Methods, devices, equipment, and storage media for inferring gene regulatory networks

Abstract

This application discloses a method, apparatus, device, and storage medium for inferring gene regulatory networks, relating to the field of bioinformatics. It utilizes a matrix factorization framework provided by a bilateral self-representation model to solve for intercellular regulatory relationship matrices, enabling more accurate causal relationship inferences for intercellular regulation. The method includes: obtaining a first gene expression matrix and a second gene expression matrix; using the first gene expression matrix as input, employing different gene regulation inference algorithms to predict regulatory relationships and constructing a global gene regulation adjacency matrix; embedding the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model for a first-stage calculation to obtain the intercellular regulatory relationship matrix; and, based on a defined cell type, performing a second-stage calculation using the second gene expression matrix and the intercellular regulatory relationship matrix to obtain the gene regulatory network for the defined cell type.

Description

Technical Field

[0001] This application relates to the field of bioinformatics technology, and in particular to a method, apparatus, device, and storage medium for inferring gene regulatory networks. Background Technology

[0002] Gene regulatory networks (GRNs) play a central role in controlling cell identity, fate determination, and disease progression. Accurately inferring GRNs from transcriptome data is a core task in systems biology. Traditional methods for inferring GRNs largely rely on transcriptional sequencing techniques for large mixed cell populations, reflecting only the average regulatory signals of the cell population and failing to reveal cell type-specific regulatory heterogeneity.

[0003] With the advent of single-cell sequencing transcription technology, it has become possible to analyze gene regulatory networks at single-cell resolution. However, the inherent high sparsity and technical noise of single-cell transcription sequencing data pose a significant challenge to reliable network inference. One approach to inferring gene regulatory networks is to use correlation analysis of gene expression levels to infer gene regulatory relationships, attempting to construct gene regulatory networks at the cell type level. However, synchronous expression of two genes does not imply a direct causal relationship between them, and this method also ignores intercellular interactions and the dynamic characteristics of regulatory relationships as a result of changes in cell state. Another approach to inferring gene regulatory networks is to use deep learning to build models that capture complex nonlinear regulatory relationships between genes, thereby inferring these relationships. However, this approach heavily relies on prior knowledge of known transcribed and target gene pairs, making it difficult to discover unknown regulatory relationships and resulting in poor general applicability in scenarios lacking comprehensive annotations. In summary, existing gene regulation inference methods cannot simultaneously capture gene regulatory relationships and intercellular regulatory dynamics, making it difficult to output gene regulatory networks with high accuracy and biological interpretability. Summary of the Invention

[0004] In view of this, this application provides a method, apparatus, device and storage medium for inferring gene regulatory networks. The main purpose is to solve the problem that existing gene regulation inference methods cannot simultaneously capture gene regulatory relationships and intercellular regulatory dynamics, and it is difficult to output gene regulatory networks with high accuracy and biological explanatory power.

[0005] The first aspect provides a method for inferring gene regulatory networks, which includes:

[0006] Obtain a first gene expression matrix and a second gene expression matrix, wherein the first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data from the same tissue;

[0007] Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix.

[0008] The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to perform the first solution calculation through the bilateral self-representation model, thereby obtaining the intercellular regulatory relationship matrix;

[0009] Based on the defined cell type, the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix are embedded into a two-sided self-representation model. The two-sided self-representation model is then used to perform a second solution calculation to obtain the gene regulatory network for the defined cell type.

[0010] Further, obtaining the first gene expression matrix and the second gene expression matrix includes:

[0011] RNA molecules from the same tissue were sequenced using different sequencing dimensions to obtain batch RNA sequencing data and single-cell RNA sequencing data.

[0012] Gene name matching and quality control were performed on the batch RNA sequencing data and single-cell RNA sequencing data to obtain a first gene expression matrix and a second gene expression matrix with the same row dimensions.

[0013] Further, the step of using the first gene expression matrix as input, employing different gene regulation inference algorithms to predict regulatory relationships, and constructing a global gene regulation adjacency matrix includes:

[0014] Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict the regulatory relationship of the first gene expression matrix to obtain the global regulatory relationship between genes. The global regulatory relationship is used to characterize whether there is a regulatory edge between genes in the first gene expression matrix.

