A single-cell spatial multi-omics analysis method for tumor microenvironment research
By performing continuous slicing of tumor tissue and constructing a graph neural network model, the problem of multimodal data alignment was solved, enabling accurate causal inference and dynamic simulation of the tumor microenvironment, and providing high-precision support for personalized treatment.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANCHANG UNIV
- Filing Date
- 2026-05-08
- Publication Date
- 2026-07-31
AI Technical Summary
Traditional methods struggle to achieve accurate spatial alignment of multimodal data on the same tissue sample, making it difficult to establish causal relationships across scales, especially in highly heterogeneous tumor tissues where it is difficult to accurately infer cellular composition.
By performing continuous slicing of the same tumor tissue sample, physical imaging, spatial transcriptome sequencing, and spatial metabolite detection data are obtained. Combined with a graph neural network model, a digital twin of the tumor microenvironment is constructed to achieve accurate alignment of multidimensional information and causal inference.
It enables the alignment of physical, genetic, protein, and metabolic information at the subcellular scale, realistically reproducing the diffusion behavior of signaling molecules in a dense matrix, supporting virtual perturbations of specific cell subpopulations, and providing high-precision decision-making basis for personalized immunotherapy strategies.
Smart Images
Figure CN122493955A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of spatial multi-omics analysis technology, and in particular to a single-cell spatial multi-omics analysis method for tumor microenvironment research. Background Technology
[0002] Spatial multi-omics analysis technology refers to a class of cutting-edge biomedical analysis technologies that, while preserving the original spatial structure of tissues, simultaneously or sequentially acquire multi-dimensional molecular information such as gene expression, protein distribution, metabolite abundance, and physical microenvironment from the same biological sample, and then use computational methods to map these different levels of data into a unified spatial coordinate system, thereby achieving precise localization and integrated analysis of cell types, functional states, and their interactions in the tissue microenvironment.
[0003] Traditional methods typically detect different molecular layers independently, lacking the ability to accurately align multimodal data in spatial location on the same tissue sample. This makes it difficult to establish causal relationships across scales. Furthermore, most spatial omics platforms are limited by resolution, making it difficult to accurately infer the cellular composition of each region while maintaining spatial location, especially in highly heterogeneous tumor tissues. Summary of the Invention
[0004] In view of the aforementioned existing problems, the present invention is proposed.
[0005] Therefore, this invention provides a single-cell spatial multi-omics analysis method for tumor microenvironment research. This solves the problem that traditional methods usually detect different molecular layers independently, lack the ability to accurately align multimodal data in spatial location on the same tissue sample, making it difficult to establish cross-scale causal relationships. Furthermore, most spatial omics platforms are limited by resolution, making it difficult to accurately infer the cellular composition of each region while maintaining spatial location, especially in highly heterogeneous tumor tissues.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a single-cell spatial multi-omics analysis method for tumor microenvironment research, comprising: Serial sections were prepared from the same tumor tissue sample to obtain tissue sections for physical imaging, spatial transcriptome sequencing, spatial protein detection, and spatial metabolite detection. High-resolution imaging of physical imaging sections and quantification of physical property parameters of the extracellular matrix are performed to generate spatially resolved physical microenvironment maps. Spatial transcriptome sections, spatial protein sections, and spatial metabolite sections were analyzed to obtain corresponding spatially resolved gene expression maps, protein activity maps, and metabolite abundance maps. Physical microenvironment maps, gene expression maps, protein activity maps, and metabolite abundance maps are registered to a unified spatial coordinate system based on tissue morphology and structure, so that data at the same spatial location correspond to each other. In each spatial unit of a unified spatial coordinate system, corresponding physical property parameters, gene expression data, protein activity data, and metabolite abundance data are fused to construct a multidimensional feature vector. Cell type deconvolution is then performed in conjunction with single-cell transcriptome reference data to obtain cell composition information within the spatial unit. Using each spatial unit as a node, multidimensional feature vectors as node features, and spatial proximity and physical connectivity as edge relationships, a graph neural network model is constructed and trained to form a digital twin of the tumor microenvironment, which is used to perform virtual perturbations and quantify the causal regulatory effects of specific cell subpopulations on the immune microenvironment.
