fractal location encoding, represented as:

[0026] ;

[0027] in, For position encoding; Embedded for time steps; Embed for optional conditional tags; and These are the spatial codes for the xy plane and the z-axis, respectively; the xy plane is the plane where the slice is located, and the z-axis is perpendicular to the xy plane.

[0028] As an improvement to the above method, it also includes:

[0029] Gene expression profiles at each location are used to generate three-dimensional virtual space expression images.

[0030] This application also provides a virtual organ simulation system based on three-dimensional stacked spatial transcriptome data, implemented using the above method, the system comprising:

[0031] A spatial adjacency network module is constructed to build a spatial adjacency network based on the coordinate information of several continuous or non-continuous spatial transcriptome slice data.

[0032] The training graph attention autoencoder module is used to learn the embedding representation of each point in the spatial adjacency network using a graph attention autoencoder.

[0033] The module for generating a 3D morphological model is used to voxelize the 3D coordinates of each slice, construct a continuum through Gaussian smoothing, and generate a complete 3D morphological model.

[0034] A potential expression generation module is used to train a potential spatial diffusion model using a three-dimensional morphological model to generate potential expressions from spatial transcriptome data.

[0035] The gene expression profile generation module is used to sample and generate embedding vectors from a trained latent spatial diffusion model for given coordinates and conditions. The embedding vectors are then input into a graph attention autoencoder to obtain the gene expression profiles at the corresponding locations.

[0036] Compared with existing technologies, the advantages of this application are:

[0037] 1. Supports high-resolution continuous 3D spatial modeling, scalable to the organ scale. This invention proposes a modeling framework that can fuse spatial information from multiple slices to reconstruct continuous 3D tissue structures. By introducing spatial adjacency graphs and voxel-level contour interpolation mechanisms, the spatial context between 2D slices is effectively integrated, achieving continuous reconstruction of 3D structures. Simultaneously, the system possesses excellent scalability, adapting to millions of spatial point datasets and supporting unified modeling and expression generation of spatial transcriptomes for entire organs or even cross-tissue regions, overcoming the limitations of existing methods that are limited to small-scale modeling of local regions or single slices.

[0038] 2. A controllable and interactive 3D representation generation simulation has been achieved. Compared to traditional representation simulation methods that rely on fixed distribution assumptions (such as Poisson or negative binomial distribution), this invention employs a latent diffusion model without distribution assumptions, which can more accurately capture the nonlinear changes and complex covariant structures of real-world spatial representations. The designed 3D spatial denoising network integrates fractal position encoding and a self-attention mechanism, which can flexibly receive structural labels or spatial conditions, thereby achieving controllable representation generation under specific regions, structural perturbations, and cross-section selection scenarios. Furthermore, the accompanying 3D visualization module supports user-defined virtual cross-section input, facilitating downstream tasks such as computer experimental simulations and tissue-level intervention testing.

[0039] 3. Overcoming the limitations of 2D modeling and static simulation, this invention provides a tissue-level generative platform. Current methods mostly remain at the 2D level, lacking 3D modeling capabilities and expression continuity control, making it difficult to support systematic modeling and multi-scale expression reconstruction of spatial transcriptomes in the "digital organ" scenario. In contrast, this invention constructs a 3D tissue structural framework by stacking slices and uses a generative model to interpolate, complete, and constrain structural conditions of expression information, realizing the transformation from "local slice data" to "complete 3D tissue model." This provides a unified, highly controllable, and scalable generative solution for tasks such as tissue digital twins, pathological modeling, and structural simulation. Attached Figure Description

[0040] Figure 1 The diagram shows the overall computational framework of a virtual organ simulation method based on three-dimensional stacked spatial transcriptome data. This framework integrates three-dimensional spatial coordinate information in both the expression representation learning and generative modeling stages, which can effectively maintain the spatial continuity between slices and realize the three-dimensional gene expression reconstruction and simulation generation at the tissue scale.