[0015] The matrix elements are assigned values ​​based on the global regulatory relationships between the genes to obtain a global gene regulation adjacency matrix. The matrix element assignment is performed on gene pairs in the global gene regulation adjacency matrix that meet the set conditions. The set conditions are that at least two gene regulation inference algorithms predict the existence of regulatory edges between genes.

[0016] Further, the embedding of the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model, and the subsequent first solution calculation using the bilateral self-representation model to obtain the intercellular regulatory relationship matrix, includes:

[0017] The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to output a first fusion representation matrix that encodes the global regulatory relationship structure and single-cell expression heterogeneity through the bilateral self-representation model;

[0018] The pseudo-inverse of the first fusion characterization matrix is ​​solved to obtain the intercellular regulatory relationship matrix.

[0019] Furthermore, before embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulatory relationship matrix into the bilateral self-representation model, and performing a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network for the given cell type, the method further includes:

[0020] Based on the defined cell type, gene expression sub-matrixes for the defined cell type are extracted from the second gene expression matrix, and inter-cell regulatory relationship sub-matrixes for the defined cell type are extracted from the inter-cell regulatory relationship matrix.

[0021] Accordingly, the gene expression sub-matrix of the specified cell type and the intercellular regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to perform a second solution calculation through the two-sided self-representation model to obtain the gene regulatory network of the specified cell type.

[0022] The step of embedding the gene expression sub-matrix of the specified cell type and the intercellular regulatory relationship sub-matrix of the specified cell type into a bilateral self-representation model, and then performing a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the specified cell type, includes:

[0023] The gene expression sub-matrix of the specified cell type and the inter-cell regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to output a second fusion representation matrix encoding single-cell expression heterogeneity and inter-cell regulatory relationship constraints through the two-sided self-representation model;

[0024] The pseudo-inverse of the second fusion characterization matrix is ​​solved to obtain the gene regulatory network for the specified cell type.

[0025] Furthermore, after embedding the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model to perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix, the method further includes:

[0026] A sparsity constraint is applied to the intercellular regulatory relationship matrix to remove elements with regulatory intensity below a set threshold, while retaining elements with strong regulatory relationships.

[0027] Further, the gene regulation network is represented as an adjacency matrix, where each element represents the regulatory strength between genes. After embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix into a bilateral self-representation model for a second solution calculation to obtain the gene regulation network for the given cell type, the method further includes:

[0028] The adjacency matrix of the gene regulatory network was normalized to eliminate differences in network size between different cell types.

[0029] Secondly, a device for inferring gene regulatory networks is provided, the device comprising:

[0030] The acquisition unit is used to acquire a first gene expression matrix and a second gene expression matrix, wherein the first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data from the same tissue;

[0031] The construction unit is used to construct a global gene regulation adjacency matrix by using the first gene expression matrix as input and different gene regulation inference algorithms to predict the regulatory relationship.

[0032] The first computing unit is used to embed the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model, so as to perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix.

[0033] The second computational unit is used to embed the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix into a two-sided self-representation model based on the set cell type, so as to perform a second solution calculation through the two-sided self-representation model to obtain the gene regulatory network of the set cell type.

[0034] Furthermore, the acquisition unit is specifically used for:

[0035] RNA molecules from the same tissue were sequenced using different sequencing dimensions to obtain batch RNA sequencing data and single-cell RNA sequencing data.

[0036] Gene name matching and quality control were performed on the batch RNA sequencing data and single-cell RNA sequencing data to obtain a first gene expression matrix and a second gene expression matrix with the same row dimensions.

[0037] Furthermore, the building unit is specifically used for:

[0038] Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict the regulatory relationship of the first gene expression matrix to obtain the global regulatory relationship between genes. The global regulatory relationship is used to characterize whether there is a regulatory edge between genes in the first gene expression matrix.

[0039] The matrix elements are assigned values ​​based on the global regulatory relationships between the genes to obtain a global gene regulation adjacency matrix. The matrix element assignment is performed on gene pairs in the global gene regulation adjacency matrix that meet the set conditions. The set conditions are that at least two gene regulation inference algorithms predict the existence of regulatory edges between genes.

