Method and apparatus for denoising spatial transcriptome data

Abstract

The application provides a method and device for denoising spatial transcriptome data. The method for denoising spatial transcriptome data comprises: representing gene expression and tissue image feature information of spatial transcriptome data through an RGB image; clustering and dividing each sampling point based on three-dimensional data information of the RGB image; and restoring real gene expression of the same type of sampling point data in each tissue region after clustering. The method and device for denoising spatial transcriptome data provided in the application compress and reduce dimensions of gene expression and tissue image feature information of spatial transcriptome data, then perform clustering and division, and then perform denoising by using the data after clustering and division, so that the real gene expression is restored, and the accuracy of downstream analysis results of data is improved.

Description

Technical Field

[0001] This application relates to the field of data mining technology, and in particular to a method and apparatus for denoising spatial transcriptome data. Background Technology

[0002] Gene expression exhibits both time and spatial specificity. Analysis of genome-wide gene expression in tissues typically employs spatial transcriptomics techniques, combining spatial information with morphological data to map the locations of all gene expression events, thus obtaining a complete gene expression map of the tissue. This provides not only the tissue's gene expression profile but also its spatial location information.

[0003] In existing technologies, spatial transcriptomics technology captures transcripts at each sampling point using microarray chips. These transcripts contain location information, but the capture of information at a sampling point is affected by messenger RNA flowing out from nearby sampling points. Furthermore, the messenger RNA of the sampling point itself also flows out to nearby sampling points, which contaminates the unique molecular identifier (UMI) count of that sampling point. This results in incomplete information captured by the sampling point, thereby reducing the accuracy of spatial transcriptomics data. Summary of the Invention

[0004] This application provides a method and apparatus for denoising spatial transcriptome data, in order to solve the technical problem that the low accuracy of spatial transcriptome data in the prior art makes it impossible to guarantee the accuracy of downstream data analysis.

[0005] In a first aspect, embodiments of this application provide a method for denoising spatial transcriptome data, including:

[0006] RGB images are used to represent gene expression and tissue image features of spatial transcriptome data.

[0007] Each sampling point is clustered based on the three-dimensional data information of the RGB image;

[0008] The true gene expression is restored from the data of similar sampling points within each clustered tissue region.

[0009] In some embodiments, the gene expression and tissue image feature information representing spatial transcriptome data via RGB images includes:

[0010] Based on the original gene expression matrix of spatial transcriptome data and the corresponding tissue image feature information, a graph neural network is used to generate a relationship graph of connection sampling points that retains the tissue image feature information.

[0011] Based on the relationship graph of sampling points, a graph autoencoder is used to learn the 3D embedding of sampling points.

[0012] Map 3D embeddings to RGB images.

[0013] In some embodiments, the original gene expression matrix based on spatial transcriptome data and the corresponding tissue image information data are used to generate a graph of connection sampling points that preserves tissue image features using a graph neural network, including:

[0014] Determine the reconstructed gene expression matrix of spatial transcriptome data based on the original gene expression matrix of spatial transcriptome data;

[0015] Based on the reconstructed gene expression matrix and the corresponding tissue image feature information data, a graph neural network is used to generate a spatial embedding for each sampling point.

[0016] Euclidean distance is used to distinguish directly adjacent sampling points;

[0017] The spatial relationships between directly adjacent sampling points are constructed into an adjacency matrix, generating a relationship graph of connected sampling points that preserves tissue image information.

[0018] In some embodiments, the encoder of the graph autoencoder consists of a two-layer graph convolutional network for learning low-dimensional graph embeddings.

[0019] The decoder of the graph autoencoder is set as the inner product between graph embeddings;

[0020] The decoder of the graph autoencoder uses sigmoid as the activation function;

[0021] The goal of the graph autoencoder learning is to minimize the cross-entropy between the adjacency matrix of the input sampling points and the reconstructed gene expression matrix.

[0022] In some embodiments, mapping the 3D embedding to an RGB image includes:

[0023] Based on the panchromatic spectrum, three values, R, G, and B, are assigned to the three-dimensional embedding of each sampling point;

[0024] An RGB image is generated based on the coordinates and diameter of each sampling point in the full-resolution organized image, as well as the RGB value assigned to each sampling point.