[0041] Figure 2 The diagram shown is a flowchart illustrating the technical implementation of this invention; it demonstrates the complete technical path from multi-slice spatial transcriptome data input, spatial adjacency construction, expression embedding learning, latent spatial diffusion modeling, decoding generation to three-dimensional expression output;

[0042] Figure 3 The diagram shown is a schematic of the system application in Example 1. Based on spatial transcriptome slices of 33 marmoset cerebellum, a complete three-dimensional cerebellum expression atlas was constructed. This atlas supports cross-sectional reconstruction from any viewpoint, enabling visualization of the spatial expression of complex brain region structures.

[0043] Figure 4 The diagram shows the system architecture supporting 3D visualization and representation analysis. It adopts a front-end and back-end separation design, integrating functional modules such as representation prediction, cross-section generation, and map rendering, providing users with a flexible 3D interactive analysis platform.

[0044] Figure 5 The diagram shows the operation flowchart for 3D visualization and expression analysis support; the red box marks the user input area, which allows users to specify cross-section parameters (such as cross-section position and normal vector) online and generate corresponding virtual slices and expression maps in real time to help explore the spatial heterogeneity and biological laws in the cerebellum of marmosets;

[0045] Figure 6 The image shows a simulated generation of three-dimensional marmoset cerebellar slices; it displays virtual slices generated by the system in the coronal, sagittal, and arbitrary angles; each column is labeled with the normal vector n of the slice; the four images on the far right are gene expression heatmaps of representative cell types in the cortex (white matter, granular layer, Purkinje cell layer, and molecular layer), demonstrating the ability to accurately model complex spatial patterns. Detailed Implementation

[0046] The technical solution of this application will be described in detail below with reference to the accompanying drawings.

[0047] like Figure 1 and Figure 2 As shown, this application provides a method and system for virtual organ simulation based on three-dimensional stacked spatial transcriptome data. This method can be used to generate high-fidelity, structurally continuous, and label-controllable three-dimensional spatial expression data, suitable for applications such as expression completion, data augmentation, heterogeneity simulation, virtual experiments, and spatial map visualization. The method includes:

[0048] Step 1: Input several continuous or non-continuous spatial transcriptome slices. Each slice contains the spatial coordinates of a location and its corresponding gene expression matrix. First, construct a Spatial Neighbor Network (SNN) based on the two-dimensional or three-dimensional coordinate information of the locations within each slice to capture local spatial structural relationships. The graph structure is defined as follows:

[0049]

[0050] in, Indicates the first The three-dimensional coordinates of each point. The radius of the defined spatial domain.

[0051] Step 2: Based on the constructed graph structure, a graph attention autoencoder is used to learn the embedding representation for each point. The model structure includes an encoder, decoder, and attention mechanism, which can model the covariant structure of representations under spatial adjacency. For data from multiple slices, a triple contrastive learning mechanism is introduced, setting positive and negative samples to train the network to align the embedding representations of points across slices, thereby mitigating batch effects and achieving spatial continuity.

[0052] Step 3: 3D Geometric Reconstruction: The 3D coordinates of each slice are voxelized and a continuum is constructed using Gaussian smoothing. This is converted to an intermediate format supporting multi-dimensional spatial data storage for subsequent processing. Slices are aligned and registered manually or automatically, and structural cavities within the organ (such as fissures and cavities) are labeled. Subsequently, contour interpolation is applied along the slice axes to generate a complete 3D morphological model, which serves as the geometric constraint and visualization reference for subsequent generation and expression.

[0053] Step 4: Generative Modeling Based on Diffusion Model: This invention employs a latent spatial diffusion model to achieve high-quality generation of latent expressions from spatial transcriptome data. The forward and reverse processes of the diffusion model follow existing techniques, progressively moving towards latent expressions. Adding noise creates:

[0054]

[0055] in, These are noise control parameters. This is standard Gaussian noise. And through a denoising network... The process of reverse noise restoration:

[0056]

[0057] in, For control items, Encoding spatial location; ; One of the innovations of this invention lies in the design of a spatial denoising network specifically for three-dimensional space. This network, based on the U-Net architecture, integrates a multi-layer self-attention mechanism to effectively capture spatial context information. To enhance the accurate modeling of 3D spatial representation, a fractal position encoding is introduced, specifically represented as:

[0058]

[0059] in, Embedded for time steps, For optional conditional tag embedding, and These represent spatial encodings for the xy-plane and z-axis, respectively. This composite embedding, after processing by a multilayer perceptron, is injected into each layer of the U-Net to achieve efficient fusion and representation of 3D spatial conditions. The training process employs standard backpropagation and mainstream optimization algorithms (such as Adam), with model parameters optimized end-to-end by minimizing prediction or reconstruction errors.