[0040] Furthermore, the first computing unit is specifically used for:

[0041] The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to output a first fusion representation matrix that encodes the global regulatory relationship structure and single-cell expression heterogeneity through the bilateral self-representation model;

[0042] The pseudo-inverse of the first fusion characterization matrix is ​​solved to obtain the intercellular regulatory relationship matrix.

[0043] Furthermore, the device also includes:

[0044] The extraction unit is used to extract the gene expression sub-matrix of the set cell type from the second gene expression matrix and the sub-matrix of the intercellular regulation relationship matrix, respectively, based on the set cell type, before embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulation relationship matrix into the bilateral self-representation model to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulation network of the set cell type.

[0045] Correspondingly, the second computing unit is specifically used to embed the gene expression sub-matrix of the set cell type and the intercellular regulatory relationship sub-matrix of the set cell type into the bilateral self-representation model, so as to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the set cell type.

[0046] The second computing unit is further used for:

[0047] The gene expression sub-matrix of the specified cell type and the inter-cell regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to output a second fusion representation matrix encoding single-cell expression heterogeneity and inter-cell regulatory relationship constraints through the two-sided self-representation model;

[0048] The pseudo-inverse of the second fusion characterization matrix is ​​solved to obtain the gene regulatory network for the specified cell type.

[0049] Furthermore, the device also includes:

[0050] The elimination unit is used to apply sparsity constraints to the intercellular regulatory relationship matrix after embedding the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model to obtain the intercellular regulatory relationship matrix through the first solution calculation of the bilateral self-representation model, so as to eliminate elements in the intercellular regulatory relationship matrix whose regulatory intensity is lower than a set threshold, and retain elements with strong regulatory relationship.

[0051] Furthermore, the gene regulatory network is characterized in the form of an adjacency matrix, where each element in the adjacency matrix represents the regulatory strength between genes. The device further includes:

[0052] The normalization processing unit is used to normalize the adjacency matrix of the gene regulation network after embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulatory relationship matrix into the bilateral self-representation model to obtain the gene regulation network of the set cell type through the second solution calculation of the bilateral self-representation model, so as to eliminate the difference in network size between different cell types.

[0053] Thirdly, a storage medium is provided on which a computer program is stored, which, when executed by a processor, implements the above-described method for inferring gene regulatory networks.

[0054] Fourthly, a device for inferring gene regulatory networks is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the above-mentioned method for inferring gene regulatory networks.

[0055] By utilizing the above technical solutions, this application provides a method, apparatus, device, and storage medium for inferring gene regulation networks. Compared with current methods that use gene expression correlation and depth models to infer gene regulation models, this application obtains a first gene expression matrix and a second gene expression matrix. The first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data from the same tissue. Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix. The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to perform a first solution calculation through the bilateral self-representation model, obtaining the intercellular regulation relationship matrix. Based on a defined cell type, a second solution calculation is performed based on the second gene expression matrix and the intercellular regulation relationship matrix to obtain the gene regulation network for the defined cell type. The entire process uses the global gene regulation adjacency matrix as a prior constraint and solves the intercellular regulatory relationship matrix through the matrix factorization framework provided by the bilateral self-representation model. The prior constraint of this global regulation adjacency matrix can limit abnormal fluctuations in intercellular regulation, while the matrix factorization framework can separate the dynamic changes and homeostatic core structure of gene expression. This allows the intercellular regulatory relationship matrix to accurately capture the dynamic homeostatic mechanisms in gene expression, enabling cells to maintain the stability of the overall regulatory structure even with temporal changes or microenvironmental differences when regulating gene expression. This provides a more accurate causal relationship inference for intercellular regulation, and the corresponding solved gene regulation network has high accuracy and strong biological explanatory power.