[0007] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the specific steps of performing high-resolution imaging of physical imaging sections and quantifying the physical property parameters of the extracellular matrix to generate a spatially resolved physical microenvironment map are as follows: The physical imaging slices were scanned using a second harmonic generation microscopy system to obtain nonlinear optical signal images of collagen fibers. The signal image is subjected to background correction and contrast enhancement processing to obtain a preprocessed image; In the preprocessed image, multiple local regions are slidably segmented with a fixed window size, and a skeletonization algorithm is performed on each local region to extract the fiber centerline; The number of fiber intersections per unit area is calculated based on the centerline set, and used as an indicator of network complexity. Construct an orientation histogram for the set of centerline orientation angles and calculate the orientation entropy. The expression is: ; in, This means dividing the interval from 0 to π radians into equal parts. A sub-interval of angles, Indicates the first The proportion of all centerline direction angles within each angle sub-interval; Network complexity metrics and orientation entropy Together with the local porosity estimated by morphological opening operations, they form a vector of physical property parameters, which is mapped to the corresponding spatial coordinates to form a physical microenvironment map.
[0008] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the skeletonization algorithm adopts an iterative refinement method, removing boundary pixels that meet the preset endpoint retention conditions in each iteration until all fiber structures converge to a single-pixel width centerline while retaining the original topological connectivity; the endpoint retention conditions include prohibiting deletion if the current pixel is a fiber endpoint or located at a three-way or higher connection.
[0009] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the specific steps for registering the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map to a unified spatial coordinate system based on tissue morphology and structure are as follows: For each functional slice, an H&E stained slice was prepared at the adjacent position as a morphological reference. Tissue contours, vascular cavities, and densely populated areas of cell nuclei in H&E images were extracted as feature anchors. Low-resolution bright-field images were extracted from the corresponding slices of the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map. An initial affine transformation model is established using feature anchor points, and then a non-rigid registration algorithm based on B-splines is used to optimize the local deformation field, so as to minimize the spatial coordinate deviation of the same anatomical structure in each modal image. The optimized deformation field is applied to each functional map, and a four-modal aligned dataset in a unified spatial coordinate system is output. The non-rigid registration algorithm achieves multimodal image alignment by minimizing the mutual information loss function, ensuring spatial consistency of data at different molecular levels at the subcellular scale.
[0010] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the steps of fusing corresponding physical property parameters, gene expression data, protein activity data, and metabolite abundance data in each spatial unit of a unified spatial coordinate system to construct a multidimensional feature vector, and then performing cell type deconvolution in conjunction with single-cell transcriptome reference data to obtain cell composition information within the spatial unit are as follows: Define the spatial unit as a square grid with fixed side lengths in a unified coordinate system; For each spatial unit, the measured values of all modes within it are extracted and standardized, and then concatenated to form a multidimensional feature vector; Load an externally available tumor single-cell transcriptome reference dataset and label cluster centers by cell type; Solve the constrained optimization problem so that the gene expression part of the spatial unit is linearly approximated by a weighted combination of reference cell types as much as possible; The output weight vector serves as the cellular composition information of the spatial unit. The objective function of the constrained optimization problem introduces non-negativity and normalization constraints to ensure that the deconvolution result has biological interpretability.
[0011] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the specific steps for constructing and training a graph neural network model to form a digital twin of the tumor microenvironment, using each spatial unit as a node, multidimensional feature vectors as node features, and spatial proximity and physical connectivity as edge relationships, are as follows: All spatial units are indexed into a set of nodes. For any two nodes, if the Euclidean distance is less than a preset threshold, an edge is created, and the edge weight is calculated. for: ; in, Represents a node With nodes Euclidean distance in a unified spatial coordinate system and Representing nodes respectively With nodes The corresponding local porosity, This is the balance coefficient; Constructing a graph structure based on node sets and edge weight matrices; A graph attention network architecture is used for multi-layer message passing, and each layer updates the node representation to integrate neighborhood micro-environment information. Using the activation state of immune cells in real samples as a supervisory signal, the model parameters are trained so that the output layer can predict changes in immune function indicators under perturbation conditions.
