Single cell data expression profile reconstruction method based on cell-gene heterogeneous bipartite graph and related equipment

Abstract

The invention discloses a single cell data expression profile reconstruction method based on a cell-gene heterogeneous bipartite graph and related equipment, and belongs to the technical field of bioinformation.The method comprises the steps that the cell-gene heterogeneous bipartite graph is constructed, and embedding of different types of gene nodes and cell nodes is updated through a graph neural network encoder; obtaining an initial cell node representation and an initial gene node representation; constructing global representation of two types of nodes on the whole graph and local representation of nodes on a closed sub-graph based on initial cell node representation and initial gene node representation, inputting the global representation and the local representation into a noise contrast loss function, scoring through a discriminator, obtaining cell node embedding and gene node embedding, and inputting the cell node embedding and the gene node embedding into a matrix decomposition model; and after an objective function is optimized, outputting a reconstructed single cell data expression profile. The method can solve the problem that in the prior art, more original information cannot be reserved when a single cell data expression profile is reconstructed, so that the cell subpopulation cannot be accurately recognized and the cell state cannot be reflected.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a single cell data expression spectrum reconstruction method based on a cell-gene heterogeneous bipartite graph and related equipment. Background Art

[0002] In the field of life science research, single-cell RNA sequencing (scRNA-seq) technology can measure gene expression and transcriptome status at the single-cell level, providing key data support for analyzing cell heterogeneity and exploring the pathogenesis of human diseases. It has become an important tool for key scientific research such as analyzing cell heterogeneity and exploring the pathogenesis of human diseases, and has greatly promoted the development of multiple disciplines such as precision medicine and cell developmental biology.

[0003] However, the processing of scRNA-seq data faces many technical challenges. At present, the existing data processing and computational methods are mainly divided into three categories: the first method, such as MAGIC (Markov Affinity-based Graph Imputation of Cells), uses data diffusion to smooth expression values ​​between cells with similar expression profiles to interpolate missing values; the second method reconstructs the expression matrix from the latent space with the help of low-rank matrices or deep learning techniques, such as DeepImpute, scVI (single-cell Variational Inference), etc., using deep neural networks to learn expression patterns, or such as scGNN (single-cell Graph Neural Network) using graph neural networks to learn cell relationships; the third method uses probabilistic models, such as SAVER (Single-cell Analysis via Expression Recovery), which uses empirical Bayes methods and specific regression models to handle data sparsity and interpolate gene expression.

[0004] Although these methods have promoted the development of scRNA-seq data processing technology to a certain extent, there are still obvious shortcomings. In particular, in terms of data interpolation accuracy, due to the lack of a gold standard for single-cell expression matrices, existing methods have limited ability to restore the true expression spectrum when simulating missing value evaluations. For example, in cluster analysis, due to the presence of "dropout" events in scRNA-seq data, traditional clustering methods are difficult to accurately identify cell subpopulations; in trajectory analysis, although existing methods can denoise to a certain extent, their ability to comprehensively improve the quality of various types of analysis is still insufficient. Therefore, how to improve the feasibility and accuracy of data analysis while retaining the original information of single-cell data has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The object of the present invention is to provide a method and related device for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph, so as to solve the problem that the prior art cannot retain more original information when reconstructing a single-cell data expression profile, and thus cannot accurately identify cell subsets and reflect cell states.

[0006] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, a method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph includes the following steps: Construct a cell-gene heterogeneous bipartite graph based on a single-cell gene expression matrix; The cell-gene heterogeneous bipartite graph updates the embeddings of different types of gene nodes and cell nodes through a graph neural network encoder to obtain an initial cell node representation and an initial gene node representation; Based on the initial cell node representation and the initial gene node representation, construct the global representations of the two types of nodes on the entire cell-gene heterogeneous bipartite graph and the local representations of the nodes on their closed subgraphs, and input them into a noise contrastive loss function. After being scored by a discriminator, cell node embeddings and gene node embeddings are obtained. The noise contrastive loss function aims to maximize the mutual information between the local representation and the global representation; Input the cell node embeddings and gene node embeddings into a matrix factorization model, and after optimizing the objective function of the matrix factorization model, output the reconstructed single-cell data expression profile.