[0056] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0057] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0058] Figure 1 This is a flowchart illustrating a method for inferring gene regulatory networks in one embodiment of this application;

[0059] Figure 2 yes Figure 1 A flowchart illustrating a specific implementation method of step 101;

[0060] Figure 3 yes Figure 1A flowchart illustrating a specific implementation method for step 102;

[0061] Figure 4 yes Figure 1 A schematic diagram of a specific implementation method for step 103;

[0062] Figure 5 This is a flowchart illustrating a method for inferring gene regulatory networks in another embodiment of this application;

[0063] Figure 6 This is a flowchart illustrating the inference process of a gene regulatory network in one embodiment of this application;

[0064] Figure 7 This is a schematic diagram of the structure of a gene regulatory network inference device in one embodiment of this application;

[0065] Figure 8 This is a schematic diagram of the device structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0066] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0067] Among related technologies, the advent of single-cell sequencing transcription technology has made it possible to analyze gene regulatory networks at single-cell resolution. However, the inherent high sparsity and technical noise of single-cell transcription sequencing data pose a significant challenge to reliable network inference. One approach to inferring gene regulatory networks is to use correlation analysis of gene expression levels to infer gene regulatory relationships, attempting to construct gene regulatory networks at the cell type level. However, synchronous expression of two genes does not imply a direct causal relationship between them, and this method also ignores intercellular interactions and the dynamic characteristics of regulatory relationships as a result of changes in cell state. Another approach to inferring gene regulatory networks is to use deep learning to build models that capture complex nonlinear regulatory relationships between genes, thereby inferring these relationships. However, this approach heavily relies on prior knowledge of known transcribed and target gene pairs, making it difficult to discover unknown regulatory relationships and resulting in poor general applicability in scenarios lacking comprehensive annotations. In summary, existing gene regulation inference methods cannot simultaneously capture gene regulatory relationships and intercellular regulatory dynamics, making it difficult to output gene regulatory networks with high accuracy and biological interpretability.

[0068] To address this problem, this embodiment provides a method for inferring gene regulatory networks, such as... Figure 1 As shown, it includes the following steps:

[0069] 101. Obtain the first gene expression matrix and the second gene expression matrix.

[0070] The first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data of the same tissue, which can be detected by batch RNA sequencing technology and is used to reflect the overall gene expression of the target tissue. The second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data of the same tissue, which can be detected by single-cell RNA sequencing technology and is used to capture the gene expression heterogeneity of individual cells within the target tissue. Considering the significant differences in cell types and gene expression profiles among different tissues, "same tissue" here can be understood as the same biological tissue origin, such as homologous physiological / pathological states and tissue functional characteristics. If the tissue origin is different, the overall tissue characteristics of the first gene expression matrix are not related to the single-cell characteristics of the second gene expression matrix, leading to heterogeneity in regulatory relationships.

[0071] Specifically, in the process of obtaining the first gene expression matrix, RNA samples from the same tissue are subjected to batch RNA sequencing to obtain RNA sequencing data. This RNA sequencing data is usually a string of disordered nucleic acid sequences. Bioinformatics analysis is required to match these sequences to the corresponding genes and calculate the expression intensity of each gene to obtain the first gene expression matrix.

[0072] Specifically, in the process of obtaining the second gene expression matrix, single-cell RNA sequencing is performed on RNA samples from the same tissue to obtain RNA sequencing data. This single-cell RNA sequencing data carries a disordered nucleic acid sequence with cell-specific barcodes, which is the same as the RNA sequencing data. This single-cell RNA sequencing data also needs to be analyzed by bioinformatics to match these sequences to the corresponding genes and calculate the expression intensity of each gene to obtain the second gene expression matrix.

[0073] 102. Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix.

[0074] In this embodiment, the gene regulation inference algorithm is a technical tool that predicts the regulatory relationships between genes based on gene expression data through statistical, machine learning, or bioinformatics models. Specific algorithms include, but are not limited to, ensemble regression tree algorithms, mutual information algorithms, and hybrid model algorithms. Different gene regulation inference algorithms belong to different inference logics and can mutually verify and complement the limitations of a single algorithm.