[0012] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the specific steps of performing virtual perturbation and quantifying the causal regulatory effects of specific cell subpopulations on the immune microenvironment are as follows: Identify spatial units where the proportion of COL11A1-positive cancer-associated fibroblasts exceeds a set threshold; In the input layer of the graph neural network, the COL11A1 expression feature value of all nodes in the set is set to zero, and the stiffness parameter in its physical microenvironment map is multiplied by a decay factor, with the decay factor being less than 1. Run the trained graph neural network model to obtain the predicted expression values of IFNG and GZMB of the nodes where CD8 positive T cells are located; Calculate the difference between the IFNG and GZMB predicted values before and after the disturbance; The difference was used as a quantitative indicator of the causal regulatory effect of COL11A1-positive cancer-associated fibroblasts on T cell function, and was used to generate subsequent personalized treatment strategies.
[0013] As a preferred embodiment of the single-cell spatial multi-omics analysis method for tumor microenvironment research described in this invention, the attenuation factor is determined through the power-law relationship between porosity and stiffness, and the specific steps are as follows: Based on the physical microenvironment maps of multiple tumor samples, local porosity was obtained. With corresponding stiffness parameters The paired data; Fitting empirical relation: ; in, and These are the positive real coefficients calibrated using the least squares method; In virtual perturbation, the porosity of the target spatial cell is changed from... Increase to And calculate the stiffness after disturbance. : ; Attenuation factor Defined as: ; Update the stiffness characteristics of the spatial element to ; In the formula, Indicates the original local porosity. This indicates the porosity set after the disturbance. Represents the original stiffness parameters. This represents the post-disturbance stiffness calculated based on the model. This is the attenuation factor used to scale physical characteristics.
[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the single-cell spatial multi-omics analysis method for tumor microenvironment research as described in the first aspect of the present invention.
[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the single-cell spatial multi-omics analysis method for tumor microenvironment research as described in the first aspect of the present invention.
[0016] The beneficial effects of this invention are as follows: By integrating physical microenvironment imaging and multi-omics spatial data, a digital twin of the tumor microenvironment with causal inference capability is constructed, realizing the leap from static description to dynamic simulation. Moreover, the method not only accurately aligns physical, genetic, protein and metabolic four-dimensional information at the subcellular scale, but also realistically restores the restricted diffusion behavior of signaling molecules in dense matrix by designing graph neural network edge weights that integrate porosity and spatial proximity. It also supports virtual perturbation of specific stromal cells and quantitatively assesses their inhibitory effect on T cell function, thereby providing a high-precision, calculable and mechanism-clear decision-making basis for the formulation of personalized immunotherapy strategies. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a flowchart of the single-cell spatial multi-omics analysis method used for tumor microenvironment research in Example 1. Detailed Implementation
[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.
[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.
[0022] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a single-cell spatial multi-omics analysis method for tumor microenvironment research, comprising the following steps: S1. Serial sections are prepared from the same tumor tissue sample to obtain tissue sections for physical imaging, spatial transcriptome sequencing, spatial protein detection, and spatial metabolite detection.
[0023] Furthermore, the slices are prepared alternately in a preset order, with each functional slice having a morphological reference slice preserved adjacent to it; different modal slices are prepared using appropriate slide types and preservation methods according to the detection requirements to maintain their molecular or structural integrity; all slices are recorded in their relative positions in the original tissue during the preparation process to ensure that the spatial correspondence can be reconstructed subsequently.
[0024] It should be noted that by alternating the arrangement of slices according to functional requirements in continuous slices and retaining morphological reference slices, combined with a systematic recording and adaptive preservation strategy of slice positions, the consistency of spatial topology and the integrity of molecular information of samples of different modalities are effectively guaranteed, laying a reliable foundation for high-precision alignment and fusion of subsequent multi-omics data.
[0025] S2. Perform high-resolution imaging on the physical imaging slices and quantify the physical property parameters of the extracellular matrix to generate a spatially resolved physical microenvironment map.