[0025] In some embodiments, clustering each sampling point based on the three-dimensional data information of the RGB image includes:

[0026] An adjacency matrix of sampling points is generated based on the three-dimensional data information of the RGB image; the weight of adjacent edges is the similarity of the RGB values ​​of two adjacent sampling points;

[0027] The adjacency matrix is ​​clustered using a graph clustering algorithm to obtain the clustering results of the sampled points.

[0028] In some embodiments, restoring the true gene expression from the same sampling point data within each clustered tissue region includes:

[0029] Gene expression in each tissue region is modeled as a function of gene expression levels in similar sampling points within that tissue region, and data from each sampling point to other sampling points and data from other sampling points to other sampling points are read.

[0030] Based on the data that seeps from each sampling point to other sampling points and the data that flows from other sampling points, the gradient descent algorithm is used to estimate the seepage rate and the size of the affected neighborhood for each type of sampling point.

[0031] Based on the exudation rate and the size of the affected neighborhood of each sampling point, the potential expression level of each sampling point is estimated using the expectation-maximization EM algorithm.

[0032] The spatial transcriptome data of each sampling point are corrected based on the potential expression level of each sampling point to restore the true gene expression of the sampling point.

[0033] Secondly, embodiments of this application provide an apparatus for denoising spatial transcriptome data, comprising:

[0034] The representation module is used to represent gene expression and tissue image feature information of spatial transcriptome data using RGB images;

[0035] The clustering module is used to cluster each sampling point based on the three-dimensional data information of the RGB image;

[0036] The recovery module is used to recover the true gene expression from the same type of sampling points within each clustered tissue region.

[0037] Thirdly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for denoising spatial transcriptome data as described in the first aspect above.

[0038] Fourthly, embodiments of this application also provide a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for denoising spatial transcriptome data as described in the first aspect above.

[0039] Fifthly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the method for denoising spatial transcriptome data as described in the first aspect above.

[0040] The method and apparatus for denoising spatial transcriptome data provided in this application compress and reduce the gene expression and tissue image feature information of spatial transcriptome data and map it to an RGB image. After clustering each sampling point based on the three-dimensional data information of the RGB image, the data of the same type of sampling points in each tissue region after clustering are restored to the true gene expression, thereby improving the accuracy of downstream data analysis and obtaining reliable analysis results. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart illustrating a method for denoising spatial transcriptome data provided in an embodiment of this application;

[0043] Figure 2 This is a schematic diagram of a device for denoising spatial transcriptome data provided in an embodiment of this application;

[0044] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0045] Spatial transcriptomics is a powerful and widely used method for analyzing genome-wide gene expression in tissues. The rise of spatial transcriptomics technology not only provides gene expression profiles of tissues but also offers tissue imaging information, making it highly valuable in the medical field.

[0046] Spatial transcriptomics technology captures transcripts at each sampling point using a microarray chip, preserving spatial information. Ideally, the UMI (Unique Metric Index) at a given sampling point would represent the gene expression at that point within the tissue. However, since the information captured at each sampling point is not entirely representative of the expression at that point, messenger RNA flowing out from nearby sampling points can cause significant contamination of UMI counts, while messenger RNA flowing from that point to nearby sampling points can lead to a loss of the true UMI count, resulting in errors in clustering and tissue segmentation, thus affecting downstream analysis.

[0047] Existing methods for denoising single-cell transcriptome data can remove contamination from certain sources present in spatial transcriptome experiments, but they are insufficient to correct gene expression profile data because they cannot correct contamination caused by the spatial arrangement of sites, such as contamination due to point exchange.

[0048] Based on the above-mentioned technical problems, this application proposes a method for denoising spatial transcriptome data. The gene expression and tissue image feature information of the spatial transcriptome data are compressed and reduced in dimensionality and then mapped to an RGB image. Based on the three-dimensional data information of the RGB image, each sampling point is clustered and divided. The data of the same type of sampling points in each tissue region after clustering are restored to restore the true gene expression, improve the accuracy of downstream data analysis, and obtain reliable analysis results.

[0049] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0050] Figure 1 This is a flowchart illustrating a method for denoising spatial transcriptome data provided in an embodiment of this application, as shown below. Figure 1 As shown in the embodiments of this application, a method for denoising spatial transcriptome data is provided, the method comprising:

[0051] Step 101: Represent the gene expression and tissue image feature information of spatial transcriptome data using RGB images.