[0075] To accurately extract reliable intergene regulatory relationships from the first gene expression matrix, at least two complementary gene regulation inference algorithms can be used to independently predict regulatory relationships on the first gene expression matrix, outputting the prediction results of each algorithm. These prediction results represent whether a regulatory edge exists between genes. Then, the prediction results of multiple algorithms are aggregated to obtain cross-validated regulatory relationships. If at least two gene regulation inference algorithms predict the existence of a regulatory edge between genes, it indicates that the cross-validated regulatory relationship is one where a regulatory edge exists between genes. Furthermore, a global gene regulation adjacency matrix is ​​constructed based on the cross-validated regulatory relationships. In the global gene regulation adjacency matrix, if the cross-validated regulatory relationship indicates the existence of a regulatory edge between genes, the matrix element is assigned a value of 1; if the cross-validated regulatory relationship indicates the absence of a regulatory edge between genes, the matrix element is assigned a value of 0.

[0076] 103. Embed the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model, and perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix.

[0077] In this embodiment, the bilateral self-representation model includes two parallel modules: a global regulatory constraint module, which receives a global gene regulatory adjacency matrix to provide prior constraints on regulatory relationships at the tissue level; and a single-cell feature extraction module, which receives a second gene expression matrix to capture the expression heterogeneity features of individual cells. Furthermore, the output features of the two modules are fused through matrix multiplication to achieve single-cell feature representation guided by regulatory constraints, outputting a fused representation matrix. Then, the pseudo-inverse of the fused representation matrix is ​​solved, and the intercellular regulatory relationship matrix is ​​obtained through matrix multiplication. Solving the pseudo-inverse effectively handles the inverse operation of non-square matrices, avoids matrix singularity problems, and ensures the stability and reliability of the output matrix.

[0078] Correspondingly, through the dual-matrix fusion and pseudo-inverse solution process, the global regulatory constraints at the tissue level can be combined with the heterogeneity characteristics at the single-cell level, so that the obtained intercellular regulatory relationship matrix has both the reliability of global regulation and the ability to capture the specific regulatory effects between single cells, providing core intermediate products for the subsequent extraction of cell type-specific regulatory networks.

[0079] 104. Based on the set cell type, the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix are embedded into the bilateral self-representation model, so as to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the set cell type.

[0080] In this embodiment, the cell type is a pre-defined cell type with specific biological characteristics selected from the tissue sample corresponding to the second gene expression matrix based on research objectives or application requirements. The gene regulatory network of this cell type reflects the intrinsic correlation between its single-cell expression heterogeneity and intercellular regulatory relationships. Here, the gene-cell association path can be obtained by multiplying the gene expression sub-matrix of the cell type with the intercellular regulatory sub-matrix, and then the direct regulatory relationship between genes can be deduced by pseudo-inverse solving.

[0081] Specifically, based on the set cell type, the single-cell index corresponding to the set cell type is selected from the second gene expression matrix to obtain the single-cell index list of the set cell type. Then, based on the single-cell index list of the set cell type, the corresponding sub-matrices are extracted from the second gene expression matrix and the intercellular regulation relationship matrix, respectively. Through sub-matrix multiplication and pseudo-inverse operation, the mapping from intercellular regulation to inter-gene regulation can be realized to obtain the gene regulation network of the set cell type.

[0082] The gene regulation network inference method provided in this application, compared with the current method of inferring gene regulation models using gene expression correlation and depth models, obtains a first gene expression matrix and a second gene expression matrix. The first gene expression matrix is ​​a gene expression matrix from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix from single-cell RNA sequencing data from the same tissue. Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix. The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to perform a first solution calculation through the bilateral self-representation model to obtain the intercellular regulation relationship matrix. Based on the set cell type, a second solution calculation is performed according to the second gene expression matrix and the intercellular regulation relationship matrix to obtain the gene regulation network of the set cell type. The entire process uses the global gene regulation adjacency matrix as a prior constraint and solves the intercellular regulatory relationship matrix through the matrix factorization framework provided by the bilateral self-representation model. The prior constraint of this global regulation adjacency matrix can limit abnormal fluctuations in intercellular regulation, while the matrix factorization framework can separate the dynamic changes and homeostatic core structure of gene expression. This allows the intercellular regulatory relationship matrix to accurately capture the dynamic homeostatic mechanisms in gene expression, enabling cells to maintain the stability of the overall regulatory structure even with temporal changes or microenvironmental differences when regulating gene expression. This provides a more accurate causal relationship inference for intercellular regulation, and the corresponding solved gene regulation network has high accuracy and strong biological explanatory power.

