Spatial transcriptomic clustering method, apparatus, device, and storage medium

Abstract

This invention relates to a spatial transcriptomics clustering method, apparatus, device, and storage medium. The method includes: acquiring raw gene expression data, spatial location data, and adjacency relationship data of the target clustering object; performing graph convolutional encoding on the raw gene expression data and adjacency relationship data to obtain expression feature data; performing spatial location encoding on the spatial location data to obtain spatial feature data; reconstructing and fusing the expression feature data and spatial feature data to obtain target gene expression data; and performing spatial clustering analysis on the target clustering object based on the target gene expression data. The technical solution of this invention can achieve joint modeling and collaborative encoding of biomolecular features and physical spatial topology, improving the accuracy, robustness, and spatial consistency of clustering results.

Description

Technical Field

[0001] This invention relates to the field of spatial transcriptomics technology, and in particular to a spatial transcriptomics clustering method, apparatus, device, and storage medium. Background Technology

[0002] Spatial transcriptomics technology overcomes the limitations of traditional single-cell transcriptomics sequencing, which loses information about the spatial distribution of cells. It can simultaneously capture the gene expression profiles and precise spatial coordinates of cells in biological tissues, providing core data support for analyzing tissue microenvironment heterogeneity, cell-cell interactions, and disease pathological mechanisms. In fields such as tumor subtyping, developmental biology, and neuroscience, cluster analysis of spatial transcriptomics data is a crucial preliminary step. Its core objective is to classify cells with similar gene expression characteristics and spatial distribution patterns, thereby identifying functional homologous units or abnormal lesion regions. With the popularization of spatial transcriptomics technology, how to efficiently and accurately achieve data clustering has become one of the core technical bottlenecks restricting its practical application.

[0003] However, existing spatial transcriptomics clustering methods largely rely on graph neural networks, which are essentially local feature aggregations. They only indirectly introduce spatial location information through adjacency graphs, making it difficult to model global spatial patterns. They fail to directly encode and extract features from spatial location data at a deep level, resulting in the utilization of spatial information remaining at the "adjacent association" level. This leads to insufficient depth in spatial information mining and ignores the strong correlation between gene expression changes and the continuous distribution of spatial coordinates in biological cells. Furthermore, the fusion of gene expression and spatial location data only uses spatial information for mapping, failing to consider the heterogeneity of the two types of data. Gene expression is high-dimensional discrete data, while spatial location is low-dimensional continuous data, leading to redundancy or information loss in the fused features, thus affecting the accuracy of clustering. Summary of the Invention

[0004] This invention provides a spatial transcriptomics clustering method, apparatus, device, and storage medium, aiming to achieve joint modeling and collaborative encoding of biomolecular features and physical spatial topology, maximizing the preservation of spatially discriminative biological information, and improving the accuracy, robustness, and spatial consistency of clustering results.

[0005] In a first aspect, embodiments of the present invention provide a spatial transcriptomics clustering method, comprising: Obtain the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object; Based on the graph convolutional coding model, the original gene expression data and the adjacency relationship data are processed by graph convolutional coding to obtain the expression feature data of the target cluster object; Based on the spatial location coding model, the spatial location data is processed by spatial location coding to obtain the spatial feature data of the target clustering object; Based on the reconstruction and fusion model, the expression feature data and the spatial feature data are reconstructed and fused to obtain the target gene expression data. Spatial clustering analysis is performed on the target cluster objects based on the target gene expression data.

[0006] Optionally, the step of reconstructing and fusing the expression feature data and the spatial feature data based on the reconstruction fusion model to obtain target gene expression data includes: Based on the reconstruction fusion model, the expression feature data and the spatial feature data are decoded and reconstructed respectively to obtain the first reconstructed expression data and the second reconstructed expression data. The first reconstructed expression data and the second reconstructed expression data are subjected to expression fusion processing to obtain the target gene expression data.

[0007] Optionally, before performing spatial clustering analysis on the target clustering objects based on the target gene expression data, the method further includes: The graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model are jointly trained to optimize and update the target gene expression data.

[0008] Optionally, the joint training of parameters for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model; The graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model are jointly trained according to the joint loss function. The joint loss function consists of a feature reconstruction loss function, a location reconstruction loss function, a representation difference loss function, and a representation consistency loss function.

[0009] Optionally, the construction of the joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the feature reconstruction loss function between the first reconstructed expression data and the original gene expression data; Construct the positional reconstruction loss function between the second reconstructed expression data and the original gene expression data.

[0010] Optionally, after obtaining the original gene expression data, spatial location data, and adjacency relationship data of the target clustering object, the method further includes: Construct the perturbed gene expression data of the target cluster objects; Based on the graph convolutional coding model, the perturbation gene expression data and the adjacency relationship data are processed by the graph convolutional coding to obtain the perturbation feature data of the target cluster object; The joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the representation difference loss function between the perturbation feature data and the representation feature data.

[0011] Optionally, after performing graph convolutional coding on the perturbed gene expression data and the adjacency relationship data based on the graph convolutional coding model to obtain the perturbed feature data of the target cluster object, and performing graph convolutional coding on the original gene expression data and adjacency relationship data based on the graph convolutional coding model to obtain the expression feature data of the target cluster object, the method further includes: Spatial neighborhood aggregation processing is performed on the perturbation feature data and the expression feature data respectively to obtain the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation; The joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the representation consistency loss function between the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation.

