Intercellular interaction reconstruction method and system based on graph convolution and long and short term memory attention network

Abstract

The invention relates to an intercellular interaction reconstruction method and system based on graph convolution and a long and short term memory attention network. The method comprises the following steps: collecting spatial transcriptome data under single-cell or subcellular resolution to obtain a gene expression matrix corresponding to each cell; determining spatial proximity among the cells, and judging whether interaction exists among the cells so as to construct a cell map; combining graph convolution and a long-short-term memory attention network to form a feature extraction network to perform feature extraction on the cell graph to obtain potential feature representation; and decoding the potential feature representation by using an inner product to generate a new adjacency matrix so as to obtain the reconstructed intercellular interaction. By introducing the combination of graph convolution and the long and short term memory attention network, spatial correlation in a cell graph structure can be captured while the interaction between the cells is learned, the interaction between the cells is reconstructed, the interaction between the cells is better known, and the tissue homeostasis of an organism is researched.

Description

Technical Field

[0001] The present invention relates to the field of cell biorecognition technology, and in particular to a method and system for reconstructing cell-to-cell interaction based on graph convolution and long short-term memory attention network. Background Art

[0002] In biological systems, cells transmit information to each other directly or indirectly by secreting signal molecules. Cell-cell interactions can be dependent interactions with short-range contacts or can be achieved through secreted signal molecules. Multicellular life depends on the coordination of cell activities, which in turn depends on cell-cell interactions between different cell types and tissues of the organism. Cell-cell interactions can coordinate the development and balance of the organism and affect the physiological functions between cells and tissues to a certain extent. In addition, intercellular regulation can maintain the normal function of tissues and organs, while their disorder may lead to a variety of diseases, including cancer, autoimmune diseases and infections. Therefore, identifying cell-cell interactions is crucial to understanding the role of individual cells in maintaining tissue homeostasis and responding to the microenvironment. The study of cell-cell interactions is highly dependent on the spatial location information of cells, and it also has high requirements for gene expression in individual cells. With the continuous development of spatially resolved transcriptome analysis technology, studies have been able to measure gene expression profiles at single-cell or subcellular resolution while retaining the spatial location information of cells.

[0003] Traditional cell-cell interaction prediction methods are often based on networks, machine learning methods, and spatial information methods. Among them, network-based cell-cell interaction prediction methods mainly use networks to calculate the ligand-receptor pair interaction scores of two cell types to study potential cell-cell interactions; machine learning-based methods mainly identify cell types through clustering, calculate the ligand-receptor pair scores of two cell types, and infer cell-cell interactions based on the scores; spatial information-based methods mainly fully characterize spatial location information and spatially adjacent cell clusters, and discover interactions between different cell types based on single-cell transcriptome data, spatial transcriptome data, etc. However, network-based and machine learning networks do not take into account the overall position distribution of cells in tissues, and are prone to false interactions; spatial information-based methods currently often do not take into account the time dependence of cell interactions, and the effect of inferring cell-cell interactions needs to be improved.

[0004] Therefore, traditional methods for obtaining cell-cell interactions often fail to take into account the time dependence of cell interactions and the overall positional distribution of cells in tissues, resulting in poor results in inferred cell-cell interactions. Summary of the invention

[0005] Based on this, in order to solve the above technical problems, a method and system for reconstructing cell-to-cell interactions based on graph convolution and long short-term memory attention network are provided, which can better capture spatial correlation, reconstruct cell-to-cell interactions, better understand cell-cell interactions, and study the tissue homeostasis of organisms.

[0006] A method for reconstructing cell-to-cell interactions based on graph convolution and long short-term memory attention network, the method comprising:

[0007] Collecting spatial transcriptome data at single-cell or sub-cellular resolution, and converting the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell;

[0008] Determining the spatial proximity between each cell according to the spatial transcriptome data, and judging whether there is interaction between the cells based on the spatial proximity to obtain a judgment result;

[0009] Taking each of the cells as a vertex of a graph structure, if the judgment result is that there is interaction between the cells, constructing edges of a graph structure for the two cells that have interaction according to a spatial connection network, taking a matrix describing cells and edges as an adjacency matrix, taking the gene expression matrix of each cell as a graph node feature of the graph structure, obtaining a node feature matrix, and constructing a cell graph;

[0010] Combining graph convolution and long short-term memory attention network to form a feature extraction network, and extracting features from the cell graph through the feature extraction network to obtain a potential feature representation;

[0011] The inner product is used to decode the latent feature representation to generate a new adjacency matrix and obtain the reconstructed cell-to-cell interactions.