[0083] In practical applications, the first and second gene expression matrices have different sequencing dimensions. Gene expression matrices obtained through different sequencing dimensions can form a bidirectional complement, effectively offsetting the inherent limitations of single-dimensional data and enabling accurate inference of cell type-specific gene regulatory networks. Specifically, for example... Figure 2 As shown, step 101 includes the following steps:

[0084] 201. RNA molecules from the same tissue were sequenced using different sequencing dimensions to obtain batch RNA sequencing data and single-cell RNA sequencing data.

[0085] 202. Gene name matching and quality control are performed on the batch RNA sequencing data and single-cell RNA sequencing data to obtain a first gene expression matrix and a second gene expression matrix with the same row dimension.

[0086] This embodiment is applicable to accurately inferring cell type-specific gene regulatory networks from batch RNA sequencing data and single-cell RNA sequencing data from the same biological tissue. In practical applications, this embodiment is applicable to a variety of biological processes that rely on cell type-specific regulation, including but not limited to hematopoietic differentiation, embryonic development, tumor microenvironment remodeling, immune cell activation, and stem cell differentiation.

[0087] In this embodiment, the core of different sequencing dimensions lies in the difference in sequencing unit granularity, including the tissue-wide dimension and the single-cell resolution dimension. By differentiating sequencing targets, gene expression signals of the entire tissue and individual cells are captured, respectively. Correspondingly, the sequencing dimension of batch RNA sequencing data is the tissue-wide dimension. Batch RNA sequencing data uses total RNA from the tissue as input, and the sequencing signal integrates the expression information of all cell types within the tissue. Specifically, the tissue-wide sequencing dimension of batch RNA sequencing can be used, with total RNA extracted from the tissue as the sample for sequencing, to obtain batch RNA sequencing data without the need for nucleic acid sequence format. Correspondingly, the sequencing dimension of single-cell RNA sequencing data is the single-cell resolution dimension. Single-cell sequencing uses a single cell as an independent sequencing unit, and the expression signal of each cell is distinguished through cell transcribing barcodes. Specifically, the single-cell resolution sequencing dimension of single-cell sequencing can be used, with single-cell RNA prepared from tissue exfoliation as the sample, and sequencing is performed using cell-specific barcode labeling technology to obtain single-cell RNA sequencing data carrying cell-specific identifiers.

[0088] Considering the inconsistency in gene sets across different sequencing dimensions, gene name matching is necessary. Without this matching, dimension compatibility cannot be achieved, leading to misalignment of gene sets. Similarly, to preserve the advantages of data from different sequencing dimensions, quality control is required. Specifically, in the process of gene name matching and quality control for batch RNA sequencing data and single-cell RNA sequencing data, the batch RNA sequencing data and single-cell RNA sequencing data are converted into batch raw expression matrices and single-cell raw expression matrices, respectively. Based on protein-coding annotation data consistent with the same version of the reference genome, the gene IDs in the batch raw expression matrices and single-cell raw expression matrices are uniformly converted into gene symbol format. Then, low-quality cells are filtered out from both the batch raw expression matrices and single-cell raw expression matrices, resulting in the first and second gene expression matrices, respectively.

[0089] In practical applications, the first gene expression matrix integrates expression signals from all cell types. Its predicted regulatory relationships reflect common regulatory patterns related to core functions, rather than specific regulation of a particular cell type. This global regulatory context ensures that subsequent extraction of single-cell-specific networks does not deviate from the core biological characteristics of the tissue. Specifically, such as... Figure 3 As shown, step 102 includes the following steps:

[0090] 301. Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict the regulatory relationship of the first gene expression matrix to obtain the global regulatory relationship between genes.

[0091] 302. Assign values ​​to matrix elements based on the global regulatory relationships between the genes to obtain the global gene regulation adjacency matrix.