[0052] Specifically, RGB images are used to represent or visualize gene expression and tissue image feature information of the dimensionality-reduced spatial transcriptome data. The spatial transcriptome data includes the original gene expression matrix for each sampling point. The tissue image feature information includes spatial location information and other relevant information.

[0053] For example, the reconstructed gene expression matrix of the spatial transcriptome data is first determined based on the original gene expression matrix of the spatial transcriptome data. Then, based on the reconstructed gene expression matrix and the corresponding spatial information data, a graph neural network (GNN) is used to generate a relational graph of connection sampling points that retains spatial information. Based on this relational graph, an autoencoder composed of a graph convolutional network of two layers is used to learn the three-dimensional embedding of the sampling points, and the three-dimensional embedding is mapped to the color channel in an RGB image. The gene expression and organization image feature information of the dimensionality-reduced spatial transcriptome data are then visualized using the RGB image.

[0054] Step 102: Cluster each sampling point based on the three-dimensional data information of the RGB image.

[0055] Specifically, an adjacency matrix of sampling points is generated based on the three-dimensional (R, G, B) data information of the RGB image. Then, a graph clustering algorithm is used to cluster the adjacency matrix to obtain the clustering result for the sampling point. Each sampling point corresponds to a spatial node.

[0056] Step 103: Restore the true gene expression of the same type of sampling points in each clustered tissue region.

[0057] Specifically, based on a pre-established flow model, the potential expression level of each sampling point (tissue region) is estimated using the Expectation Maximization (EM) algorithm for clustered sampling points (tissue regions). Then, the true gene expression of each sampling point is recovered based on the estimated potential expression level. The tissue region is a region within a tissue slice image. The model is established as a function of gene expression within each tissue region and the amount of gene expression present in that region.

[0058] For example, gene expression in each tissue region is modeled as a function of the gene expression levels present in that tissue region. Data on leakage from each sampling point to other sampling points and data on inflow from other sampling points are read. Then, based on this data, the gradient descent algorithm is used to estimate the leakage rate and the size of the affected neighborhood for each type of sampling point. Based on the leakage rate and the size of the affected neighborhood for each type of sampling point, the EM algorithm is used to estimate the potential expression level, thereby correcting the spatial transcriptome data.

[0059] The method for denoising spatial transcriptome data provided in this application compresses and reduces the dimensionality of gene expression and tissue image feature information of spatial transcriptome data and maps it to an RGB image. Based on the three-dimensional data information of the RGB image, each sampling point is clustered and divided. The sampling point data of each tissue region after clustering is restored to the true gene expression, thereby improving the accuracy of downstream data analysis results.

[0060] In some embodiments, the gene expression and tissue image feature information representing spatial transcriptome data via RGB images includes:

[0061] Based on the original gene expression matrix of spatial transcriptome data and the corresponding tissue image feature information, a graph neural network is used to generate a relationship graph of connection sampling points that retains the tissue image feature information.

[0062] Based on the relationship graph of sampling points, a graph autoencoder is used to learn the 3D embedding of sampling points.

[0063] Map 3D embeddings to RGB images.

[0064] Specifically, the original gene expression matrix of the spatial transcriptome data is obtained by normalizing the sequencing library size and performing a logarithmic transformation using the number of unique molecular identifiers (UMIs) per million (UPM).

[0065] For example, the raw gene expression matrix of the spatial transcriptome data, as input, is embedded into a low-dimensional vector using an autoencoder. The encoder learns the embedding from the raw gene expression matrix of the input spatial transcriptome data (by default, the raw gene expression matrix of the top 2000 highly variable genes is selected), and then the decoder performs matrix reconstruction to obtain the reconstructed gene expression matrix of the spatial transcriptome data. Minimizing the mean squared error (MSE) is used as the loss function, with the goal of maximizing the similarity between the original matrix and the reconstructed matrix during training.

[0066] After determining the reconstructed gene expression matrix of the spatial transcriptome data, a graph neural network (GNN) is used to generate a relationship graph of connection sampling points that preserves the tissue image feature information, based on the reconstructed gene expression matrix and the corresponding tissue image feature information. The tissue image feature information may include the spatial location of the reconstructed gene expression matrix and the spatial relationship between the reconstructed gene expression matrix and other reconstructed gene expression matrices.