[0012] In one embodiment, the gene expression matrix corresponding to each cell is obtained by conversion based on the spatial transcriptome data, including:

[0013] Searching for fuzzy-defined cells in the spatial transcriptome data, and removing the fuzzy-defined cells from the spatial transcriptome data to obtain updated spatial transcriptome data;

[0014] The updated spatial transcriptome data is converted to obtain the gene expression matrix corresponding to each cell.

[0015] In one embodiment, after converting the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell, the method further includes:

[0016] A logarithmic change normalization operation was performed on each of the gene expression matrices.

[0017] In one embodiment, constructing a cell map comprises:

[0018] Determine a vertex cell that is a vertex of a graph structure, and find a cell that interacts with the vertex cell to construct a target edge of the graph structure;

[0019] Using the matrix describing the vertex cells and the target edges as the target adjacency matrix, and determining the target node feature matrix of the graph structure according to the gene expression matrix corresponding to the vertex cells;

[0020] The vertices, target edges, target adjacency matrix, and target node feature matrix are combined and constructed to obtain a cell graph.

[0021] In one embodiment, a graph convolution and a long short-term memory attention network are combined to form a feature extraction network, including:

[0022] The graph convolution and long short-term memory attention network are combined in series to form a feature extraction network.

[0023] In one embodiment, the long short-term memory attention network includes a long short-term memory network, an activation function, an attention mechanism, and a skip connection;

[0024] The long short-term memory network is used to learn the temporal dependency of the interaction between cells; the attention mechanism is used to enhance the feature representation of the network; the jump connection is used to add the output feature of the long short-term memory network to the output feature of the attention mechanism;

[0025] The graph convolution is used to capture the spatial correlation of the graph structure and learn the relationship between cells

[0026] In one embodiment, extracting features from the cell map by the feature extraction network to obtain a potential feature representation includes:

[0027] Inputting the cell map into the feature extraction network, optimizing the feature matrix in the cell map through the long short-term memory attention network in the feature extraction network, and obtaining an optimized feature matrix;

[0028] The optimized feature matrix and the adjacency matrix are input into the graph convolutional network together, the spatial correlation of the graph structure is captured by the graph convolutional network, and the potential feature representation is obtained according to the spatial correlation of the graph structure.

[0029] In one embodiment, obtaining a potential feature representation according to the spatial correlation of the graph structure includes:

[0030] The spatial correlation of the graph structure is processed by the activation function and added to the input feature matrix using a skip connection to capture features at different levels;

[0031] The features at different levels are input again into the long short-term memory attention network and the graph convolutional network for encoding optimization, and are regularized to obtain potential feature representation.

[0032] A cell-to-cell interaction reconstruction system based on graph convolution and long short-term memory attention network, the system comprising:

[0033] A data collection module is used to collect spatial transcriptome data at single-cell or sub-cellular resolution, and convert the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell;

[0034] An interaction judgment module, used to determine the spatial proximity between each cell according to the spatial transcriptome data, and judge whether there is interaction between the cells based on the spatial proximity to obtain a judgment result;

[0035] A cell graph construction module, used to use each of the cells as a vertex of a graph structure, and if the judgment result is that there is interaction between cells, construct edges of a graph structure for two cells that have interaction according to a spatial connection network, use a matrix describing cells and edges as an adjacency matrix, use the gene expression matrix of each cell as a graph node feature of the graph structure, obtain a node feature matrix, and construct a cell graph;

[0036] A feature extraction module, used to combine graph convolution and long short-term memory attention network to form a feature extraction network, and extract features from the cell graph through the feature extraction network to obtain a potential feature representation;

[0037] The cell-to-cell interaction reconstruction module is used to decode the latent feature representation using the inner product, generate a new adjacency matrix, and obtain the reconstructed cell-to-cell interaction.