[0092] In this embodiment, the global regulatory relationship is used to characterize whether there are regulatory edges between genes in the first gene expression matrix. It is understood that different gene regulation inference algorithms will produce different prediction results for the regulatory relationships in the first gene expression matrix. To avoid the limitations of using a single gene regulation inference algorithm, each gene regulation inference algorithm independently predicts the regulatory relationships in the first gene expression matrix. By aggregating the regulatory relationships predicted by different gene regulation inference algorithms, a more accurate global regulatory relationship between genes can be obtained.

[0093] Specifically, in the process of summarizing the regulatory relationships predicted by different gene regulation inference algorithms, if at least two gene regulation inference algorithms predict the existence of regulatory edges between genes, it is determined that there are regulatory edges between genes corresponding to the corresponding elements in the global gene regulation adjacency matrix; conversely, if only a single gene regulation inference algorithm predicts the existence of regulatory edges between genes, it is determined that there are no regulatory edges between genes corresponding to the corresponding elements in the global gene regulation adjacency matrix.

[0094] For example, the gene regulation inference algorithm includes algorithm A, algorithm B, and algorithm C. Algorithm A, algorithm B, and algorithm C are used to predict the regulatory relationship of the first gene expression matrix. If algorithm A and algorithm B predict that there is a regulatory edge between genes, then it is determined that there is a regulatory edge between genes. If only algorithm C predicts that there is a regulatory edge between genes, then it is determined that there is no regulatory edge between genes.

[0095] Correspondingly, in the process of assigning matrix elements based on the global regulatory relationships between genes, the matrix element assignment is performed on gene pairs in the global gene regulation adjacency matrix that meet the set conditions. The set conditions are that at least two gene regulation inference algorithms predict the existence of regulatory edges between genes. Specifically, the rows and columns of the global gene regulation network correspond to the gene sets in the first matrix expression matrix, and their row and column indices are gene symbols. The row and column order is consistent with the row order of the first gene expression matrix. For any element in the matrix, there is a regulatory relationship between genes. If the global regulatory relationship determines that there is a regulatory edge between genes, the element is assigned a value of 1; if the global regulatory relationship determines that there is no regulatory edge between genes, the element is assigned a value of 0.

[0096] Furthermore, to translate global regulatory relationships at the gene level into differential regulatory representations at the cellular level, it is necessary to integrate global regulatory constraints with single-cell heterogeneity. This means allowing global regulatory relationships to constrain the expression characteristics of single cells, and allowing single-cell expression characteristics to imbue global regulatory relationships with cell-specific characteristics. Specifically, for example... Figure 4 As shown, step 103 includes the following steps:

[0097] 401. Embed the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model, so as to output a first fusion representation matrix encoding the global regulatory relationship structure and single-cell expression heterogeneity through the bilateral self-representation model.

[0098] 402. Solve the pseudo-inverse of the first fusion characterization matrix to obtain the intercellular regulatory relationship matrix.

[0099] In this embodiment, the bilateral self-representation model is an improved unsupervised learning model based on an autoencoder framework. Its core features include two parallel input branches (a regulation constraint branch and a feature input branch) and a shared encoding layer. Through collaborative optimization of self-supervised reconstruction loss and regulation consistency loss, it achieves feature fusion of two types of heterogeneous data, outputting a fused representation matrix that simultaneously preserves global structural constraints and local heterogeneity features. Specifically, the bilateral self-representation model can be expressed using the following formula:

[0100]

[0101] in, This is the second gene expression matrix, which is the original gene expression matrix of the single-cell data. This is a global gene regulation adjacency matrix, containing shared regulatory relationships across cell types. Let be the intercellular regulatory relationship matrix to be solved.

[0102] Furthermore, the first fusion characterization matrix is ​​a matrix obtained by fusing the second gene expression matrix with the global gene regulation adjacency matrix. This matrix can non-linearly integrate the global regulatory structure constraints at the gene level with the personalized expression characteristics at the cell level. Each column corresponds to a fusion feature vector of regulation and expression in a single cell, and each row corresponds to the coding dimension of a global regulatory structure.