[0012] Secondly, embodiments of the present invention provide a spatial transcriptomics clustering device, comprising: The data acquisition module is used to acquire the original gene expression data, spatial location data, and adjacency relationship data of the target clustering object; The expression feature module is used to perform graph convolutional coding on the original gene expression data and the adjacency relationship data based on a graph convolutional coding model to obtain the expression feature data of the target cluster object; The spatial feature module is used to perform spatial location encoding processing on the spatial location data based on the spatial location encoding model to obtain the spatial feature data of the target clustering object; The reconstruction and fusion module is used to reconstruct and fuse the expression feature data and the spatial feature data based on the reconstruction and fusion model to obtain the target gene expression data. The clustering analysis module is used to perform spatial clustering analysis on the target cluster objects based on the target gene expression data.

[0013] Thirdly, embodiments of the present invention provide a spatial transcriptomics clustering device, comprising: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the spatial transcriptomics clustering method provided in any embodiment of the present invention.

[0014] Fourthly, embodiments of the present invention provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a spatial transcriptomics clustering method as provided in any embodiment of the present invention.

[0015] This invention provides a spatial transcriptomics clustering method, apparatus, device, and storage medium. It acquires the original gene expression data, spatial location data, and adjacency relationship data of the target clustering object. Based on a graph convolutional coding model, it performs graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the expression feature data of the target clustering object. Based on a spatial location coding model, it performs spatial location coding on the spatial location data to obtain the spatial feature data of the target clustering object. Finally, based on a reconstruction and fusion model, it reconstructs and fuses the expression feature data and spatial feature data to obtain the target gene expression data. Thus, it performs spatial clustering analysis on the target clustering object based on the target gene expression data. This solves the problems of insufficient spatial information mining and unreasonable fusion methods of gene expression and spatial location features in existing clustering processes. It achieves joint modeling and collaborative coding of biomolecular features and physical spatial topology, maximizing the preservation of spatially discriminative biological information and improving the accuracy, robustness, and spatial consistency of the clustering results. Attached Figure Description

[0016] Figure 1 This is a flowchart of a spatial transcriptomics clustering method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of a spatial transcriptomics clustering method provided in Embodiment 2 of the present invention; Figure 3 This is a flowchart of a spatial transcriptomics clustering method provided in Embodiment 2 of the present invention; Figure 4 This is a schematic diagram of the structure of a spatial transcriptomics clustering device provided in Embodiment 3 of the present invention; Figure 5 This is a schematic diagram of the structure of a spatial transcriptomics clustering device provided in Embodiment 4 of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0018] Example 1 Figure 1 This is a flowchart of a spatial transcriptomics clustering method provided in Embodiment 1 of the present invention. This embodiment is applicable to the spatial transcriptomics clustering of sequencing units in biological tissues. The method can be executed by a spatial transcriptomics clustering device, which can be implemented by hardware and / or software and is generally integrated into an electronic device, such as a computer device. The method specifically includes: Step 110: Obtain the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object.

[0019] The target clustering objects can be sequencing units that make up biological tissue slices, including single cells and / or spots, serving as objects for spatial transcriptomics clustering. Raw gene expression data can be data describing the expression levels of all or some specific genes in each target clustering object. Spatial location data can be data describing the spatial location of each target clustering object. Adjacency relationship data can be data describing the adjacency relationships of each target clustering object in its spatial location.

[0020] Specifically, the original gene expression data and spatial location data can be determined based on the spatial transcriptome data of the target cluster objects. Furthermore, based on the locations of each target cluster object described in the spatial location data, target cluster objects with spatial adjacency relationships can be extracted and recorded, constructing adjacency relationship data.

[0021] Step 120: Based on the graph convolutional coding model, perform graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the expression feature data of the target cluster object.

[0022] The graph convolutional coding model can be a pre-built and trained model for processing input data using graph convolutional coding. Graph convolutional coding processing can include operations that extract adjacent association features of gene expression from the original input gene expression data and adjacency relationship data, and map them to a specific dimension. The expression feature data can be a specific-dimensional representation of the adjacent association features of gene expression obtained through graph convolutional coding processing.

[0023] Specifically, by performing graph convolutional encoding on the raw gene expression data and adjacency relationship data input to the graph convolutional encoding model, the gene expression features and spatial neighborhood relationship features of the target cluster object can be extracted and fused to generate expression feature data of a specific dimension. The dimension of the expression feature data can be pre-set according to the needs of subsequent steps, preferably set to the same dimension as the spatial location features, so as to facilitate subsequent collaborative processing and fusion with the spatial location feature representation. Based on the graph convolutional encoding operation, a graph convolutional encoding model with corresponding functions can be pre-constructed and trained.

[0024] Step 130: Based on the spatial location coding model, perform spatial location coding processing on the spatial location data to obtain the spatial feature data of the target clustering object.

[0025] The spatial location encoding model is a pre-built and trained model used to encode the spatial location of the input data. Spatial location encoding can include extracting spatial location features from the input spatial location data and mapping them to a specific dimension. The spatial feature data can be a specific-dimensional representation of the spatial location features obtained through spatial location encoding.

[0026] Specifically, by performing spatial location encoding on the spatial location data input to the spatial location encoding model, spatial location features of the target clustering objects can be extracted, and spatial feature data of a specific dimension can be generated. The dimension of the spatial feature data can be pre-set according to the needs of subsequent steps; preferably, it can be consistent with the dimension of the feature data to facilitate collaborative processing and fusion between the two. Based on the spatial location encoding operation, a spatial location encoding model with corresponding functions can be pre-constructed and trained.

[0027] Step 140: Based on the reconstruction and fusion model, the expression feature data and spatial feature data are reconstructed and fused to obtain the target gene expression data.

[0028] The reconstruction and fusion model can be a pre-built and trained model used to reconstruct and fuse input data. The reconstruction and fusion process can include operations such as graph convolutional encoding and spatial location encoding to decode the expression feature data and spatial feature data, obtaining the corresponding gene expression data, and then fusing the gene expression data into target gene expression data. The target gene expression data can be gene expressions that simultaneously reflect the adjacency association features and spatial location features of the gene expressions of the target cluster object.