[0007] In some embodiments, the step of updating the embeddings of different types of gene nodes and cell nodes through a graph neural network encoder in the cell-gene heterogeneous bipartite graph specifically includes: Input the cell-gene bipartite graph into the graph neural network encoder, and aggregate the feature information of two-hop neighbors in the k layer network of the graph neural network encoder to update the embeddings of different types of gene nodes and cell nodes, and obtain an initial cell node representation and an initial gene node representation.

[0008] In some embodiments, the step of constructing the global representations of the two types of nodes on the entire cell-gene heterogeneous bipartite graph and the local representations of the nodes on their closed subgraphs based on the initial cell node representation and the initial gene node representation, and inputting them into a noise contrastive loss function. After being scored by a discriminator, cell node embeddings and gene node embeddings are obtained, specifically includes: For the edges of the cell-gene heterogeneous bipartite graph with the initial cell node representation and the initial gene node representation and their corresponding hLeap closed subgraph, allocate weights through the attention mechanism to obtain local representations, and at the same time perform an average operation on all nodes of the cell category and gene category of the cell-gene heterogeneous bipartite graph with the initial cell node representation and initial gene node representation to obtain global representations; Input the global representation and the local representation into the noise contrast loss function for discriminator scoring to obtain cell node embeddings and gene node embeddings.

[0009] In some embodiments, the step of inputting the global representation and the local representation into the noise contrast loss function for discriminator scoring to obtain cell node embeddings and gene node embeddings specifically includes: Use the cell-gene heterogeneous bipartite graph as a positive sample, and generate negative samples from the cell-gene heterogeneous bipartite graph through a perturbation function; Use the global representation and the local representation of the positive sample as a pair of samples, and use the global representation and the local representation of the negative sample as another pair of samples; After scoring the two pairs of samples through the discriminator, obtain cell node embeddings and gene node embeddings.

[0010] In some embodiments, the noise contrast loss function maximizes the mutual information between the local representation and the global representation through the JS divergence.

[0011] In some embodiments, the stochastic gradient descent algorithm is used to optimize the objective function of the matrix factorization model.

[0012] In a second aspect, a single-cell data expression profile reconstruction system based on a cell-gene heterogeneous bipartite graph includes: A cell-gene heterogeneous bipartite graph construction module for constructing a cell-gene heterogeneous bipartite graph based on a single-cell gene expression matrix; An initial embedding module for updating the embeddings of different types of gene nodes and cell nodes of the cell-gene heterogeneous bipartite graph through a graph neural network encoder to obtain an initial cell node representation and an initial gene node representation; A mutual information maximization module for constructing global representations of two types of nodes on the entire cell-gene heterogeneous bipartite graph and local representations of nodes on its closed subgraph based on the initial cell node representation and the initial gene node representation, and inputting them into the noise contrast loss function for discriminator scoring to obtain cell node embeddings and gene node embeddings, where the noise contrast loss function aims to maximize the mutual information between the local representation and the global representation; A single-cell data expression profile reconstruction module for inputting the cell node embeddings and gene node embeddings into a matrix factorization model, and after optimizing the objective function of the matrix factorization model, outputting the reconstructed single-cell data expression profile.

[0013] In a third aspect, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph are implemented.

[0014] In a fourth aspect, a computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph are implemented.

[0015] In a fifth aspect, a computer program product includes a computer program, characterized in that when the computer program is executed by a processor, the steps of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph are implemented.

[0016] Compared with the prior art, the present invention has the following beneficial effects: In the process of reconstructing the expression matrix, the present invention constructs local representation and global representation and determines the objective function, and uses the attention mechanism and noise contrast loss to enhance the mutual information between the local and global representations, so that the reconstructed single-cell data expression profile can retain more original information, and thus can more accurately identify cell subpopulations during cluster analysis.

[0017] Furthermore, the present invention uses a graph neural network to aggregate the feature information of two-hop neighbors to update the embeddings of cell and gene nodes, deeply mines the graph structure information, obtains more effective data representations, and can more accurately recover the true expression profile. Especially in the evaluation of simulated missing values, the accuracy of data imputation is significantly improved.