[0060] Step 5: Decoding and 3D Generation of Representation Output: Given coordinates and conditions (optional), sample and generate embeddings from the trained 3D spatial diffusion model. The final generated embedding vector The data is input into a trained decoder to obtain the gene expression profile at the corresponding location. By combining spatial coordinate input, virtual spatial representation data can be simulated and generated at any three-dimensional location, enabling representation interpolation completion or new structure prediction.

[0061] Step 6: 3D Visualization: Users can specify any spatial plane (defined by a point and a normal vector), calculate the intersection of this plane and the 3D model, and generate a virtual spatial representation image on the cross-section. 3D visualization can be used for structural analysis, tissue anatomy, and computer simulation (in silico) experiments.

[0062] Example 1

[0063] Using a virtual organ simulation method based on three-dimensional stacked spatial transcriptome data, a three-dimensional expression generation model was created based on 3D expression generation from 33 coronal slices of the cerebellum of marmosets.

[0064] like Figure 3 As shown, this embodiment uses the cerebellum of a marmoset as the research object, with spatial transcriptome data from 33 coronal sections as input. The section spacing is 250 μm, and the single-point resolution is 25 μm, exhibiting significant spatial anisotropy. This embodiment demonstrates the ability of this invention to achieve three-dimensional expression modeling and virtual section generation under large-scale, high-sparse data conditions.

[0065] (1) Data preparation and spatial alignment

[0066] After standard preprocessing, the raw data yielded a total of 1,919,455 spatial locations and 2,371 hypervariable genes. To establish a three-dimensional coordinate system, the open-source tool 3D Slicer was used to perform coarse registration on 33 slices, initially reconstructing the three-dimensional anatomical structure of the cerebellum and extracting the three-dimensional spatial coordinates of each location.

[0067] (2) Spatial adjacency graph construction

[0068] For each slice, a spatial adjacency graph is constructed based on the 3D coordinates of its points. Points with an Euclidean distance less than a set radius δ = 75 μm are considered adjacent. The adjacency graph is stored as a sparse adjacency matrix for subsequent use by the graph neural network.

[0069] (3) Representation embedding learning and cross-slice alignment

[0070] A graph attention autoencoder is used as the representation embedding model, and its structure includes:

[0071] The encoder consists of a three-layer graph attention network, with each layer having an output dimension of 128, 64, and 32.

[0072] Decoder: Based on dual tasks of adjacency reconstruction and representation reconstruction;

[0073] Loss function: includes reconstruction error and triple contrast loss, the latter used to align similar regions across slices;

[0074] Training method: The Adam optimizer was used, with an initial learning rate of 1e-3, a training batch size of 512, and 300 iterations.

[0075] (4) Three-dimensional geometric modeling and organizational structure reconstruction

[0076] The aligned 3D coordinates were voxelized (25 μm³ resolution) and spatially smoothed using a Gaussian kernel. Contour reconstruction of missing regions between slices was performed using an interpolation algorithm to form a continuous cerebellar volume model. To adapt to the representation generation task, this model was converted into a structurally constrained mesh and stored in a standard intermediate format.

[0077] (5) Diffusion model modeling and expression generation

[0078] This invention employs a three-dimensional latent space diffusion model to learn the generation distribution of representation data. Key components are as follows:

[0079] Forward process: Gaussian noise is gradually added to the embedding space to generate a perturbation trajectory;

[0080] Denoising network: Based on the 3D U-Net architecture, it integrates multi-layer self-attention modules. The specific architecture is shown in the table below.

[0081] Position encoding: Fractal position embedding is introduced and injected into each layer of U-Net to enhance 3D conditional modeling;

[0082] Training method: The optimization objective is L2 reconstruction error, the training epochs are 800, the Adam optimizer is used, the initial learning rate is 1e-4, and the learning rate is cosine decay.