[0029] Specifically, through reconstruction and fusion processing, expression feature data and spatial feature data can be reconstructed into the original gene expression form, and finally fused into target gene expression data for expression. The decoding process of expression feature data and spatial feature data can include decoding operations corresponding to graph convolutional coding and spatial location coding, respectively.

[0030] In an optional implementation, based on a reconstruction fusion model, expression feature data and spatial feature data are reconstructed and fused to obtain target gene expression data. This may include: based on the reconstruction fusion model, decoding and reconstructing expression feature data and spatial feature data respectively to obtain first reconstructed expression data and second reconstructed expression data; and performing expression fusion processing on the first reconstructed expression data and the second reconstructed expression data to obtain target gene expression data.

[0031] The decoding and reconstruction process can involve decoding the expression feature data and spatial feature data according to graph convolutional coding and spatial location coding operations, respectively, to obtain the corresponding gene expression data. The first reconstructed expression data and the second reconstructed expression data can be the gene expression data corresponding to the expression feature data and spatial feature data, respectively. The expression fusion process can be the operation of fusing the first reconstructed expression data and the second reconstructed expression data into the target gene expression data.

[0032] The above implementation design features independent encoding modules adapted to both types of data while ensuring the fusionability of encoded features. Ultimately, it can naturally integrate the two heterogeneous information representation methods of discrete gene expression and continuous space, and enable them to complement and correct each other during the learning process. This ensures that the final learned representation retains spatially discriminative biological variation information to the greatest extent, achieving more accurate, robust, and spatially consistent clustering results.

[0033] Step 150: Perform spatial clustering analysis on the target cluster objects based on the target gene expression data.

[0034] Spatial clustering analysis can be an operation for spatial transcriptomics clustering.

[0035] Specifically, the final target gene expression data has the original expression form and can reflect the adjacency association characteristics and spatial location characteristics of the gene expression of the target cluster objects. Therefore, spatial clustering analysis of the target cluster objects based on the target gene expression data can achieve clustering analysis that takes into account both the adjacency association characteristics and spatial location characteristics of gene expression.

[0036] The technical solution of this embodiment obtains the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object. Based on a graph convolutional coding model, it performs graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the expression feature data of the target cluster object. Based on a spatial location coding model, it performs spatial location coding on the spatial location data to obtain the spatial feature data of the target cluster object. Based on a reconstruction and fusion model, it performs reconstruction and fusion on the expression feature data and spatial feature data to obtain the target gene expression data. Thus, spatial clustering analysis is performed on the target cluster object based on the target gene expression data. This solves the problems of insufficient spatial information mining and unreasonable fusion methods of gene expression and spatial location features in existing clustering processes. It realizes joint modeling and collaborative coding of biomolecular features and physical spatial topology, preserves spatially discriminative biological information to the greatest extent, and improves the accuracy, robustness, and spatial consistency of clustering results.

[0037] Example 2 Figure 2 This is a flowchart of a spatial transcriptomics clustering method provided in Embodiment 2 of the present invention. This embodiment further refines the above technical solution by including, before performing spatial clustering analysis on the target clustering object based on the target gene expression data: jointly training the graph convolutional coding model, the spatial location coding model, and the reconstructed fusion model; and updating the target gene expression data based on the jointly trained graph convolutional coding model, spatial location coding model, and reconstructed fusion model. Specifically, this method includes: Step 210: Obtain the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object.

[0038] In one alternative implementation, the original gene expression data can be a gene expression matrix, the spatial location data can be a spatial coordinate matrix, and the adjacency relationship data can be a graph structure obtained based on KNN (K-Nearest Neighbors algorithm).

[0039] In an optional implementation, obtaining the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object may include: obtaining a gene expression matrix and a spatial coordinate matrix; performing highly variable gene screening and standardization on the gene expression matrix to obtain a processed gene expression matrix; constructing a spatial adjacency matrix based on the spatial coordinate matrix and adding self-loop edges to form adjacency relationship data.

[0040] Specifically, spatial transcriptome data can include gene expression matrices. and spatial coordinate matrix , where N represents the number of cells, M represents the number of genes, and the spatial coordinate matrix S has 2 columns, which are used to record the x and y coordinates in the spatial coordinates.

[0041] Further, optionally, the gene expression matrix can be subjected to highly variable gene screening and standardization to obtain a processed gene expression matrix. This can include: using the Scanpy library in Python to screen the gene expression matrix for highly variable genes, retaining the head. A subset of genes, such as the Top 3000 genes, is selected and normalized to obtain the processed gene expression matrix. .

[0042] Further, optionally, constructing a spatial adjacency matrix based on the spatial coordinate matrix and adding self-loop edges can include: constructing a k-nearest neighbor graph based on the spatial coordinate matrix S to obtain a symmetric spatial adjacency matrix. Where 0 represents that there is no connection between two cells, and 1 represents that there is a connection between two cells; add self-loop edges I to the spatial adjacency matrix A to form the adjacency relationship data.

[0043] In an optional implementation, after obtaining the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object, the method may further include: constructing perturbed gene expression data of the target cluster object; and performing graph convolutional coding processing on the perturbed gene expression data and adjacency relationship data based on a graph convolutional coding model to obtain perturbed feature data of the target cluster object.

[0044] The perturbation gene expression data can be data obtained by randomly perturbing the cellular representation order in the original gene expression data. The perturbation feature data can be a specific-dimensional representation of the adjacent association features of gene expression described by the perturbation gene expression data obtained through graph convolutional coding.

[0045] Specifically, by randomly perturbing the cell representation order in the original gene expression data, perturbed gene expression data can be obtained. Then, based on the same graph convolutional coding model, graph convolutional coding processing can be performed on the perturbed gene expression data and the adjacency relationship data to obtain the perturbed feature data of the target cluster object. The rationality, accuracy and robustness of graph convolutional coding processing based on the graph convolutional coding model can be obtained by examining the relationship between the perturbed feature data and the expression feature data.