[0018] Furthermore, the present invention reconstructs the expression matrix based on matrix factorization and optimizes the objective function using the stochastic gradient descent algorithm, which can more accurately reflect the changes in cell states and provide reliable data support for the trajectory analysis task. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is the application flowchart of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph provided in this embodiment; Figure 2 is the RMSE result of the imputed data of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph and other comparative imputation methods in simulated datasets with different zero-expression rates provided in this embodiment; Figure 3 is the flowchart of the method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph provided by the present invention; Figure 4 This is the structural diagram of the single-cell data expression profile reconstruction system based on the cell-gene heterogeneous bipartite graph provided by this embodiment. Specific implementation manner

[0020] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the accompanying drawings. The content described is an explanation of the present invention rather than a limitation.

[0021] It should be noted that the terms "comprising" and "having" and any variations thereof in the description and claims of the present invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, systems, products or devices.

[0022] As Figure 1 and Figure 3 shown, this embodiment provides a single-cell data expression profile reconstruction method based on a cell-gene heterogeneous bipartite graph, including the following steps: S1. Construct a cell-gene heterogeneous bipartite graph based on the single-cell gene expression matrix; Specifically, to construct a cell-gene heterogeneous bipartite graph, starting from the single-cell gene expression matrix, construct a heterogeneous bipartite graph . Among them, represents the cell set, represents the gene set, is the edge set. Through the adjacency matrix , describe the interaction between nodes. If gene j is expressed in cell i , then , otherwise it is . This step lays the foundation for subsequent analysis and intuitively presents the association between cells and genes.

[0023] S2. The cell-gene heterogeneous bipartite graph updates the embeddings of different types of gene nodes and cell nodes through a graph neural network encoder to obtain the initial cell node representation and the initial gene node representation; Specifically, use a graph neural network encoder to learn the initial embedding. Based on the principle of the Graph Neural Network (GNN), build a graph encoder. In the k layer network of the encoder, update the embeddings of gene nodes and cell nodes of different node types by aggregating the feature information of second-order neighbors. For example, when deriving the cell node embedding , first with the help of Obtain , and then obtain from this , and finally obtain ; Similarly, the gene node embedding can be deduced. This process fully exploits the graph structure information and provides an effective data representation for subsequent analysis. Taking the deduction of the cell node embedding as an example, the calculation process is as follows: S2.1, According to the formula , calculate , where is the Relu activation function, is the weight matrix, is the mean operation, is k the embedding representation of the i -th cell in the is node in the set, represents the first-order neighbor of the gene node ; S2.2, Use to obtain according to the formula to get , where is the Relu activation function, is the weight matrix, is the mean operation, is the embedding representation of the gene k in the -th layer obtained in S2.1, which can be regarded as having certain cell characteristics, is node in the set, represents the first-order neighbor of the cell node ; S2.3, Through obtain , where is the weight matrix, [] is the concatenation operation.

[0024] S2.4, Similarly, in the process of deducing the gene node embedding , first calculate according to the formula , and the formula is , where is the weight matrix, is the Relu activation function, is k the embedding representation of the n -th gene in the -1 layer, is Denote cell nodes 's first-order neighbors; then through the formula obtain , where is the weight matrix, is the embedding representation of the cell k in the layer obtained in the previous step, which can be regarded as having certain gene characteristics, is a node in the set, denotes the first-order neighbors of the gene node ; finally, through the formula derive the gene node embedding , where is the weight matrix, and [] is the concatenation operation. The above calculation steps make full use of the characteristics of GNN to deeply mine the information in the graph structure and provide a powerful data representation for subsequent analysis.