[0038] The above-mentioned cell-to-cell interaction reconstruction method and system based on graph convolution and long short-term memory attention network, by introducing a combination of graph convolution and long short-term memory attention network, can capture the spatial correlation in the cell graph structure while learning cell-to-cell interactions, thereby training the ability to extract and optimize features, improving the overall computational efficiency and accuracy, better capturing spatial correlation, reconstructing cell-to-cell interactions, and better understanding cell-cell interactions and studying the tissue homeostasis of organisms. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1A diagram of an application environment of a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network in one embodiment;

[0040] Figure 2 A schematic diagram of a flow chart of a method for reconstructing cell-to-cell interactions based on graph convolution and long short-term memory attention network in one embodiment;

[0041] Figure 3 A schematic diagram of a network framework for implementing feature extraction enhancement in one embodiment;

[0042] Figure 4 A structural block diagram of a cell-to-cell interaction reconstruction system based on graph convolution and long short-term memory attention network in one embodiment;

[0043] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can collect spatial transcriptome data at single cell or subcellular resolution, and convert the spatial transcriptome data to obtain the gene expression matrix corresponding to each cell; the computer device 110 can determine the spatial proximity between each cell based on the spatial transcriptome data, and judge whether there is interaction between cells based on the spatial proximity to obtain a judgment result; the computer device 110 can use each cell as a vertex of a graph structure. If the judgment result is that there is interaction between cells, the edges of the graph structure are constructed for the two cells that have interaction according to the spatial connection network, and the matrix describing the cells and the edges is used as the adjacency matrix, and the gene expression matrix of each cell is used as the graph node feature of the graph structure to obtain a node feature matrix, and construct a cell graph; the computer device 110 can use a graph convolution and a long short-term memory attention network to combine to form a feature extraction network, and extract features from the cell graph through the feature extraction network to obtain a potential feature representation; the computer device 110 can use an inner product to decode the potential feature representation, generate a new adjacency matrix, and obtain the reconstructed cell-to-cell interaction. The computer device 110 may be, but is not limited to, various personal computers, laptop computers, smart phones, robots, unmanned aerial vehicle devices, etc.

[0046] In one embodiment, Figure 2 As shown, a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network is provided, comprising the following steps:

[0047] Step 202 , collect spatial transcriptome data at single-cell or sub-cellular resolution, and convert the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell.

[0048] The computer device can collect and acquire data. Specifically, in this embodiment, the computer device can collect spatial transcriptome data at single cell or sub-cellular resolution.

[0049] In one embodiment, a method for reconstructing cell-to-cell interactions based on graph convolution and long short-term memory attention network can also include a data processing process, the specific process including: searching for fuzzy defined cells in the spatial transcriptome data, and removing the fuzzy defined cells from the spatial transcriptome data to obtain updated spatial transcriptome data; converting based on the updated spatial transcriptome data to obtain a gene expression matrix corresponding to each cell.

[0050] In one embodiment, the computer device may perform a logarithmic change normalization operation on each gene expression matrix.

[0051] Specifically, in this embodiment, the computer device can collect spatial transcriptome datasets at single-cell or sub-cellular resolution, remove ambiguously defined cells, and perform logarithmic change normalization operations on the gene expression matrix.

[0052] Step 204, determining the spatial proximity between each cell according to the spatial transcriptome data, and judging whether there is interaction between the cells based on the spatial proximity to obtain a judgment result.

[0053] The computer equipment can determine whether cells are interacting with each other based on their spatial proximity.

[0054] Step 206, taking each cell as a vertex of the graph structure, if the judgment result is that there is interaction between the cells, constructing the edges of the graph structure for the two cells that have interaction according to the spatial connection network, taking the matrix describing the cells and the edges as the adjacency matrix, taking the gene expression matrix of each cell as the graph node feature of the graph structure, obtaining the node feature matrix, and constructing the cell graph.

[0055] Computer equipment can regard each cell as a vertex of a graph structure. If there is interaction, there is an edge between the two cells. If there is no interaction, there is no edge between the two cells. The edges of the graph structure are constructed according to the spatial connection network. The matrix describing cells and edges is called the adjacency matrix. The gene expression matrix of each cell in the graph is regarded as the graph node feature, and together with the adjacency matrix, a cellular level graph structure, namely the cell graph, is constructed.