[0046] Specifically, optional, for the gene expression matrix after processing the original gene expression data. Constructing perturbed gene expression data for the target clustering objects can include: randomly shuffling the representation order of cells in the gene expression matrix X to generate perturbed feature data as perturbation enhancement features. subscript The data is identified as a perturbation enhancement feature generated by performing a specific perturbation operation on the original expression matrix X, i.e., randomly shuffling the cell order.

[0047] Further optionally, the graph convolutional coding model can be a pre-defined graph convolutional neural network; correspondingly, based on the graph convolutional coding model, graph convolutional coding processing is performed on the perturbation gene expression data and adjacency relationship data to obtain the perturbation feature data of the target clustering object, which may include: enhancing the perturbation features. The spatial adjacency matrix A is input into the graph convolutional neural network to obtain the perturbed graph cell latent representation output by the graph convolutional neural network. As perturbation feature data, the perturbation map cell latent representation The subscript 'a' in the text indicates perturbation-based enhancement features. It was calculated, rather than based on the original gene expression matrix. Calculated.

[0048] Specifically, graph convolutional coding models can be used to map high-dimensional genetic features into latent representations of a specific dimension D, such as 64 dimensions. For the input... and Using GCN (Graph Convolutional Network) for encoding, high-dimensional gene features are mapped to low-dimensional representations, resulting in perturbation graph cell latent representations. .

[0049] Step 220: Based on the graph convolutional coding model, perform graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the expression feature data of the target cluster object.

[0050] In one optional implementation, the graph convolutional coding model can be a pre-defined graph convolutional neural network. Accordingly, based on the graph convolutional coding model, graph convolutional coding processing is performed on the original gene expression data and adjacency relation data to obtain the expression feature data of the target cluster object. This can include: inputting the gene expression matrix X and the spatial adjacency matrix A into the graph convolutional neural network to obtain the original graph cell latent representation Z output by the graph convolutional neural network as the expression feature data.

[0051] Specifically, graph convolutional coding models can be used to map high-dimensional genetic features into latent representations of a specific dimension D, such as 64 dimensions. For the input... and GCN is used for encoding to map high-dimensional gene features to the original cellular latent representation. , where N represents the number of cells.

[0052] In an optional implementation, after performing graph convolutional coding on the perturbed gene expression data and adjacency relationship data based on the graph convolutional coding model to obtain the perturbed feature data of the target cluster object, and performing graph convolutional coding on the original gene expression data and adjacency relationship data based on the graph convolutional coding model to obtain the expression feature data of the target cluster object, the implementation may further include: performing spatial neighborhood aggregation processing on the perturbed feature data and expression feature data to obtain the perturbed graph neighborhood aggregation representation and the original graph neighborhood aggregation representation.

[0053] Among them, spatial neighborhood aggregation processing is to aggregate the expression features of cells in its neighborhood based on spatial adjacency for each cell, so that each cell obtains expression features that fuse neighborhood information, thereby obtaining perturbation graph neighborhood aggregation representation and original graph neighborhood aggregation representation corresponding to perturbation feature data and expression feature data, respectively.

[0054] Specifically, through spatial neighborhood aggregation processing, the perturbation feature data and expression feature data extracted from each target cluster object, i.e., cells, can be transformed into a neighborhood aggregation representation of each target cluster object after fusing its spatial neighborhood information, realizing feature extraction and representation at different levels: cell unit level and graph structure neighborhood level. Specifically, optionally, the expression feature data is the original graph cell latent representation Z, and the perturbation feature data is the perturbation graph cell latent representation. Spatial neighborhood aggregation is performed on the perturbation feature data and the expression feature data respectively to obtain the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation, which may include: the latent representation of the perturbation graph cells. An average readout layer and a sigmoid activation function are used for graph structure awareness, and neighborhood information is aggregated to obtain a perturbed graph neighborhood aggregation representation that incorporates the neighborhood information. The original image cell latent representation Z is processed using an average readout layer and a Sigmoid activation function for graph structure perception, aggregating neighborhood information to obtain the original image neighborhood aggregated representation with fused neighborhood information. .

[0055] Step 230: Based on the spatial location coding model, perform spatial location coding processing on the spatial location data to obtain the spatial feature data of the target clustering object.

[0056] In an optional implementation, for spatial location data as a spatial coordinate matrix S, the spatial location encoding model can be a pre-defined sinusoidal representation network. Correspondingly, based on the spatial location encoding model, performing spatial location encoding processing on the spatial location data to obtain the spatial feature data of the target clustering object can include: inputting the spatial coordinate matrix S into the sinusoidal representation network, performing multi-layer nonlinear mapping, and obtaining a potential spatial coordinate representation aligned with the gene feature dimension. As spatial feature data.

[0057] Specifically, SIREN (Sinusoidal Representation Networks) can be configured to contain at least four weight matrix layers, all of which have sine activation functions, and their mapping process satisfies the following formula: in, Represents the sine function. , , and Let be the weight matrix, where, , and Let S represent the hidden layer dimension of each mapping, and S be the spatial coordinate matrix. This is the latent representation of the spatial coordinates for the output. Given the input spatial coordinate matrix... Based on the above formula, the intermediate latent representation of spatial coordinates is obtained, thus yielding the latent representation of spatial coordinates. It is a hidden representation aligned with gene features, which can enhance spatial structure modeling.