[0025] S3. Based on the initial cell node representation and the initial gene node representation, construct the global representations of the two types of nodes on the entire graph and the local representations of the nodes on the closed subgraph, and input them into the noise contrast loss function for scoring by the discriminator to obtain the cell node embedding and the gene node embedding. The noise contrast loss function aims to maximize the mutual information between the local representation and the global representation; Specifically, for a given edge and its corresponding hop closed subgraph , use the attention mechanism to assign weights to obtain the final local representation . At the same time, by averaging all the nodes within the cell and gene categories, generate two prototype representations, and then obtain the global representation by concatenating the two prototype representations. Convert the objective function of maximizing the mutual information between the local representation and the global representation into a noise contrast loss, and use the discriminator to score the local representation and the global representation, and minimize the loss function by adjusting the parameters to enhance the mutual information between the local representation and the global representation; The local representation is constructed around the given edge and the corresponding hop closed subgraph . The hop closed subgraph refers to , where represents the neighborhood of the given cell node c , represents the neighborhood of the given gene node , is the distance function, Take odd numbers to fit the heterogeneous bipartite graph structure. Using the attention mechanism, given a node c , for its neighborhood nodes , calculate the attention weights through the following formula: (1) where, and are training matrices, represents the embedding of the c -th gene node in the neighborhood i of node , represents the embedding of the given node . Calculate the similarity between node and its neighborhood node in the form of dot product, represents the exponential function, used for weight normalization.

[0026] Similarly, for node g and its neighborhood nodes , calculate the attention weights through the following formula: (2) where, represents the embedding of the g -th cell node in the neighborhood i of node , represents the embedding of the given node g . Finally, obtain the local attention representation: (3) where, COM is the concatenation operation, is the sigmoid activation function.

[0027] The global representation is obtained by taking the mean operation on all nodes of the same type. The global representation of cell type nodes is: (4) where, is the -th cell node i in the set 's embedding, is the set of all cell nodes.

[0028] Similarly, the global representation of gene type nodes can be obtained as: (5) Among them, is the embedding of the i th gene node in which is the set of all gene nodes.

[0029] Furthermore, the global representation on the entire bipartite graph is obtained e : (6) Among them, COM is the concatenation operation, and

[0030] generates a matrix A with the same dimension as the adjacency matrix S . Each element S of the matrix follows a 0-1 distribution. Using the Bernoulli distribution as a random number generator, the formula is as follows: (7) Among them, represents the disruption rate. Then, the perturbation function P is used to S disrupt A to re-obtain the negative samples: (8) Among them, represents the exclusive OR operation, represents the generated negative samples, and

[0031] represents the corresponding set of negative sample edges. (9) Among them, is the local representation, e is the global representation, is the set of negative sample edges, is the set of positive sample edges, D is a bilinear mapping discriminator: . Finally, the cell node embedding and the gene node embedding are output.

[0032] S4. Input the cell node embedding and gene node embedding into a matrix factorization model. After optimizing the objective function of the matrix factorization model, output the reconstructed single-cell data expression profile.

[0033] Specifically, the matrix factorization model incorporating the cell node embedding and gene node embedding is as follows: (10) where represents the predicted value of the input gene expression matrix in , is the node embedding of the i th cell obtained, is the node embedding of the j th gene obtained, and are the latent factors paired with the embedding. Set the objective function , and use the Stochastic Gradient Descent (SGD) algorithm to optimize it, gradually adjust the model parameters, and finally complete the reconstruction of the single-cell data expression profile; The formula of the objective function is: (11) where is the regularization parameter, represents the masking operator, and the formula is as follows: (12) In this embodiment, through the above S1 - S4, a cell-gene heterogeneous bipartite graph is constructed from the single-cell gene expression matrix, the connection relationship between cells and genes is clarified by the adjacency matrix, then a graph encoder based on the GNN principle is used to calculate and update the cell and gene node embeddings by aggregating second-order neighbor feature information, then a local-global representation is constructed, the weights of the local representation are determined by means of the attention mechanism, the global representation is obtained from the prototype representation, the objective function is transformed into a noise contrastive loss, negative samples are generated by the perturbation function, scored by the discriminator and the loss is minimized to enhance the mutual information; finally, based on the matrix factorization model, combined with specific formulas and the objective function, the parameters are optimized using the stochastic gradient descent algorithm to achieve the reconstruction of the single-cell data expression profile. As Figure 2As shown, the Splatter tool is used to generate 4 scRNA-seq simulation datasets. Each dataset consists of 500 cells and 1000 genes and is divided into four cell subpopulations. Four sets of different data (48%, 55%, 63%, 78%) are generated by setting different missing rate parameters. For the simulation data with these 4 different missing rate parameters, a comparative experiment is conducted on the gene expression matrices generated by the method provided in this embodiment, namely scHGE (Imputaion model of single-cell data based on bipartite heterogeneous graph embedding), and 5 representative imputation methods. To evaluate the accuracy of gene expression recovery by different methods, the Rooted Mean Square Error (RMSE) is used as the evaluation index. As Figure 2 can be seen from the results in