[0067] For example, a GNN is used to generate a spatial embedding for each spatial sampling point using the preprocessed reconstructed gene expression matrix and the corresponding tissue image feature information. Then, directly adjacent spatial sampling sites are distinguished, and this direct adjacency spatial relationship is constructed into an adjacency matrix. Finally, a relationship graph connecting spatial sampling points that retains tissue image feature information is generated.

[0068] After obtaining the relational graph of the connected sampling points that preserves the feature information of the tissue image, in order to preserve the topological relationships in the graph, a graph autoencoder is used to learn the three-dimensional embedding of the sampling points based on the relational graph of the sampling points; the learned low-dimensional embedding is used to represent and preserve the potential relationships from the spatial transcriptome data model graph.

[0069] In some embodiments, the encoder of the graph autoencoder consists of a two-layer graph convolutional network for learning low-dimensional graph embeddings.

[0070] The decoder of the graph autoencoder is set as the inner product between graph embeddings;

[0071] The decoder of the graph autoencoder uses sigmoid as the activation function;

[0072] The goal of the graph autoencoder learning is to minimize the cross-entropy between the adjacency matrix of the input sampling points and the reconstructed gene expression matrix.

[0073] For example, the encoder of a graph autoencoder consists of a two-layer graph convolutional network (GCN) used to learn low-dimensional graph embeddings. The output dimensions of the first and second GCN layers are set to 32 and 3, respectively, and are assigned RGB values ​​based on the three color channels. The learning rate is set to 0.001. The decoder of the graph autoencoder is defined as the inner product between the graph embeddings, using sigmoid as the activation function. The goal of the graph autoencoder is to minimize the cross-entropy between the adjacency matrix of the input space sampling points and the reconstructed matrix.

[0074] Finally, the 3D embedding of the sampling points is mapped onto the RGB image.

[0075] For example, the 3D embedding of sampling points is mapped to the red, green, and blue channels in the RGB space, with each sampling point corresponding to a set of RGB values, and an RGB image is generated based on the RGB values ​​of all sampling points.

[0076] The method for denoising spatial transcriptome data provided in this application obtains a sampling point relationship graph that preserves spatial information by reconstructing gene expression from the spatial transcriptome data, and uses a graph autoencoder to learn the three-dimensional embedding of the sampling points to generate an RGB image to visualize the embedding. This enables the sampling points to be clustered by combining gene expression and spatial information, improving the accuracy of clustering and thus achieving a better denoising effect.

[0077] In some embodiments, the original gene expression matrix based on spatial transcriptome data and the corresponding tissue image feature information data are used to generate a graph of connection sampling points that preserves tissue image feature information using a graph neural network, including:

[0078] Determine the reconstructed gene expression matrix of spatial transcriptome data based on the original gene expression matrix of spatial transcriptome data;

[0079] Based on the reconstructed gene expression matrix and the corresponding tissue image feature information data, a graph neural network is used to generate a spatial embedding for each sampling point.

[0080] Euclidean distance is used to distinguish directly adjacent sampling points;

[0081] The spatial relationships between directly adjacent sampling points are constructed into an adjacency matrix, generating a relational graph of connected sampling points that preserves tissue image feature information.

[0082] Specifically, the raw gene expression matrix of the spatial transcriptome data is first preprocessed by inputting it into an autoencoder, embedding the spatial transcriptome expression data into a low-dimensional vector. The encoder learns the embedding, and the decoder reconstructs the matrix to obtain the reconstructed gene expression matrix of the spatial transcriptome data. A loss function is used to ensure the accuracy of the reconstruction.

[0083] After determining the reconstructed gene expression matrix of the spatial transcriptome data, a graph neural network is used to generate a spatial embedding for each spatial sampling point based on the reconstructed gene expression matrix and the corresponding tissue image feature information data.

[0084] Then, Euclidean distance is used to distinguish directly adjacent sampling points. An adjacency matrix is ​​constructed based on the spatial relationships between directly adjacent spatial sampling points, and a relationship graph of connected spatial sampling points that preserves tissue image feature information is generated.