[0058] The above implementation applies SIREN to spatial coordinate encoding, injecting spatial location information into the model in a learnable and non-linear manner, going beyond simple Euclidean distance or adjacency relationships. Due to the characteristics of its periodic activation function, SIREN can more efficiently learn high-frequency details and complex geometric structures in continuous space, thereby generating coordinate latent representations that are better semantically aligned with high-dimensional genetic features, significantly enhancing the model's ability to perceive spatial continuity and heterogeneity.

[0059] Step 240: Based on the reconstruction and fusion model, the expression feature data and spatial feature data are reconstructed and fused to obtain the target gene expression data.

[0060] In an optional implementation, based on the reconstruction fusion model, the expression feature data and spatial feature data are decoded and reconstructed separately to obtain first reconstructed expression data and second reconstructed expression data. This may include: using a graph convolutional neural network to decode the latent cell representation Z of the original image to obtain the first reconstructed expression data as a first gene expression matrix. ; Utilizing a feedforward network containing a sine function layer for latent representation of spatial coordinates Decoding yields the second reconstructed expression data, which is the second gene expression matrix. Accordingly, expression fusion processing is performed on the first and second reconstructed expression data to obtain target gene expression data. This may include fusing the first gene expression matrix and the second gene expression matrix to obtain the target gene expression data as the final reconstructed gene expression matrix. .

[0061] Optionally, the first gene expression matrix and the second gene expression matrix can be fused to obtain the target gene expression data as the final reconstructed gene expression matrix. This can include: summing and averaging the first gene expression matrix and the second gene expression matrix to obtain the final reconstructed gene expression matrix. .

[0062] Specifically, by using a graph convolutional neural network to decode the latent cell representation Z in the original image, the first gene expression matrix can be obtained. , ; Utilizing a single-layer SIREN for latent representation of spatial coordinates Decoding is performed to reconstruct the original gene expression space, resulting in the second reconstructed expression data, which is the second gene expression matrix. , Subsequently, regarding and Summing and averaging yields the final reconstructed gene expression matrix. , .in, Indicates GCN network, Let A represent the sine function, and A be the adjacency matrix. and Let represent the weight matrix, where This represents the hidden layer dimension of the mapping.

[0063] Step 250: Jointly train the parameters of the graph convolutional coding model, spatial location coding model, and reconstruction fusion model to optimize and update the target gene expression data.

[0064] Joint parameter training can be an operation that optimizes the parameters in the graph convolutional coding model, spatial location coding model, and reconstruction fusion model to make the output results of each model more accurate. Optimization update processing can be an operation that updates the target gene expression data with the corresponding results obtained from the jointly trained graph convolutional coding model, spatial location coding model, and reconstruction fusion model.

[0065] Specifically, before performing spatial clustering analysis based on the target gene expression data, the logical relationships between the outputs of each model obtained in the aforementioned steps can be analyzed to determine whether they are reasonable and conform to the definition. Based on the deviations between the results, the parameters of each model can be jointly trained to obtain more accurate target gene expression data after optimization and updating, which serves as the basis for spatial clustering analysis.

[0066] In one optional implementation, joint parameter training of the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model may include: constructing a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model; and performing joint parameter training of the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model based on the joint loss function.

[0067] The joint loss function consists of the feature reconstruction loss function, the location reconstruction loss function, the representation difference loss function, and the representation consistency loss function.

[0068] Specifically, the feature reconstruction loss function can be a function describing the bias caused by the graph convolutional coding model encoding the original gene expression data and adjacency relationship data into expression feature data, and the reconstruction fusion model reconstructing the expression feature data into the corresponding gene expression data. Based on the feature reconstruction loss function, the model can be trained to accurately reconstruct the input features.

[0069] The location reconstruction loss function can be a function that describes the bias caused by the spatial location encoding model encoding spatial location data into spatial feature data, and the reconstruction fusion model reconstructing spatial feature data into corresponding gene expression data. Based on the location reconstruction loss function, the quality of the model's internal representation can be trained.

[0070] The representational difference loss function can be considered a function describing the bias in the recognition of gene expression features of each target cluster object during the process of a graph convolutional coding model encoding raw gene expression data and adjacency relation data into expression feature data. Based on the representational difference loss function, the model can be trained to distinguish different sample representations.

[0071] The representation consistency loss function can be considered a function describing the identification bias of a graph convolutional coding model in encoding raw gene expression data and adjacency relationship data into expression feature data for all target cluster objects. Based on the representation consistency loss function, the robustness of the model to graph structure perturbations can be trained.

[0072] In one optional implementation, constructing a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model may include: constructing a feature reconstruction loss function between the first reconstructed expression data and the original gene expression data; and constructing a location reconstruction loss function between the second reconstructed expression data and the original gene expression data.

[0073] Specifically, optionally, constructing a feature reconstruction loss function between the first reconstructed expression data and the original gene expression data may include: calculating the original gene expression matrix X and the first gene expression matrix X. The mean squared error between them is used as the feature reconstruction loss function. .

[0074] Further optional, feature reconstruction loss function The following formula can be used for calculation: in, This represents the original gene expression in the i-th cell. This represents the reconstructed gene expression in the i-th cell. It compares the original gene expression matrix X with the reconstructed first gene expression matrix. Calculating MSE (Mean Squared Error) measures the mean squared error between the original input features and the model's output features, ensuring that the model can accurately reconstruct the input features.

[0075] Specifically, optionally, constructing a positional reconstruction loss function between the second reconstructed expression data and the original gene expression data may include: calculating the original gene expression matrix X and the second gene expression matrix X. The Frobenius norm difference between them serves as the location reconstruction loss function. .

[0076] Further optional, location reconstruction loss function The following formula can be used for calculation: in, This represents the original gene expression in the i-th cell. The coordinates of the i-th cell represent the reconstructed gene expression. The coordinates are reconstructed using L2 norm constraints. The difference from the original feature X ensures the quality of the representation within the model.