[0026] Furthermore, a second harmonic generation microscopy system was used to scan the physical imaging slices to obtain nonlinear optical signal images of collagen fibers. Background correction and contrast enhancement are performed on the signal image to obtain a preprocessed image; In the preprocessed image, multiple local regions are slidably segmented with a fixed window size, and a skeletonization algorithm is performed on each local region to extract the fiber centerline; The number of fiber intersections per unit area is calculated based on the centerline set, and used as an indicator of network complexity. Construct an orientation histogram for the set of centerline orientation angles and calculate the orientation entropy. The expression is: ; in, This means dividing the interval from 0 to π radians into equal parts. A sub-interval of angles, Indicates the first The proportion of all centerline direction angles within each angle sub-interval; Network complexity metrics and orientation entropy Together with the local porosity estimated by morphological opening operations, they form a vector of physical property parameters, which is mapped to the corresponding spatial coordinates to form a physical microenvironment map; The skeletonization algorithm uses an iterative refinement method. In each iteration, boundary pixels that meet the preset endpoint retention conditions are removed until all fiber structures converge to a single-pixel-width centerline while preserving the original topological connectivity. The endpoint retention conditions include prohibiting deletion if the current pixel is a fiber endpoint or located at a three-way or higher connection.
[0027] It should be noted that by performing high-resolution nonlinear optical imaging on collagen fiber structures, and combining image preprocessing, skeleton extraction, and multi-dimensional physical parameter quantification, the microscopic tissue characteristics of the extracellular matrix can be comprehensively characterized. The constructed physical microenvironment atlas truly reflects the spatial heterogeneity of the tumor matrix, providing key input for understanding the influence of physical factors on cell behavior.
[0028] S3. Spatial transcriptome slices, spatial protein slices, and spatial metabolite slices are analyzed to obtain corresponding spatially resolved gene expression maps, protein activity maps, and metabolite abundance maps.
[0029] Furthermore, each modal slice is captured using a matching high-throughput spatial detection technology, including in situ labeling and sequencing, multiplex protein imaging, or mass spectrometry. After detection, molecular information is mapped to a unified spatial coordinate framework through image processing and signal correction, and multi-omics maps are generated according to a consistent spatial unit division method, providing a foundation for subsequent data fusion.
[0030] It should be noted that by employing spatial detection techniques that match the characteristics of different molecular levels and performing signal mapping and standardization within a unified coordinate framework, the consistency and comparability of gene, protein, and metabolite information at spatial scales are ensured, thereby improving the accuracy and biological interpretability of multi-omics data fusion.
[0031] S4. Register the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map to a unified spatial coordinate system based on tissue morphology and structure, so that data at the same spatial location correspond to each other.
[0032] Furthermore, an H&E-stained section was prepared at the adjacent position of each functional section as a morphological reference; Tissue contours, vascular cavities, and densely populated areas of cell nuclei in H&E images were extracted as feature anchors. Low-resolution bright-field images were extracted from the corresponding slices of the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map. An initial affine transformation model is established using feature anchor points, and then a non-rigid registration algorithm based on B-splines is used to optimize the local deformation field, so as to minimize the spatial coordinate deviation of the same anatomical structure in each modal image. The optimized deformation field is applied to each functional map, and a four-modal aligned dataset in a unified spatial coordinate system is output. Non-rigid registration algorithms achieve multimodal image alignment by minimizing the mutual information loss function, ensuring spatial consistency of data at different molecular levels at the subcellular scale.
[0033] It should be noted that by using stable anatomical features extracted from adjacent H&E stained sections as anchor points and combining rigid and non-rigid registration strategies to optimize local deformation, the spatial misalignment problem caused by tissue deformation or section differences is effectively overcome, achieving precise alignment of cross-modal data on fine anatomical structures and providing spatial consistency assurance for multidimensional microenvironment analysis.
[0034] S5. In each spatial unit of a unified spatial coordinate system, the corresponding physical property parameters, gene expression data, protein activity data, and metabolite abundance data are fused to construct a multidimensional feature vector. Cell type deconvolution is then performed in conjunction with single-cell transcriptome reference data to obtain cell composition information within the spatial unit.