[0085] For example, by calculating the Euclidean distance between every two spatial sampling points, directly adjacent spatial sampling points are distinguished. This direct adjacency spatial relationship of spatial sampling points is constructed into an adjacency matrix. Based on the direct adjacency spatial relationship and other spatial positional relationships of spatial sampling points, a relational graph of sampling points with spatial information is generated. In the 10xVisium spatial gene expression platform, each spatial sampling point has a maximum of six directly adjacent spatial sampling sites. Therefore, in the corresponding relational graph, the maximum node degree of all nodes is 6.

[0086] The method for denoising spatial transcriptome data provided in this application constructs an adjacency matrix based on the tissue image feature information corresponding to the reconstructed gene expression matrix, and generates a relational graph of connected sampling points that retains the tissue image feature information. This enables better correction of noise in spatial transcriptome data using spatial information, improves the denoising effect, and allows for downstream analysis based on accurate spatial transcriptome data, ensuring the reliability of downstream analysis.

[0087] In some embodiments, mapping the 3D embedding to an RGB image includes:

[0088] Based on the panchromatic spectrum, three values, R, G, and B, are assigned to the three-dimensional embedding of each sampling point;

[0089] An RGB image is generated based on the coordinates and diameter of each sampling point in the full-resolution organized image, as well as the RGB value assigned to each sampling point.

[0090] Specifically, the panchromatic spectrum refers to the spectrum ranging from 0 to 255 pixels. The full-resolution tissue image can be obtained from a 10x Visium platform. A set of RGB values, including R, G, and B values, is assigned to each sampling point based on the panchromatic spectrum's 3D embedding; that is, a color is assigned. Then, based on the coordinates and diameter of each sampling point in the full-resolution tissue image, combined with the RGB values ​​assigned to each sampling point, an RGB image is generated.

[0091] For example, the 3D embedding is intuitively mapped to the red, green, and blue channels in the image's RGB space. Normalized to the RGB color space based on the full-spectrum color (pixel range from 0 to 255), each point's embedding is assigned a unique color. The coordinates and diameter of each spatial sampling point in the image are organized using the full resolution provided by the 10x Visium platform. All points are plotted using the synthetic color assigned to each spatial sampling point, and a full-size RGB image is reconstructed to describe the spatial representation in the original spatial coordinate system.

[0092] The method for denoising spatial transcriptome data provided in this application assigns a unique color to each sampling point, enabling the sampling point to be displayed in two-dimensional space. By combining the coordinates and diameter of each sampling point, a full-size RGB image is reconstructed, allowing the spatial relationship and gene enlargement information of each sampling point to be visualized and presented in the RGB image. This improves the accuracy of clustering based on the RGB image, thereby reducing noise and obtaining more accurate spatial transcriptome data.

[0093] In some embodiments, clustering each sampling point based on the three-dimensional data information of the RGB image includes:

[0094] An adjacency matrix of sampling points is generated based on the three-dimensional data information of the RGB image; the weight of adjacent edges is the similarity of the RGB values ​​of two adjacent sampling points;

[0095] The adjacency matrix is ​​clustered using a graph clustering algorithm to obtain the clustering results of the sampled points.

[0096] The method for denoising spatial transcriptome data provided in this application improves the accuracy of cluster analysis results by constructing an adjacency matrix to cluster the dimensionality-reduced sampling points, thereby more effectively denoising each component and ensuring the reliability of downstream data analysis.

[0097] In some embodiments, restoring the true gene expression from the same sampling point data within each clustered tissue region includes:

[0098] Gene expression in each tissue region is modeled as a function of gene expression levels in similar sampling points within that tissue region, and data from each sampling point to other sampling points and data from other sampling points to other sampling points are read.

[0099] Based on the data that seeps from each sampling point to other sampling points and the data that flows from other sampling points, the gradient descent algorithm is used to estimate the seepage rate and the size of the affected neighborhood for each type of sampling point.

[0100] Based on the exudation rate and the size of the affected neighborhood of each sampling point, the potential expression level of each sampling point is estimated using the expectation-maximization algorithm.

[0101] The spatial transcriptome data of each sampling point are corrected based on the potential expression level of each sampling point to restore the true gene expression of the sampling point.