[0077] The above implementation, through the dual constraints of feature reconstruction loss and coordinate reconstruction loss, requires the model not only to recover the original information from the gene latent representation but also to decode the gene features from the coordinate encoding. This cross-reconstruction task, as a powerful self-supervised signal, ensures that the learned latent space contains rich and complementary biological information, avoiding information loss.

[0078] In an optional implementation, constructing a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model may include: constructing a representation difference loss function between the perturbation feature data and the representation feature data.

[0079] Specifically, optionally, constructing a representation difference loss function between perturbation feature data and expression feature data may include: calculating the latent representation of the perturbation map cells. The negative value of the cosine similarity between the latent cell representation Z and the original image is used as the representation of the difference loss function. .

[0080] Further optional, representing the difference loss function The following formula can be used for calculation: Where N represents the total number of cells, and Let Z be the original image and the embedding vector of the interference image, respectively, for the i-th cell. The cosine similarity is used to measure the latent representation Z of the original image and the latent representation Z of the interference image. The similarity helps distinguish between different sample representations.

[0081] In one optional implementation, constructing a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model may include: constructing a representation consistency loss function between the perturbed graph neighborhood aggregation representation and the original graph neighborhood aggregation representation.

[0082] Specifically, optionally, constructing a representation consistency loss function between the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation may include: calculating the perturbation graph neighborhood aggregation representation. The negative value of the cosine similarity between the neighborhood aggregation of the original image and G is used as the consistency loss function. .

[0083] Further optional, representing the consistency loss function The following formula can be used for calculation: Where N represents the total number of cells, and Let G be the original image and the neighboring image of the perturbed image, respectively, representing the fused embedding vectors of the i-th cell. Cosine similarity is used to measure the difference between the original image-level neighborhood representation G and the perturbed image-level neighborhood representation G. Consistency improves the model's robustness to graph structure perturbations.

[0084] The above implementation method, by generating random perturbation graphs and designing representation difference loss and representation consistency loss, forces the model to learn core features that are insensitive to input noise but robust to graph structure information. This effectively avoids overfitting and improves the model's generalization ability on complex and noisy real data.

[0085] Correspondingly, the joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model may include: constructing the joint loss function as follows: ;in, For feature reconstruction loss function, For the location reconstruction loss function, To represent the difference loss function, To represent the consistency loss function, , and These are the weight hyperparameters. The joint loss function combines the various loss functions, balancing the importance of each loss component through a weighted summation.

[0086] Step 260: Perform spatial clustering analysis on the target cluster objects based on the target gene expression data.

[0087] In one optional implementation, spatial clustering analysis of the target cluster objects based on the target gene expression data may include: using the mclust algorithm for spatial clustering analysis when the number of clusters is known; and using the Louvain algorithm for spatial clustering analysis when the number of clusters is unknown.

[0088] The above implementation achieves the final clustering feature as an average fusion of graph convolutional decoding features and coordinate decoding features, which essentially integrates gene expression information and deep spatial structure information. The resulting feature matrix has lower noise and stronger discriminative power, thus providing a higher quality input for clustering, directly leading to higher clustering purity, clearer domain boundaries, and more accurate cell type identification.

[0089] For example, Figure 3 This is a flowchart illustrating a spatial transcriptomics clustering method provided in Embodiment 2 of the present invention. Figure 3 As shown, spatial transcriptomics clustering methods can specifically include data preprocessing steps, graph convolutional coding steps, spatial coordinate coding steps, decoding and reconstruction steps, and spatial clustering analysis steps.

[0090] The data preprocessing steps specifically include: inputting spatial transcriptome data, including a gene expression matrix X and a spatial coordinate matrix S; performing highly variable gene screening and standardization on the gene expression matrix X to obtain a processed gene expression matrix; constructing a spatial adjacency matrix A based on the spatial coordinate matrix S and adding self-loop edges; and generating perturbation enhancement features by randomly shuffling the cell representation order in the processed gene expression matrix. .

[0091] Furthermore, the graph convolutional encoding step specifically includes: inputting the processed gene expression matrix X and spatial adjacency matrix A into the first graph convolutional neural network to obtain the original graph cell latent representation Z; and adding perturbation enhancement features. The spatial adjacency matrix A is input into the second graph convolutional neural network to obtain the latent representation of the perturbed graph cells. Spatial neighborhood aggregation is performed on the latent cell representation Z of the original image to obtain the aggregated neighborhood representation G of the original image; the latent cell representation of the perturbed image is... Spatial neighborhood aggregation is performed to obtain the perturbation graph neighborhood aggregation representation. .

[0092] Furthermore, the spatial coordinate encoding step specifically includes: inputting the spatial coordinate matrix S into a sinusoidal representation network, performing multi-layer nonlinear mapping, and obtaining a spatial coordinate latent representation aligned with the gene feature dimension. .

[0093] Furthermore, the decoding and reconstruction steps specifically include: performing a first decoding operation on the original cell latent representation Z to reconstruct the first gene expression matrix. ; Potential representation of spatial coordinates Perform a second decoding operation to reconstruct the second gene expression matrix. ; Fusion of the first gene expression matrix With the second gene expression matrix The final reconstructed gene expression matrix is ​​obtained. The first decoding operation involves using a graph convolutional network to decode the latent representation Z of the cells in the original image; the second decoding operation involves using a feedforward network containing a sine function layer to decode the latent representation of spatial coordinates. Decode; fuse into the first gene expression matrix With the second gene expression matrix Sum and then take the average.

[0094] Furthermore, the multi-objective joint optimization step specifically includes: constructing a multi-objective loss function that includes feature reconstruction loss, coordinate reconstruction loss, representation difference loss, and representation consistency loss; and jointly training the first graph convolutional neural network, the second graph convolutional neural network, the sine representation network, and the parameters involved in the decoding and reconstruction step based on the multi-objective loss function.