[0034] , the RMSE of the method provided in this embodiment is the smallest, indicating that it has the highest accuracy for gene expression recovery. Therefore, the reconstruction method provided in this embodiment has the following advantages:

[0035] (1) By constructing a cell-gene heterogeneous bipartite graph, the connection relationship between cells and genes is accurately described by the adjacency matrix, intuitively presenting the association between the two, and laying a solid foundation for subsequent analysis. When learning the initial embedding, the graph neural network (GNN) is used to aggregate the feature information of two-hop neighbors to update the cell and gene node embeddings, deeply mining the graph structure information to obtain a more effective data representation. By making full use of the graph structure information, compared with traditional methods, the true expression profile can be recovered more accurately. For example, in the simulation of missing value evaluation, the existing methods have poor ability to recover the true expression profile, while in this embodiment, through a unique graph structure construction and information mining method, the accuracy of data imputation is improved.

[0036] Such as Figure 4As shown in the figure, this embodiment also provides a single-cell data expression profile reconstruction system based on a cell-gene heterogeneous bipartite graph, including: A cell-gene heterogeneous bipartite graph construction module for constructing a cell-gene heterogeneous bipartite graph based on a single-cell gene expression matrix; An initial embedding module for updating the embeddings of different types of gene nodes and cell nodes of the cell-gene heterogeneous bipartite graph through a graph neural network encoder to obtain an initial cell node representation and an initial gene node representation; A mutual information maximization module for constructing global representations of two types of nodes on the entire cell-gene heterogeneous bipartite graph and local representations of nodes on its closed subgraph based on the initial cell node representation and the initial gene node representation, and inputting them into a noise contrast loss function for scoring by a discriminator to obtain cell node embeddings and gene node embeddings, where the noise contrast loss function aims to maximize the mutual information between the local representation and the global representation; A single-cell data expression profile reconstruction module for inputting the cell node embeddings and gene node embeddings into a matrix factorization model, and after optimizing the objective function of the matrix factorization model, outputting the reconstructed single-cell data expression profile.

[0037] In the embodiments of the present invention, the division of modules is illustrative, only a logical function division. In actual implementation, there may be other division methods. In addition, in each embodiment of the present invention, each functional module may be integrated in a processor, may also exist independently physically, or two or more modules may be integrated in one module. The above integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0038] In this embodiment, a computer device is also provided. The computer device includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes a calculation component and an iteration component, and can perform model calculation and model update). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function. The processor described in the embodiment of the present invention can be used for the operation of a method for reconstructing the single-cell data expression profile based on a cell-gene heterogeneous bipartite graph.

[0039] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, one or more instructions suitable for being loaded and executed by the processor are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the method for reconstructing the single-cell data expression profile based on a cell-gene heterogeneous bipartite graph in the above embodiment.

[0040] This embodiment also provides a computer program product. The computer program product includes a computer program. When the computer program is executed by the processor, the corresponding steps of the method for reconstructing the single-cell data expression profile based on a cell-gene heterogeneous bipartite graph in the above embodiment are implemented.

[0041] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0042] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0043] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means realizes the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0044] These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable devices provide steps for realizing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 block or multiple blocks.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: still can modify the specific implementation manners of the present invention or make equivalent replacements, and any modification or equivalent replacement that does not depart from the spirit and scope of the present invention shall be covered by the protection scope of the claims of the present invention.

Claims

1. A method for reconstructing single-cell data expression profiles based on cell-gene heterogeneous bipartite graphs, characterized in that: The following steps are involved: Construct a cell-gene heterogeneous bipartite graph based on the single-cell gene expression matrix; The cell-gene heterogeneous bipartite graph updates the embedding of different types of gene nodes and cell nodes through a graph neural network encoder to obtain an initial cell node representation and an initial gene node representation; Based on the initial cell node representation and the initial gene node representation, a global representation of two types of nodes on the entire cell-gene heterogeneous bipartite graph and a local representation of nodes on their closed subgraphs are constructed, and the global representations are input into a noise contrast loss function and scored by a discriminator to obtain cell node embedding and gene node embedding. The noise contrast loss function aims to maximize the mutual information between the local representation and the global representation. The cell node embedding and gene node embedding are input into a matrix decomposition model, and after optimizing the objective function of the matrix decomposition model, the reconstructed single-cell data expression spectrum is output.