[0056] Specifically, in one embodiment, a method for reconstructing cell-to-cell interactions based on graph convolution and long short-term memory attention network can also include a process of constructing a cell graph, and the specific process includes: determining vertex cells that are vertices of the graph structure, and finding cells that interact with the vertex cells to construct target edges of the graph structure; using the matrix describing the vertex cells and the target edges as the target adjacency matrix, and determining the target node feature matrix of the graph structure based on the gene expression matrix corresponding to the vertex cells; and combining the vertices, target edges, target adjacency matrix, and target node feature matrix to obtain a cell graph.

[0057] Step 208, using graph convolution and long short-term memory attention network to form a feature extraction network, and extracting features from the cell graph through the feature extraction network to obtain a potential feature representation.

[0058] In one embodiment, the computer device may combine the graph convolution and the long short-term memory attention network in series to form a feature extraction network, wherein the feature extraction network may perform feature extraction and feature enhancement.

[0059] In this embodiment, the computer device can construct a graph convolution and a long short-term memory attention network to extract features from the constructed cell map. The feature extraction network used is composed of a long short-term memory attention network (module) and a graph convolution network connected in series.

[0060] In one embodiment, the long short-term memory attention network includes a long short-term memory network, an activation function, an attention mechanism, and a jump connection; the long short-term memory network is used to learn the temporal dependency of interactions between cells; the attention mechanism is used to enhance the feature representation of the network; the jump connection is used to add the output features of the long short-term memory network to the output features of the attention mechanism; and the graph convolution is used to capture the spatial correlation of the graph structure and learn the relationship between cells.

[0061] The long short-term memory attention module and the graph convolutional network are connected in series. The network framework of this part is shown in the figure below. Figure 3As shown. The feature matrix X and the adjacency matrix A are input into the feature extraction network formed by the series combination of the graph convolution and the long short-term memory attention network. They first enter the linear layer for processing, then pass through the long short-term memory attention module, and then process through the graph convolution and activation function, and perform iterative operations; finally, they pass through the long short-term memory attention module and the graph convolution, and after regularization processing, the potential features are obtained. Among them, the long short-term memory attention module is composed of a long short-term memory network, an activation function, an attention mechanism, and a jump connection. The long short-term memory network can learn the time dependence of cell-cell interactions. In order to make the network pay more attention to useful features, in this embodiment, an attention mechanism can be introduced into the long short-term memory network to enable the network to strengthen the feature weights related to cell-cell interactions and enhance the feature representation of the network.

[0062] In one embodiment, a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network is provided, which may also include a process of feature extraction through a feature extraction network. The specific process includes: inputting the cell graph into the feature extraction network, optimizing the feature matrix in the cell graph through the long short-term memory attention network in the feature extraction network, and obtaining an optimized feature matrix; inputting the optimized feature matrix together with the adjacency matrix into the graph convolution network, capturing the spatial correlation of the graph structure through the graph convolution network, and obtaining a potential feature representation based on the spatial correlation of the graph structure.

[0063] In this embodiment, the long short-term memory attention module uses a jump connection to add the output features of the attention mechanism to the output features of the long short-term memory network to avoid the gradient disappearance affecting the feature representation, thereby affecting the overall model performance. In short, the long short-term memory network with an attention mechanism can better focus on effective information and improve overall computing efficiency. The feature matrix optimized by the long short-term attention memory module is input into the graph convolutional network encoding module together with the adjacency matrix constructed in the second step to further optimize the encoding of the features.

[0064] In one embodiment, a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network can also include a process of obtaining a potential feature representation, and the specific process includes: the spatial correlation of the graph structure is processed by an activation function and added to the input feature matrix using a jump connection to capture features at different levels; the features at different levels are input again into the long short-term memory attention network and the graph convolution network for encoding optimization, and are regularized to obtain a potential feature representation.