[0102] Specifically, gene expression in each tissue region is modeled as a function of gene expression levels. Data on leakage from each sampling point to other sampling points and data on inflow from other sampling points are read. Then, based on the read data, the leakage rate and the size of the affected neighborhood of each type of sampling point are estimated using the gradient descent algorithm. Based on the leakage rate and the size of the affected neighborhood of each type of sampling point, the potential expression level of each sampling point is estimated using the expectation-maximization algorithm (EM algorithm). Based on the potential expression level of each sampling point, the spatial transcriptome data of each sampling point is corrected to restore the true gene expression of the sampling point, i.e., the denoised gene expression.

[0103] By using the EM algorithm to estimate potential expression levels, noise reduction is achieved on the sampling points of clustered tissue regions, and expression data is corrected to obtain more accurate spatial transcriptome data.

[0104] The method for denoising spatial transcriptome data provided in this application combines spatial information of sampling sites to simulate the diffusion of RNA molecules between sampling sites, thereby correcting noise in the data using spatial information. A deep learning framework is used to dimensionality-reduced, compressed, and visualize the tissue image feature information and gene expression information of each point. Given inputs such as gene expression or RNA rate, a three-dimensional embedding with a spatially preserved graph neural network is learned from spatial transcriptomics, and the embedding is visualized by mapping it to color channels in an RGB image.

[0105] After obtaining the RGB image, the sampling points are clustered and denoised using the three-dimensional data information in the RGB image, which greatly improves the denoising effect and the accuracy of transcriptome data, and is beneficial to the analysis of downstream idle data.

[0106] Figure 2This is a schematic diagram of a device for denoising spatial transcriptome data provided in an embodiment of this application, as shown below. Figure 2 As shown in the figure, this application provides an apparatus for denoising spatial transcriptome data, including a representation module 201, a clustering module 202, and a recovery module 203.

[0107] The representation module 201 is used to represent gene expression and tissue image feature information of spatial transcriptome data through RGB images.

[0108] The clustering module 202 is used to cluster each sampling point based on the three-dimensional data information of the RGB image.

[0109] The recovery module 203 is used to recover the true gene expression from the same type of sampling point data in each tissue region after clustering.

[0110] In some embodiments, the representation module includes:

[0111] The first generation unit is used to generate a graph diagram of connection sampling points that retains tissue image feature information based on the original gene expression matrix of spatial transcriptome data and the corresponding tissue image feature information using a graph neural network.

[0112] The learning unit is used to learn the 3D embedding of the sampling points based on the relationship graph of the sampling points using a graph autoencoder.

[0113] Mapping unit, used to map 3D embeddings to RGB images.

[0114] In some embodiments, the generation unit includes:

[0115] Determine subunits for determining the reconstructed gene expression matrix of spatial transcriptome data based on the original gene expression matrix of spatial transcriptome data;

[0116] The first generation subunit is used to generate a spatial embedding for each sampling point based on the reconstructed gene expression matrix and the corresponding tissue image feature information data using a graph neural network.

[0117] The molecular unit is used to distinguish directly adjacent sampling points using Euclidean distance;

[0118] The second generation subunit constructs an adjacency matrix for the direct adjacent spatial relationships of sampling points, generating a relational graph of connected sampling points that preserves tissue image feature information.

[0119] In some embodiments, the encoder of the graph autoencoder consists of a two-layer graph convolutional network for learning low-dimensional graph embeddings.

[0120] The decoder of the graph autoencoder is set as the inner product between graph embeddings;

[0121] The decoder of the graph autoencoder uses sigmoid as the activation function;

[0122] The goal of the graph autoencoder learning is to minimize the cross-entropy between the adjacency matrix of the input sampling points and the reconstructed gene expression matrix.

[0123] In some embodiments, the mapping unit includes:

[0124] The allocation subunit is used to assign three values, R, G, and B, to each sampling point based on the three-dimensional embedding of the panchromatic spectrum.

[0125] The third generation subunit is used to generate an RGB image based on the coordinates and diameter of each sampling point in the full-resolution organized image, as well as the RGB value assigned to each sampling point.

[0126] In some embodiments, the clustering module includes:

[0127] The second generation unit is used to generate an adjacency matrix of sampling points based on the three-dimensional data information of the RGB image; the weight of the adjacent edge is the similarity of the RGB values ​​of two adjacent sampling points;

[0128] Clustering units are used to cluster the adjacency matrix using graph clustering algorithms to obtain the clustering results of the sampled points.