[0083] Model enter Architecture aisle parameter 3D U-Net Architecture 1×64 NoSkip → Attn → Attn → Attn →Attn → Attn → Attn → NoSkip 64 → 128 → 128→ 256 ~11.81M

[0084] Where NoSkip represents a convolutional layer without skip-connections; Attn represents a layer using a self-attention mechanism (Attention). In other embodiments, the number of Attn layers can be selected appropriately based on the complexity of the data.

[0085] (6) Virtual slice generation and expression decoding

[0086] After training, latent embeddings are sampled from the diffusion model and input into the decoder to generate the corresponding gene expression profile. This invention supports conditional sampling at any three-dimensional spatial point to generate virtual representation data, which is then automatically rendered as virtual slice images.

[0087] In this embodiment, the system automatically generated 321 virtual coronal slices with an interval of 1 / 10 of the original resolution, achieving three-dimensional representation prediction and reconstruction of more than 20 million spatial points, demonstrating the effectiveness of the method in terms of spatial continuity and large-scale representation completion.

[0088] (7) 3D visualization and analysis support

[0089] like Figure 4 and Figure 5 As shown, to support the visualization and interactive access of the results generated by this invention, this embodiment deploys a three-dimensional visualization module based on a front-end and back-end separation architecture, which is used to display the reconstructed organizational structure and support the generation and rendering of expressions at arbitrary spatial locations.

[0090] The front-end module uses a web visualization framework to build a 3D interactive interface, loads a 3D tissue model, and supports user input of custom cross-sectional parameters (including the center position and normal vector of the cross-section). The interface supports functions such as heatmap display and gene channel switching.

[0091] Backend module: After receiving the frontend request, it calls the trained diffusion model on the server side, generates the expression spectrum of the corresponding point based on the input space cross section, and performs preliminary processing (such as standardization, compression, and color mapping).

[0092] Data communication mechanism: The front-end and back-end achieve low-latency data exchange through a lightweight asynchronous communication middleware, supporting high-concurrency request and response; the middleware plays a coordinating role in request queue management, cache control and data transmission generation.

[0093] Acceleration and Computing Resources: Expression generation tasks can run on GPU-enabled computing nodes, improving response efficiency through resource scheduling strategies and adapting to multi-user interaction scenarios.

[0094] After users select or draw any 3D cross-section in the front-end interface, the system performs spatial point extraction, representation generation, and image rendering in the background, returning the results to the front-end in real time, achieving a "what you see is what you get" 3D representation visualization experience. This module provides the invention with complete interactive representation generation and organization exploration capabilities, which can be widely used in tissue structure analysis and virtual imaging simulation.

[0095] Example 2

[0096] This application also provides a virtual organ simulation system based on three-dimensional stacked spatial transcriptome data, implemented using the above method, the system comprising:

[0097] A spatial adjacency network module is constructed to build a spatial adjacency network based on the coordinate information of several continuous or non-continuous spatial transcriptome slice data.

[0098] The training graph attention autoencoder module is used to learn the embedding representation of each point in the spatial adjacency network using a graph attention autoencoder.

[0099] The module for generating a 3D morphological model is used to voxelize the 3D coordinates of each slice, construct a continuum through Gaussian smoothing, and generate a complete 3D morphological model.

[0100] A potential expression generation module is used to train a potential spatial diffusion model using a three-dimensional morphological model to generate potential expressions from spatial transcriptome data.

[0101] The gene expression profile generation module is used to sample and generate embedding vectors from a trained latent spatial diffusion model for given coordinates and conditions. The embedding vectors are then input into a graph attention autoencoder to obtain the gene expression profiles at the corresponding locations.

[0102] This application may also provide a computer device, including: at least one processor, memory, at least one network interface, and a user interface. The various components in this device are coupled together via a bus system. It is understood that the bus system is used to implement communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus.

[0103] The user interface can include a display, keyboard, or clicking device. Examples include a mouse, trackball, touchpad, or touchscreen.

[0104] It is understood that the memory in the embodiments disclosed in this application may be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory may be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memories described herein are intended to include, but are not limited to, these and any other suitable types of memory.

[0105] In some implementations, the memory stores elements such as executable modules or data structures, or subsets thereof, or extended sets thereof: operating systems and applications.

[0106] The operating system includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application programs include various applications, such as media players and browsers, used to implement various application functions. Programs implementing the methods of the embodiments of this disclosure can be included in the application programs.