[0065] When using the feature extraction network for feature extraction, the specific process may include: inputting the cell graph into the feature extraction network, and the input node feature matrix is ​​first optimized by the long short-term memory attention module; the optimized feature matrix and the adjacency matrix are input into the graph convolutional coding module together to capture the spatial correlation of the graph structure; and then the activation function is processed and added to the input feature matrix using jump connections to capture features at different levels, avoid network degradation, and better promote information transmission, so that information can be better disseminated and shared in the network; the feature matrix optimized by the first-stage coding layer is input again into the long short-term memory attention module and the graph convolutional network for coding optimization, and the potential feature representation is obtained after regularization processing.

[0066] In this embodiment, the long short-term memory network can learn the time dependence of cell-cell interactions, and at the same time introduce an attention mechanism to better focus on effective information and improve overall computational efficiency. The feature matrix and adjacency matrix optimized by the long short-term memory attention module are input into the graph convolutional coding module together to capture the spatial correlation of the graph structure and learn the relationship between cells. The above operation is repeated again to further enhance the weight of features related to cell-cell interactions, and finally generate potential feature representations.

[0067] Step 210 , use the inner product to decode the latent feature representation, generate a new adjacency matrix, and obtain the reconstructed cell-to-cell interactions.

[0068] Among them, the latent feature representation extracted and enhanced by the graph convolution and long short-term memory attention module integrates spatial information and temporal dependency. Then, the computer device can decode the latent feature representation using the inner product to generate a new adjacency matrix, and the cell-cell interaction reconstruction method based on the graph convolution and long short-term memory attention network obtains the reconstructed cell-cell interaction.

[0069] In one embodiment, a cell-to-cell interaction reconstruction model is created and trained according to a provided cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network, and an attention mechanism module is introduced into the long short-term memory network to make it pay more attention to the learning of effective information. At the same time, combined with the graph convolution network, the cell-to-cell interaction reconstruction model can capture the spatial correlation in the graph structure while learning the temporal pattern, thereby improving the model's ability to extract and optimize features; improving the computational efficiency and accuracy of the overall model, and at the same time improving the robustness and noise resistance of the model; and being able to help researchers understand the cell communication network and reveal the cell-to-cell signal transduction mechanism.

[0070] Specifically, in this embodiment, when evaluating the reconstruction model of cell-cell interaction, the area under the receiver operating characteristic curve (Area Under the Receiver Operating Characteristic Curve, AUROC), average precision (Average Precision, AP) and accuracy (Accuracy, ACC) are mainly used to evaluate the performance of the model. AUROC is calculated based on the receiver operating characteristic curve (Receiver Operating Characteristic Curve, ROC). The ROC curve is drawn based on the true positive rate (True Positive Rate, TPR) and the false positive rate (False Positive rate, FPR). TPR is the recall rate (Recall). AP measures the performance of the model under all possible thresholds. The precision-recall curve is the weighted average of the precision achieved at each threshold, and the average precision is the area under the precision-recall curve. The calculation formula is as follows: Among them, TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative. If there is an edge between cells, it is positive, and if there is no edge between cells, it is negative; N is the number of thresholds, and i represents the i-th threshold.

[0071] In order to verify the influence of feature extraction and enhancement of each module in the network on the performance of the cell-cell interaction reconstruction model, ablation experiments were conducted on each module composition in the seqFISH dataset of the mouse visual cortex, and the training set and test set were divided in a ratio of 7:3. The AUROC, AP, and ACC indicators were used to evaluate the ablation experiment results, and the results are shown in the following table:

[0072]

[0073]

[0074] As shown in the table above, in the performance evaluation of the cell-cell interaction reconstruction model, the existence of cell-cell interaction is positive, and the absence of interaction is negative. It is converted into a binary classification task, and the reconstructed cell-cell interaction is evaluated using AP, AUROC, and ACC evaluation indicators; the adjacency matrix obtained by the inner product decoding of the decoder is visualized to obtain the reconstructed cell-cell interaction map.