[0095] Furthermore, the spatial clustering analysis steps specifically include: based on the final reconstructed gene expression matrix... Spatial clustering analysis was performed on the cells.

[0096] The technical solution of this embodiment obtains the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object. Based on a graph convolutional coding model, it performs graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the expression feature data of the target cluster object. Based on a spatial location coding model, it performs spatial location coding on the spatial location data to obtain the spatial feature data of the target cluster object. Finally, based on a reconstruction and fusion model, it performs reconstruction and fusion on the expression feature data and spatial feature data to obtain the target gene expression data. Therefore, spatial clustering analysis is performed on the target cluster object based on the target gene expression data. This solves the problems of insufficient spatial information mining and unreasonable fusion methods of gene expression and spatial location features in existing clustering processes. It achieves joint modeling and collaborative coding of biomolecular features and physical spatial topology, maximizing the preservation of spatially discriminative biological information and improving the accuracy, robustness, and spatial consistency of clustering results. Furthermore, it provides a joint optimization strategy for the encoding and decoding models, balancing feature fidelity, structural consistency, and anti-interference, making the model training process more stable, relatively less sensitive to hyperparameter changes, and producing more reproducible results.

[0097] Example 3 Figure 4 This is a schematic diagram of the structure of a spatial transcriptomics clustering device provided in Embodiment 3 of the present invention, as shown below. Figure 4 As shown, this spatial transcriptomics clustering device includes: a data acquisition module 410, an expression feature module 420, a spatial feature module 430, a reconstruction and fusion module 440, and a clustering analysis module 450, wherein... Data acquisition module 410 is used to acquire the original gene expression data, spatial location data and adjacency relationship data of the target clustering object; The expression feature module 420 is used to perform graph convolutional coding on the original gene expression data and the adjacency relationship data based on a graph convolutional coding model to obtain the expression feature data of the target cluster object; The spatial feature module 430 is used to perform spatial location encoding processing on the spatial location data based on the spatial location encoding model to obtain the spatial feature data of the target clustering object; The reconstruction and fusion module 440 is used to perform reconstruction and fusion processing on the expression feature data and the spatial feature data based on the reconstruction and fusion model to obtain target gene expression data. Clustering analysis module 450 is used to perform spatial clustering analysis on the target clustering object based on the target gene expression data.

[0098] The technical solution of this embodiment obtains the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object. Based on a graph convolutional coding model, it performs graph convolutional coding on the original gene expression data and adjacency relationship data to obtain the adjacency feature data of the target cluster object. Based on a spatial location coding model, it performs spatial location coding on the spatial location data to obtain the spatial feature data of the target cluster object. Based on a reconstruction and fusion model, it performs reconstruction and fusion on the adjacency feature data and spatial feature data to obtain the target gene expression data. Thus, spatial clustering analysis is performed on the target cluster object based on the target gene expression data. This solves the problems of insufficient spatial information mining and unreasonable fusion methods of gene expression and spatial location features in existing clustering processes. It realizes joint modeling and collaborative coding of biomolecular features and physical spatial topology, preserves spatially discriminative biological information to the greatest extent, and improves the accuracy, robustness, and spatial consistency of clustering results.

[0099] Optionally, the reconstruction fusion module 440 may include: a decoding and reconstruction unit, used to perform decoding and reconstruction processing on the expression feature data and the spatial feature data respectively based on the reconstruction fusion model to obtain first reconstructed expression data and second reconstructed expression data; and an expression fusion unit, used to perform expression fusion processing on the first reconstructed expression data and the second reconstructed expression data to obtain the target gene expression data.

[0100] Optionally, the spatial transcriptomics clustering device may further include: a joint training module for jointly training the parameters of the graph convolutional coding model, the spatial location coding model, and the reconstructed fusion model to optimize and update the target gene expression data.

[0101] Optionally, the joint training module may include: a function construction unit for constructing a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model; and a model training unit for performing joint parameter training on the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model based on the joint loss function; wherein the joint loss function consists of a feature reconstruction loss function, a location reconstruction loss function, a representation difference loss function, and a representation consistency loss function.

[0102] Optionally, the function construction unit can be used to: construct the feature reconstruction loss function between the first reconstructed expression data and the original gene expression data; and construct the position reconstruction loss function between the second reconstructed expression data and the original gene expression data.

[0103] Optionally, the spatial transcriptomics clustering device may further include: a perturbation feature module for constructing perturbation gene expression data of the target cluster object; and performing graph convolutional coding processing on the perturbation gene expression data and the adjacency relationship data based on the graph convolutional coding model to obtain perturbation feature data of the target cluster object.

[0104] Optionally, the function construction unit can be used to: construct the representation difference loss function between the perturbation feature data and the representation feature data.

[0105] Optionally, the spatial transcriptomics clustering device may further include: a spatial neighborhood aggregation module, used to perform spatial neighborhood aggregation processing on the perturbation feature data and the expression feature data respectively, to obtain a perturbation graph neighborhood aggregation representation and an original graph neighborhood aggregation representation.

[0106] Optionally, the function construction unit can be used to: construct the representation consistency loss function between the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation.

[0107] Optionally, the clustering analysis module 450 can be used to perform spatial clustering analysis on the target clustering object based on the target gene expression data.

[0108] The spatial transcriptomics clustering device provided in the embodiments of the present invention can execute the spatial transcriptomics clustering method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0109] Example 4 Figure 5 This is a schematic diagram of the structure of a spatial transcriptomics clustering device provided in Embodiment 4 of the present invention, as shown below. Figure 5 As shown, the spatial transcriptomics clustering device includes a processor 510, a memory 520, an input device 530, and an output device 540; the number of processors 510 in the spatial transcriptomics clustering device can be one or more. Figure 5 Taking a processor 510 as an example; the processor 510, memory 520, input device 530, and output device 540 in the spatial transcriptomics clustering device can be connected via a bus or other means. Figure 5 Taking the example of a connection between China and Israel via a bus.