[0035] Furthermore, the spatial unit is defined as a square grid with fixed side lengths in a unified coordinate system; For each spatial unit, the measured values of all modes within it are extracted and standardized, and then concatenated to form a multidimensional feature vector; Load an externally available tumor single-cell transcriptome reference dataset and label cluster centers by cell type; Solve the constrained optimization problem so that the gene expression part of the spatial unit is linearly approximated by a weighted combination of reference cell types as much as possible; The output weight vector serves as the cellular composition information of the spatial unit. The objective function of the constraint optimization problem is constrained by introducing nonnegativity and normalization constraints to ensure that the deconvolution result has biological interpretability.
[0036] It should be noted that by integrating multimodal measurements within a unified spatial unit and introducing high-quality single-cell reference data for constrained deconvolution, not only is high-resolution inference of cell type composition achieved, but the results are also ensured to conform to biological priors. This allows for the accurate reconstruction of the distribution and proportion of various cells in the tumor microenvironment while preserving the spatial context.
[0037] S6. Using each spatial unit as a node, multidimensional feature vectors as node features, and spatial proximity and physical connectivity as edge relationships, construct and train a graph neural network model to form a digital twin of the tumor microenvironment, which is used to perform virtual perturbations and quantify the causal regulatory effects of specific cell subpopulations on the immune microenvironment.
[0038] Furthermore, all spatial units are indexed as a set of nodes. For any two nodes, if the Euclidean distance is less than a preset threshold, an edge is created, and the edge weight is calculated. for: ; in, Represents a node With nodes Euclidean distance in a unified spatial coordinate system and Representing nodes respectively With nodes The corresponding local porosity, This is the balance coefficient; Constructing a graph structure based on node sets and edge weight matrices; A graph attention network architecture is used for multi-layer message passing, and each layer updates the node representation to integrate neighborhood micro-environment information. Using the activation state of immune cells in real samples as a supervisory signal, the model parameters are trained so that the output layer can predict changes in immune function indicators under perturbation conditions. Identify spatial units where the proportion of COL11A1-positive cancer-associated fibroblasts exceeds a set threshold; In the input layer of the graph neural network, the COL11A1 expression feature value of all nodes in the set is set to zero, and the stiffness parameter in its physical microenvironment map is multiplied by a decay factor, with the decay factor being less than 1. Run the trained graph neural network model to obtain the predicted expression values of IFNG and GZMB of the nodes where CD8 positive T cells are located; Calculate the difference between the IFNG and GZMB predicted values before and after the disturbance; The difference was used as a quantitative indicator of the causal regulatory effect of COL11A1-positive cancer-associated fibroblasts on T cell function, and was used to generate subsequent personalized treatment strategies.
[0039] The attenuation factor is determined through the power-law relationship between porosity and stiffness, and the specific steps are as follows: Based on the physical microenvironment maps of multiple tumor samples, local porosity was obtained. With corresponding stiffness parameters The paired data; Fitting empirical relation: ; in, and These are the positive real coefficients calibrated using the least squares method; In virtual perturbation, the porosity of the target spatial cell is changed from... Increase to And calculate the stiffness after disturbance. : ; Attenuation factor Defined as: ; Update the stiffness characteristics of the spatial element to ; In the formula, Indicates the original local porosity. This indicates the porosity set after the disturbance. Represents the original stiffness parameters. This represents the post-disturbance stiffness calculated based on the model. This is the attenuation factor used to scale physical characteristics.
[0040] It should be noted that by modeling spatial proximity and physical connectivity together as graph-structured edge relationships and combining them with graph neural networks for end-to-end training, the model can not only learn the complex interaction patterns of the microenvironment, but also quantitatively assess the causal effects of specific stromal cells on immune function under virtual perturbation. This provides a computable and interventionist digital platform for mechanism analysis and the generation of personalized treatment strategies.