2. A method for reconstructing single-cell data expression profiles based on cell-gene heterogeneous bipartite graphs according to claim 1, characterized in that: The cell-gene heterogeneous bipartite graph updates the embedding steps of different types of gene nodes and cell nodes through a graph neural network encoder, specifically including: The cell-gene bipartite graph is input into the graph neural network encoder. k The feature information of two-hop neighbors is aggregated in the layer network to update the embedding of different types of gene nodes and cell nodes to obtain the initial cell node representation and the initial gene node representation.

3. The method for reconstructing single-cell data expression profile based on cell-gene heterogeneous bipartite graph according to claim 1, characterized in that: The steps of constructing a global representation of two types of nodes on the entire cell-gene heterogeneous bipartite graph and a local representation of nodes on its closed subgraph based on the initial cell node representation and the initial gene node representation, and inputting them into the noise contrast loss function and scoring them through the discriminator to obtain the cell node embedding and gene node embedding specifically include: For the edges of the cell-gene heterogeneous bipartite graph with the initial cell node representation and the initial gene node representation and their corresponding h Jump to the closed subgraph, assign weights through the attention mechanism to obtain a local representation, and perform an average operation on all nodes of the cell category and gene category of the cell-gene heterogeneous bipartite graph with the initial cell node representation and the initial gene node representation to obtain a global representation; The global representation and the local representation are input into the noise contrast loss function and scored by the discriminator to obtain cell node embedding and gene node embedding.

4. The method for reconstructing single-cell data expression profile based on cell-gene heterogeneous bipartite graph according to claim 3, characterized in that: The step of inputting the global representation and the local representation into the noise contrast loss function and scoring them through a discriminator to obtain cell node embedding and gene node embedding specifically includes: Using the cell-gene heterogeneous bipartite graph as a positive sample, and generating a negative sample by using a perturbation function on the cell-gene heterogeneous bipartite graph; The global representation and the local representation of the positive sample are taken as a set of sample pairs, and the global representation and the local representation of the negative sample are taken as another set of sample pairs; The two groups of sample pairs are scored by the discriminator to obtain cell node embedding and gene node embedding.

5. The method for reconstructing single-cell data expression profile based on cell-gene heterogeneous bipartite graph according to claim 1, characterized in that: The noise contrast loss function maximizes the mutual information between the local representation and the global representation through JS divergence.

6. The method for reconstructing single-cell data expression profile based on cell-gene heterogeneous bipartite graph according to claim 1, characterized in that: The stochastic gradient descent algorithm is used to optimize the objective function of the matrix decomposition model.

7. A single cell data expression profile reconstruction system based on cell-gene heterogeneous bipartite graph, characterized in that: include: Cell-gene heterogeneous bipartite graph construction module, used to construct cell-gene heterogeneous bipartite graph based on single-cell gene expression matrix; An initial embedding module, used for updating the embedding of different types of gene nodes and cell nodes in the cell-gene heterogeneous bipartite graph through a graph neural network encoder to obtain an initial cell node representation and an initial gene node representation; A mutual information maximization module is used to construct a global representation of two types of nodes on the entire cell-gene heterogeneous bipartite graph and a local representation of nodes on its closed subgraph based on the initial cell node representation and the initial gene node representation, and input them into a noise contrast loss function to obtain cell node embedding and gene node embedding after being scored by a discriminator. The noise contrast loss function aims to maximize the mutual information between the local representation and the global representation; The single-cell data expression spectrum reconstruction module is used to input the cell node embedding and the gene node embedding into the matrix decomposition model, and after optimizing the objective function of the matrix decomposition model, output the reconstructed single-cell data expression spectrum.

8. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor implements the steps of a method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that: When the computer program is executed by a processor, the steps of a method for reconstructing a single-cell data expression profile based on a cell-gene heterogeneous bipartite graph as described in any one of claims 1 to 6 are implemented.

Images