[0110] The memory 520, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the spatial transcriptomics clustering method in this embodiment of the invention (e.g., the data acquisition module 410, expression feature module 420, spatial feature module 430, reconstruction and fusion module 440, and cluster analysis module 450 in the spatial transcriptomics clustering device). The processor 510 executes various functional applications and data processing of the spatial transcriptomics clustering device by running the software programs, instructions, and modules stored in the memory 520, thereby realizing the aforementioned spatial transcriptomics clustering method.

[0111] The memory 520 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 520 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 520 may further include memory remotely located relative to the processor 510, which can be connected to a spatial transcriptomics clustering device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0112] Input device 530 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the spatial transcriptomics clustering device. Output device 540 may include a display device such as a screen.

[0113] Example 5 Embodiment 5 of the present invention also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a spatial transcriptomics clustering method, including: Obtain the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object; Based on the graph convolutional coding model, the original gene expression data and the adjacency relationship data are processed by graph convolutional coding to obtain the expression feature data of the target cluster object; Based on the spatial location coding model, the spatial location data is processed by spatial location coding to obtain the spatial feature data of the target clustering object; Based on the reconstruction and fusion model, the expression feature data and the spatial feature data are reconstructed and fused to obtain the target gene expression data. Spatial clustering analysis is performed on the target cluster objects based on the target gene expression data.

[0114] Of course, the computer-executable instructions provided in the embodiments of the present invention are not limited to the method operations described above, but can also perform related operations in the spatial transcriptomics clustering method provided in any embodiment of the present invention.

[0115] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0116] It is worth noting that in the embodiments of the above-mentioned spatial transcriptomics clustering device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0117] Although the present invention has been described in detail above with general descriptions, specific embodiments, and experiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A spatial transcriptomics clustering method, characterized in that, include: Obtain the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object; Based on the graph convolutional coding model, the original gene expression data and the adjacency relationship data are processed by graph convolutional coding to obtain the expression feature data of the target cluster object; Based on the spatial location coding model, the spatial location data is processed by spatial location coding to obtain the spatial feature data of the target clustering object; Based on the reconstruction and fusion model, the expression feature data and the spatial feature data are reconstructed and fused to obtain the target gene expression data. Spatial clustering analysis is performed on the target cluster objects based on the target gene expression data.

2. The method according to claim 1, characterized in that, The reconstruction and fusion model-based process reconstructs and fuses the expression feature data and the spatial feature data to obtain target gene expression data, including: Based on the reconstruction fusion model, the expression feature data and the spatial feature data are decoded and reconstructed respectively to obtain the first reconstructed expression data and the second reconstructed expression data. The first reconstructed expression data and the second reconstructed expression data are subjected to expression fusion processing to obtain the target gene expression data.

3. The method according to claim 2, characterized in that, Before performing spatial clustering analysis on the target clustering object based on the target gene expression data, the method further includes: The graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model are jointly trained to optimize and update the target gene expression data.

4. The method according to claim 3, characterized in that, The joint training of parameters for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct a joint loss function for the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model; The graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model are jointly trained according to the joint loss function. The joint loss function consists of a feature reconstruction loss function, a location reconstruction loss function, a representation difference loss function, and a representation consistency loss function.

5. The method according to claim 4, characterized in that, The joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the feature reconstruction loss function between the first reconstructed expression data and the original gene expression data; Construct the positional reconstruction loss function between the second reconstructed expression data and the original gene expression data.

6. The method according to claim 4, characterized in that, After obtaining the original gene expression data, spatial location data, and adjacency relationship data of the target cluster object, the method further includes: Construct the perturbed gene expression data of the target cluster objects; Based on the graph convolutional coding model, the perturbation gene expression data and the adjacency relationship data are processed by the graph convolutional coding to obtain the perturbation feature data of the target cluster object; The joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the representation difference loss function between the perturbation feature data and the representation feature data.

7. The method according to claim 6, characterized in that, After performing graph convolutional coding on the perturbed gene expression data and the adjacency relationship data based on the graph convolutional coding model to obtain the perturbed feature data of the target cluster object, and performing graph convolutional coding on the original gene expression data and adjacency relationship data based on the graph convolutional coding model to obtain the expression feature data of the target cluster object, the method further includes: Spatial neighborhood aggregation processing is performed on the perturbation feature data and the expression feature data respectively to obtain the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation; The joint loss function for constructing the graph convolutional coding model, the spatial location coding model, and the reconstruction fusion model includes: Construct the representation consistency loss function between the perturbation graph neighborhood aggregation representation and the original graph neighborhood aggregation representation.

8. A spatial transcriptomics clustering device, characterized in that, include: The data acquisition module is used to acquire the original gene expression data, spatial location data, and adjacency relationship data of the target clustering object; The expression feature module is used to perform graph convolutional coding on the original gene expression data and the adjacency relationship data based on a graph convolutional coding model to obtain the expression feature data of the target cluster object; The spatial feature module is used to perform spatial location encoding processing on the spatial location data based on the spatial location encoding model to obtain the spatial feature data of the target clustering object; The reconstruction and fusion module is used to reconstruct and fuse the expression feature data and the spatial feature data based on the reconstruction and fusion model to obtain the target gene expression data. The clustering analysis module is used to perform spatial clustering analysis on the target cluster objects based on the target gene expression data.

9. A spatial transcriptomics clustering device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the spatial transcriptomics clustering method as described in any one of claims 1-7.

10. A storage medium containing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a computer processor, are used to perform the spatial transcriptomics clustering method as described in any one of claims 1-7.

Images