[0075] It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0076] In one embodiment, Figure 4 As shown, a cell-to-cell interaction reconstruction system based on graph convolution and long short-term memory attention network is provided, including: a data collection module 410, an interaction judgment module 420, a cell map construction module 430, a feature extraction module 440 and a cell-to-cell interaction reconstruction module 450, wherein:

[0077] A data collection module 410 is used to collect spatial transcriptome data at a single cell or sub-cellular resolution, and convert the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell;

[0078] An interaction determination module 420 is used to determine the spatial proximity between each cell according to the spatial transcriptome data, and determine whether there is interaction between the cells based on the spatial proximity to obtain a determination result;

[0079] The cell graph construction module 430 is used to use each cell as a vertex of the graph structure. If the judgment result is that there is interaction between the cells, the edges of the graph structure are constructed for the two cells that have interaction according to the spatial connection network, the matrix describing the cells and the edges is used as the adjacency matrix, the gene expression matrix of each cell is used as the graph node feature of the graph structure, the node feature matrix is ​​obtained, and the cell graph is constructed;

[0080] A feature extraction module 440 is used to combine graph convolution and long short-term memory attention network to form a feature extraction network, and extract features from the cell graph through the feature extraction network to obtain a potential feature representation;

[0081] The cell-to-cell interaction reconstruction module 450 is used to decode the latent feature representation using the inner product, generate a new adjacency matrix, and obtain the reconstructed cell-to-cell interaction.

[0082] In one embodiment, the data collection module 410 is also used to find fuzzy defined cells in the spatial transcriptome data, and remove the fuzzy defined cells from the spatial transcriptome data to obtain updated spatial transcriptome data; based on the updated spatial transcriptome data, conversion is performed to obtain the gene expression matrix corresponding to each cell.

[0083] In one embodiment, the data collection module 410 is further configured to perform a logarithmic change normalization operation on each gene expression matrix.

[0084] In one embodiment, the cell graph construction module 430 is also used to determine the vertex cells that are vertices of the graph structure, and to find cells that interact with the vertex cells to construct the target edges of the graph structure; the matrix describing the vertex cells and the target edges is used as the target adjacency matrix, and the target node feature matrix of the graph structure is determined according to the gene expression matrix corresponding to the vertex cells; the vertices, target edges, target adjacency matrix, and target node feature matrix are combined and constructed to obtain a cell graph.

[0085] In one embodiment, the feature extraction module 440 is also used to combine the graph convolution and the long short-term memory attention network in series to form a feature extraction network.

[0086] In one embodiment, the long short-term memory attention network includes a long short-term memory network, an activation function, an attention mechanism, and a jump connection; the long short-term memory network is used to learn the temporal dependency of interactions between cells; the attention mechanism is used to enhance the feature representation of the network; the jump connection is used to add the output features of the long short-term memory network to the output features of the attention mechanism; and the graph convolution is used to capture the spatial correlation of the graph structure and learn the relationship between cells.

[0087] In one embodiment, the feature extraction module 440 is also used to input the cell map into the feature extraction network, optimize the feature matrix in the cell map through the long short-term memory attention network in the feature extraction network, and obtain an optimized feature matrix; the optimized feature matrix is ​​input into the graph convolution network together with the adjacency matrix, and the spatial correlation of the graph structure is captured through the graph convolution network, and the potential feature representation is obtained according to the spatial correlation of the graph structure.

[0088] In one embodiment, the feature extraction module 440 is also used to process the spatial correlation of the graph structure through an activation function and add it to the input feature matrix using a jump connection to capture features at different levels; the features at different levels are input again into the long short-term memory attention network and the graph convolution network for encoding optimization, and are regularized to obtain potential feature representation.

[0089] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse, etc.

[0090] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0091] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network are implemented.

[0092] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of a cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network are implemented.

[0093] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0094] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0095] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network, characterized in that: The method comprises: Collecting spatial transcriptome data at single-cell or sub-cellular resolution, and converting the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell; Determining the spatial proximity between each cell according to the spatial transcriptome data, and judging whether there is interaction between the cells based on the spatial proximity to obtain a judgment result; Taking each of the cells as a vertex of a graph structure, if the judgment result is that there is interaction between the cells, constructing edges of a graph structure for the two cells that have interaction according to a spatial connection network, taking a matrix describing cells and edges as an adjacency matrix, taking the gene expression matrix of each cell as a graph node feature of the graph structure, obtaining a node feature matrix, and constructing a cell graph; Combining graph convolution and long short-term memory attention network to form a feature extraction network, and extracting features from the cell graph through the feature extraction network to obtain a potential feature representation; The inner product is used to decode the latent feature representation to generate a new adjacency matrix and obtain the reconstructed cell-to-cell interactions.

2. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 1, characterized in that: The gene expression matrix corresponding to each cell is obtained by conversion based on the spatial transcriptome data, including: Searching for fuzzy-defined cells in the spatial transcriptome data, and removing the fuzzy-defined cells from the spatial transcriptome data to obtain updated spatial transcriptome data; The updated spatial transcriptome data is converted to obtain the gene expression matrix corresponding to each cell.

3. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 1, characterized in that: After converting the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell, the method further includes: A logarithmic change normalization operation was performed on each of the gene expression matrices.

4. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 1, characterized in that: The cell map is constructed, including: Determine a vertex cell that is a vertex of a graph structure, and find a cell that interacts with the vertex cell to construct a target edge of the graph structure; Using the matrix describing the vertex cells and the target edges as the target adjacency matrix, and determining the target node feature matrix of the graph structure according to the gene expression matrix corresponding to the vertex cells; The vertices, target edges, target adjacency matrix, and target node feature matrix are combined and constructed to obtain a cell graph.

5. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 1, characterized in that: Use graph convolution and long short-term memory attention network to form a feature extraction network, including: The graph convolution and long short-term memory attention network are combined in series to form a feature extraction network.

6. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 5, characterized in that: The long short-term memory attention network includes a long short-term memory network, an activation function, an attention mechanism, and a skip connection; The long short-term memory network is used to learn the temporal dependency of the interaction between cells; the attention mechanism is used to enhance the feature representation of the network; the jump connection is used to add the output feature of the long short-term memory network to the output feature of the attention mechanism; The graph convolution is used to capture the spatial correlation of the graph structure and learn the relationship between cells.

7. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 6, characterized in that: The feature extraction network is used to extract features from the cell map to obtain a potential feature representation, including: Inputting the cell map into the feature extraction network, optimizing the feature matrix in the cell map through the long short-term memory attention network in the feature extraction network, and obtaining an optimized feature matrix; The optimized feature matrix and the adjacency matrix are input into the graph convolutional network together, the spatial correlation of the graph structure is captured by the graph convolutional network, and the potential feature representation is obtained according to the spatial correlation of the graph structure.

8. The cell-to-cell interaction reconstruction method based on graph convolution and long short-term memory attention network according to claim 6, characterized in that: The potential feature representation is obtained according to the spatial correlation of the graph structure, including: The spatial correlation of the graph structure is processed by the activation function and added to the input feature matrix using a skip connection to capture features at different levels; The features at different levels are input again into the long short-term memory attention network and the graph convolutional network for encoding optimization, and are regularized to obtain potential feature representation.

9. A cell-cell interaction reconstruction system based on graph convolution and long short-term memory attention network, characterized in that: The system comprises: A data collection module is used to collect spatial transcriptome data at single-cell or sub-cellular resolution, and convert the spatial transcriptome data to obtain a gene expression matrix corresponding to each cell; An interaction judgment module, used to determine the spatial proximity between each cell according to the spatial transcriptome data, and judge whether there is interaction between the cells based on the spatial proximity to obtain a judgment result; A cell graph construction module, used to use each of the cells as a vertex of a graph structure, and if the judgment result is that there is interaction between cells, construct edges of a graph structure for two cells that have interaction according to a spatial connection network, use a matrix describing cells and edges as an adjacency matrix, use the gene expression matrix of each cell as a graph node feature of the graph structure, obtain a node feature matrix, and construct a cell graph; A feature extraction module, used to combine graph convolution and long short-term memory attention network to form a feature extraction network, and extract features from the cell graph through the feature extraction network to obtain a potential feature representation; The cell-to-cell interaction reconstruction module is used to decode the latent feature representation using the inner product, generate a new adjacency matrix, and obtain the reconstructed cell-to-cell interaction.

10. The cell-to-cell interaction reconstruction system based on graph convolution and long short-term memory attention network according to claim 9, characterized in that: The data collection module is further used to search for fuzzy defined cells in the spatial transcriptome data, and remove the fuzzy defined cells from the spatial transcriptome data to obtain updated spatial transcriptome data; The updated spatial transcriptome data is converted to obtain the gene expression matrix corresponding to each cell.

Images