[0129] In some embodiments, the recovery module includes:

[0130] The reading unit is used to model the gene expression in each tissue region as a function of the gene expression levels present in similar sampling points within the tissue region, and to read the data that seeps from each sampling point to other sampling points and the data that flows in from other sampling points.

[0131] The first estimation unit is used to estimate the leakage rate and the size of the affected neighborhood of each type of sampling point based on the data that seeps out of each sampling point to other sampling points and the data that flows in from other sampling points, using the gradient descent algorithm.

[0132] The second estimation unit is used to estimate the potential expression level of each sampling point based on the exudation rate of each type of sampling point and the size of the affected neighborhood using the expectation-maximization EM algorithm.

[0133] The correction unit is used to correct the spatial transcriptome data of each sampling point based on the potential expression level of each sampling point, thereby restoring the true gene expression of the sampling point. Specifically, the above-described method and apparatus for denoising spatial transcriptome data provided in this application embodiment can implement all the method steps implemented in the above-described method embodiment for denoising spatial transcriptome data, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0134] It should be noted that the division of units / modules in the above embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0135] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 3 As shown, the electronic device may include: a processor 301, a communication interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other via the communication bus 304. The processor 301 can call logical instructions in the memory 303 to execute a method for denoising spatial transcriptome data, the method including:

[0136] RGB images are used to represent gene expression and tissue image features of spatial transcriptome data.

[0137] Each sampling point is clustered based on the three-dimensional data information of the RGB image;

[0138] The true gene expression is restored from the data of similar sampling points within each clustered tissue region.

[0139] Specifically, the processor 301 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0140] When the logical instructions in memory 303 can be implemented as software functional units and sold or used as independent products, they can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] In some embodiments, a computer program product is also provided, the computer program product including a computer program that can be stored on a non-transitory computer-readable storage medium, and when the computer program is executed by a processor, the computer is able to perform the method for denoising spatial transcriptome data provided in the above-described method embodiments, the method comprising:

[0142] RGB images are used to represent gene expression and tissue image features of spatial transcriptome data.

[0143] Each sampling point is clustered based on the three-dimensional data information of the RGB image;

[0144] The true gene expression is restored from the data of similar sampling points within each clustered tissue region.

[0145] Specifically, the computer program product provided in this application embodiment can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0146] In some embodiments, a computer-readable storage medium is also provided, the computer-readable storage medium storing a computer program, the computer program being configured to cause a computer to perform the method for denoising spatial transcriptome data provided in the above method embodiments, the method comprising:

[0147] RGB images are used to represent gene expression and tissue image features of spatial transcriptome data.

[0148] Each sampling point is clustered based on the three-dimensional data information of the RGB image;

[0149] The true gene expression is restored from the data of similar sampling points within each clustered tissue region.

[0150] Specifically, the computer-readable storage medium provided in the embodiments of this application can implement all the method steps implemented in the above method embodiments and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiments and the beneficial effects will not be described in detail.

[0151] It should be noted that the computer-readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic storage (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical storage (e.g., CD, DVD, BD, HVD), and semiconductor storage (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD)).

[0152] It should also be noted that the terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, and the number of objects is not limited. For example, the first object can be one or more.

[0153] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0154] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0155] In this application, "determining B based on A" means that factor A must be considered when determining B. It is not limited to "B can be determined based solely on A," but should also include: "determining B based on A and C," "determining B based on A, C, and E," "determining C based on A, and further determining B based on C," etc. It can also include using A as a condition for determining B, for example, "when A satisfies the first condition, B is determined using the first method"; or "when A satisfies the second condition, B is determined," or "when A satisfies the third condition, B is determined based on the first parameter," etc. Of course, it can also be a condition where A is a factor in determining B, for example, "when A satisfies the first condition, C is determined using the first method, and B is further determined based on C," etc.