[0107] In the above embodiments, the processor can also invoke programs or instructions stored in memory, specifically programs or instructions stored in an application program, for the following purposes:

[0108] Follow the steps described above.

[0109] The above methods can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by software instructions. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic diagrams disclosed above. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the disclosed methods can be directly implemented by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory; the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above methods.

[0110] It is understood that the embodiments described in this application can be implemented using hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in this application, or combinations thereof.

[0111] For software implementation, the technology of this application can be implemented by executing the functional modules (e.g., procedures, functions, etc.) of this application. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or outside the processor.

[0112] This application may also provide a non-volatile storage medium for storing a computer program. When the computer program is executed by a processor, it can implement the steps in the above method embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and are not intended to limit it. Although this application has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of this application do not depart from the spirit and scope of the technical solutions of this application, and should all be covered within the scope of the claims of this application.

Claims

1. A method for simulating virtual organs based on three-dimensional stacked spatial transcriptome data, comprising: Step 1: Construct a spatial adjacency network based on the coordinate information of several continuous or non-continuous spatial transcriptome slice data; Step 2: Use a graph attention autoencoder to learn the embedding representation of each point in the spatial adjacency network; Step 3: Voxelize the three-dimensional coordinates of each slice, construct a continuum using Gaussian smoothing, and generate a complete three-dimensional morphological model; Step 4: Train the potential spatial diffusion model using a three-dimensional morphological model to generate the potential expression of spatial transcriptome data; Step 5: For given coordinates and conditions, sample and generate embedding vectors from the trained latent spatial diffusion model, input the embedding vectors into the graph attention autoencoder, and obtain the gene expression profile of the corresponding location.

2. The virtual organ simulation method based on three-dimensional stacked spatial transcriptome data according to claim 1, characterized in that, Step 2 also includes: For data from multiple slices, a triplet contrastive learning mechanism is introduced, which sets positive and negative samples to train the network to align the embedding representations across slice points.

3. The virtual organ simulation method based on three-dimensional stacked spatial transcriptome data according to claim 1, characterized in that, Step 3 also includes: The slides are aligned and registered manually or automatically to mark the structural cavities inside the organ.

4. The virtual organ simulation method based on three-dimensional stacked spatial transcriptome data according to claim 1, characterized in that, The potential spatial diffusion model is based on the U-Net architecture and includes a forward process and a backward process; The forward process proceeds by progressively moving towards the potential expression Adding noise creates the following, represented as: ； in, This represents the potential expression at step t; Indicates noise control parameters; Indicates standard Gaussian noise; The backward process is passed through a noise reduction network. The inverse reconstruction of noise is expressed as: ； in, For control items; Encoding spatial location; ; ; The location encoding of the potential spatial diffusion model is a fractal location encoding, represented as: ； in, For position encoding; Embedded for time steps; Embed for optional conditional tags; and These are the spatial codes for the xy plane and the z-axis, respectively; the xy plane is the plane where the slice is located, and the z-axis is perpendicular to the xy plane.

5. The virtual organ simulation method based on three-dimensional stacked spatial transcriptome data according to claim 1, characterized in that, Also includes: Gene expression profiles at each location are used to generate three-dimensional virtual space expression images.

6. A virtual organ simulation system based on three-dimensional stacked spatial transcriptome data, implemented according to the method of any one of claims 1-5, characterized in that, The system includes: A spatial adjacency network module is constructed to build a spatial adjacency network based on the coordinate information of several continuous or non-continuous spatial transcriptome slice data. The training graph attention autoencoder module is used to learn the embedding representation of each point in the spatial adjacency network using a graph attention autoencoder. The module for generating a 3D morphological model is used to voxelize the 3D coordinates of each slice, construct a continuum through Gaussian smoothing, and generate a complete 3D morphological model. A latent expression generation module is used to train a latent spatial diffusion model using a 3D morphological model to generate latent expressions from spatial transcriptome data; and The gene expression profile generation module is used to sample and generate embedding vectors from a trained latent spatial diffusion model for given coordinates and conditions. The embedding vectors are then input into a graph attention autoencoder to obtain the gene expression profiles at the corresponding locations.

Images