[0041] In summary, this invention integrates physical microenvironment imaging with multi-omics spatial data to construct a digital twin of the tumor microenvironment with causal inference capabilities, achieving a leap from static description to dynamic simulation. Furthermore, the method not only accurately aligns physical, genetic, protein, and metabolic information at the subcellular scale, but also realistically recreates the restricted diffusion behavior of signaling molecules in a dense matrix through a graph neural network edge weight design that integrates porosity and spatial proximity. It also supports virtual perturbation of specific stromal cells to quantitatively assess their inhibitory effect on T cell function, thus providing a high-precision, calculable, and mechanistically clear decision-making basis for the formulation of personalized immunotherapy strategies.
[0042] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A single-cell spatial multi-omics analysis method for tumor microenvironment research, characterized in that: include: Serial sections were prepared from the same tumor tissue sample to obtain tissue sections for physical imaging, spatial transcriptome sequencing, spatial protein detection, and spatial metabolite detection. High-resolution imaging of physical imaging sections and quantification of physical property parameters of the extracellular matrix are performed to generate spatially resolved physical microenvironment maps. Spatial transcriptome sections, spatial protein sections, and spatial metabolite sections were analyzed to obtain corresponding spatially resolved gene expression maps, protein activity maps, and metabolite abundance maps. Physical microenvironment maps, gene expression maps, protein activity maps, and metabolite abundance maps are registered to a unified spatial coordinate system based on tissue morphology and structure, so that data at the same spatial location correspond to each other. In each spatial unit of a unified spatial coordinate system, corresponding physical property parameters, gene expression data, protein activity data, and metabolite abundance data are fused to construct a multidimensional feature vector. Cell type deconvolution is then performed in conjunction with single-cell transcriptome reference data to obtain cell composition information within the spatial unit. Using each spatial unit as a node, multidimensional feature vectors as node features, and spatial proximity and physical connectivity as edge relationships, a graph neural network model is constructed and trained to form a digital twin of the tumor microenvironment, which is used to perform virtual perturbations and quantify the causal regulatory effects of specific cell subpopulations on the immune microenvironment.
2. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 1, characterized in that: The specific steps for performing high-resolution imaging of physical imaging slices and quantifying the physical property parameters of the extracellular matrix to generate a spatially resolved physical microenvironment map are as follows: The physical imaging slices were scanned using a second harmonic generation microscopy system to obtain nonlinear optical signal images of collagen fibers. The signal image is subjected to background correction and contrast enhancement processing to obtain a preprocessed image; In the preprocessed image, multiple local regions are slidably segmented with a fixed window size, and a skeletonization algorithm is performed on each local region to extract the fiber centerline; The number of fiber intersections per unit area is calculated based on the centerline set, and used as an indicator of network complexity. Construct an orientation histogram for the set of centerline orientation angles and calculate the orientation entropy. The expression is: ; in, This means dividing the interval from 0 to π radians into equal parts. A sub-interval of angles, Indicates the first The proportion of all centerline direction angles within each angle sub-interval; Network complexity metrics and orientation entropy Together with the local porosity estimated by morphological opening operations, they form a vector of physical property parameters, which is mapped to the corresponding spatial coordinates to form a physical microenvironment map.
3. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 2, characterized in that: The skeletonization algorithm employs an iterative refinement method, removing boundary pixels that meet preset endpoint retention conditions in each iteration until all fiber structures converge to a single-pixel-width centerline while preserving the original topological connectivity. The endpoint retention conditions include prohibiting deletion if the current pixel is a fiber endpoint or located at a three-way or higher connection.
4. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 3, characterized in that: The specific steps for registering the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map to a unified spatial coordinate system based on tissue morphology are as follows: For each functional slice, an H&E stained slice was prepared at the adjacent position as a morphological reference. Tissue contours, vascular cavities, and densely populated areas of cell nuclei in H&E images were extracted as feature anchors. Low-resolution bright-field images were extracted from the corresponding slices of the physical microenvironment map, gene expression map, protein activity map, and metabolite abundance map. An initial affine transformation model is established using feature anchor points, and then a non-rigid registration algorithm based on B-splines is used to optimize the local deformation field, so as to minimize the spatial coordinate deviation of the same anatomical structure in each modal image. The optimized deformation field is applied to each functional map, and a four-modal aligned dataset in a unified spatial coordinate system is output. The non-rigid registration algorithm achieves multimodal image alignment by minimizing the mutual information loss function, ensuring spatial consistency of data at different molecular levels at the subcellular scale.
5. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 4, characterized in that: The specific steps are as follows: In each spatial unit of a unified spatial coordinate system, corresponding physical property parameters, gene expression data, protein activity data, and metabolite abundance data are fused to construct a multidimensional feature vector. This vector is then combined with single-cell transcriptome reference data for cell type deconvolution to obtain cellular composition information within the spatial unit. Define the spatial unit as a square grid with fixed side lengths in a unified coordinate system; For each spatial unit, the measured values of all modes within it are extracted and standardized, and then concatenated to form a multidimensional feature vector; Load an externally available tumor single-cell transcriptome reference dataset and label cluster centers by cell type; Solve the constrained optimization problem so that the gene expression part of the spatial unit is linearly approximated by a weighted combination of reference cell types as much as possible; The output weight vector serves as the cellular composition information of the spatial unit. The objective function of the constrained optimization problem introduces non-negativity and normalization constraints to ensure that the deconvolution result has biological interpretability.
6. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 5, characterized in that: The specific steps for constructing and training a graph neural network model to form a digital twin of the tumor microenvironment, using each spatial unit as a node, multidimensional feature vectors as node features, and spatial proximity and physical connectivity as edge relationships, are as follows: All spatial units are indexed into a set of nodes. For any two nodes, if the Euclidean distance is less than a preset threshold, an edge is created, and the edge weight is calculated. for: ; in, Represents a node With nodes Euclidean distance in a unified spatial coordinate system and Representing nodes respectively With nodes The corresponding local porosity, This is the balance coefficient; Constructing a graph structure based on node sets and edge weight matrices; A graph attention network architecture is used for multi-layer message passing, and each layer updates the node representation to integrate neighborhood micro-environment information. Using the activation state of immune cells in real samples as a supervisory signal, the model parameters are trained so that the output layer can predict changes in immune function indicators under perturbation conditions.
7. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 6, characterized in that: The specific steps for performing virtual perturbation and quantifying the causal regulatory effects of specific cell subsets on the immune microenvironment are as follows: Identify spatial units where the proportion of COL11A1-positive cancer-associated fibroblasts exceeds a set threshold; In the input layer of the graph neural network, the COL11A1 expression feature value of all nodes in the set is set to zero, and the stiffness parameter in its physical microenvironment map is multiplied by a decay factor, with the decay factor being less than 1. Run the trained graph neural network model to obtain the predicted expression values of IFNG and GZMB of the nodes where CD8 positive T cells are located; Calculate the difference between the IFNG and GZMB predicted values before and after the disturbance; The difference was used as a quantitative indicator of the causal regulatory effect of COL11A1-positive cancer-associated fibroblasts on T cell function, and was used to generate subsequent personalized treatment strategies.
8. The single-cell spatial multi-omics analysis method for tumor microenvironment research as described in claim 7, characterized in that: The attenuation factor is determined through the power-law relationship between porosity and stiffness, and the specific steps are as follows: Based on the physical microenvironment maps of multiple tumor samples, local porosity was obtained. With corresponding stiffness parameters The paired data; Fitting empirical relation: ； in, and These are the positive real coefficients calibrated using the least squares method; In virtual perturbation, the porosity of the target spatial cell is changed from... Increase to And calculate the stiffness after disturbance. : ； Attenuation factor Defined as: ； Update the stiffness characteristics of the spatial element to ; In the formula, Indicates the original local porosity. This indicates the porosity set after the disturbance. Represents the original stiffness parameters. This represents the post-disturbance stiffness calculated based on the model. This is the attenuation factor used to scale physical characteristics.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the single-cell spatial multi-omics analysis method for tumor microenvironment research as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the single-cell spatial multi-omics analysis method for tumor microenvironment research as described in any one of claims 1 to 8.