[0156] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0157] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-executable instructions. These computer-executable instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0158] These processor-executable instructions may also be stored in a processor-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the processor-readable memory produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0159] These processors can execute instructions that can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0160] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for denoising spatial transcriptome data, characterized in that, include: RGB images are used to represent gene expression and tissue image features of spatial transcriptome data. Each sampling point is clustered based on the three-dimensional data information of the RGB image; The true gene expression is restored from the data of similar sampling points within each tissue region after clustering; The gene expression and tissue image feature information of spatial transcriptome data represented by RGB images includes: Based on the original gene expression matrix of spatial transcriptome data and the corresponding tissue image feature information, a graph neural network is used to generate a relationship graph of connection sampling points that retains the tissue image feature information. Based on the relationship graph of sampling points, a graph autoencoder is used to learn the 3D embedding of sampling points. Map 3D embeddings to RGB images.

2. The method for denoising spatial transcriptome data according to claim 1, characterized in that, The original gene expression matrix based on spatial transcriptome data and the corresponding tissue image feature information data are used to generate a graph of connection sampling points that preserves tissue image feature information using a graph neural network, including: Determine the reconstructed gene expression matrix of spatial transcriptome data based on the original gene expression matrix of spatial transcriptome data; Based on the reconstructed gene expression matrix and the corresponding tissue image feature information data, a graph neural network is used to generate a spatial embedding for each sampling point. Euclidean distance is used to distinguish directly adjacent sampling points; The spatial relationships between directly adjacent sampling points are constructed into an adjacency matrix, generating a relational graph of connected sampling points that preserves tissue image feature information.

3. The method for denoising spatial transcriptome data according to claim 2, characterized in that, The encoder of the graph autoencoder consists of a two-layer graph convolutional network for learning low-dimensional graph embeddings. The decoder of the graph autoencoder is set as the inner product between graph embeddings; The decoder of the graph autoencoder uses sigmoid as the activation function; The goal of the graph autoencoder is to minimize the cross-entropy between the adjacency matrix of the input sampling points and the reconstructed gene expression matrix.

4. The method for denoising spatial transcriptome data according to claim 1, characterized in that, The process of embedding and mapping 3D images to RGB images includes: Based on the panchromatic spectrum, three values, R, G, and B, are assigned to the three-dimensional embedding of each sampling point; An RGB image is generated based on the coordinates and diameter of each sampling point in the full-resolution organized image, as well as the RGB value assigned to each sampling point.

5. The method for denoising spatial transcriptome data according to claim 1, characterized in that, The clustering of each sampling point based on the three-dimensional data information of the RGB image includes: An adjacency matrix of sampling points is generated based on the three-dimensional data information of the RGB image; the weight of adjacent edges is the similarity of the RGB values ​​of two adjacent sampling points; The adjacency matrix is ​​clustered using a graph clustering algorithm to obtain the clustering results of the sampled points.

6. The method for denoising spatial transcriptome data according to claim 1, characterized in that, The process of restoring true gene expression from similar sampling points within each clustered tissue region includes: Gene expression in each tissue region is modeled as a function of gene expression levels in similar sampling points within that tissue region, and data from each sampling point to other sampling points and data from other sampling points to other sampling points are read. Based on the data that seeps from each sampling point to other sampling points and the data that flows from other sampling points, the gradient descent algorithm is used to estimate the seepage rate and the size of the affected neighborhood for each type of sampling point. Based on the exudation rate and the size of the affected neighborhood of each sampling point, the potential expression level of each sampling point is estimated using the expectation-maximization EM algorithm. The spatial transcriptome data of each sampling point are corrected based on the potential expression level of each sampling point to restore the true gene expression of the sampling point.

7. A device for denoising spatial transcriptome data, characterized in that, include: The representation module is used to represent gene expression and tissue image feature information of spatial transcriptome data using RGB images; The clustering module is used to cluster each sampling point based on the three-dimensional data information of the RGB image; The recovery module is used to recover the true gene expression from the same type of sampling points within each tissue region after clustering; The gene expression and tissue image feature information of spatial transcriptome data represented by RGB images includes: Based on the original gene expression matrix of spatial transcriptome data and the corresponding tissue image feature information, a graph neural network is used to generate a relationship graph of connection sampling points that retains the tissue image feature information. Based on the relationship graph of sampling points, a graph autoencoder is used to learn the 3D embedding of sampling points. Map 3D embeddings to RGB images.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for denoising spatial transcriptome data as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for denoising spatial transcriptome data as described in any one of claims 1 to 6.

Images