[0103] Specifically, the pseudo-inverse of the first fusion representation matrix can be solved using the following formula:

[0104]

[0105] in, Representation matrix The false rebellion.

[0106] It should be noted that both the second gene expression matrix and the intercellular regulatory relationships contain global data from multiple single cells, covering various cell types. If global data is used directly for subsequent analysis, the regulatory networks of different cell types will mix multiple regulatory patterns, resulting in a lack of heterogeneity in the subsequently inferred regulatory networks. Furthermore, as... Figure 5 As shown, prior to step 104, the method further includes the following steps:

[0107] 501. Based on the set cell type, extract the gene expression sub-matrix of the set cell type from the second gene expression matrix and extract the inter-cell regulatory relationship sub-matrix of the set cell type from the inter-cell regulatory relationship matrix.

[0108] Accordingly, in step 104, the gene expression sub-matrix of the specified cell type and the intercellular regulatory relationship sub-matrix of the specified cell type are embedded into a bilateral self-representation model, so as to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the specified cell type.

[0109] In this embodiment, the cell type is defined as a target cell population explicitly specified based on the clustering results of the intercellular regulatory relationship matrix and prior biological knowledge. In other words, the cell type is defined as a specific cell type jointly contained in the second gene expression matrix and the intercellular regulatory relationship matrix.

[0110] Specifically, in the process of extracting sub-matrixes, the target cell index of the set cell type can be obtained, and the target cell index can be mapped to the second gene expression matrix and the inter-cell regulatory relationship matrix respectively. The gene expression sub-matrix of the set cell type can be extracted from the second gene expression matrix, and the inter-cell regulatory relationship sub-matrix of the set cell type can be extracted from the inter-cell regulatory relationship matrix.

[0111] It is understandable that the second gene expression matrix and the intercellular regulatory relationship matrix are two types of complementary data. By extracting the gene expression sub-matrix and the intercellular regulatory relationship sub-matrix of the specified cell type, both the specific expression characteristics of the target cell and the regulatory similarity constraints within the target cell are preserved.

[0112] Similar to the first solution calculation process, in the second solution calculation, the gene expression sub-matrix and the inter-cell regulatory relationship sub-matrix of the specified cell type can be embedded into a two-sided self-representation model. This two-sided self-representation model outputs a second fusion representation matrix encoding single-cell expression heterogeneity and inter-cell regulatory constraints. The pseudo-inverse of the second fusion representation matrix is ​​then performed to obtain the gene regulatory network of the specified cell type. The difference is that the second fusion representation matrix is ​​a matrix obtained by fusing the gene expression sub-matrix and the inter-cell regulatory relationship sub-matrix of the specified cell type. This matrix can non-linearly integrate the inter-cell regulatory constraints of cell types with the personalized expression characteristics of cells. Each column corresponds to the fusion feature vector of regulation and expression of a single cell in the specified cell type, and each row corresponds to the core encoding dimension of the global regulatory structure.

[0113] Accordingly, the pseudo-inverse of the second fusion representation matrix can be obtained using the following formula:

[0114]

[0115] in, To define the gene expression matrix for cell types, To define the intercellular regulatory relationships submatrix for cell types, To define the gene regulatory network for each cell type, this network is represented as an adjacency matrix, where each element represents the regulatory strength between genes. For example, element... Indicates gene On genes The intensity of regulation.

[0116] Understandably, the elements of the intercellular regulatory relationship matrix are obtained based on the pseudo-inverse solution of the fusion representation matrix. This process inevitably contains low-intensity elements with random or weak biological associations, such as spurious regulatory similarities caused by single-cell expression noise. These elements have no actual biological significance, and retaining them would dilute the core regulatory signal, leading to biases in subsequent cell clustering or submatrix extraction. Such elements can be extracted by setting a threshold through sparsity constraints, retaining only regulatory relationships with intensity higher than the threshold, thus reducing the interference of noise on subsequent analysis. Correspondingly, after embedding the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model for the first solution calculation to obtain the intercellular regulatory relationship matrix, sparsity constraints can be applied to the intercellular regulatory relationship matrix to remove elements with regulatory intensity lower than the set threshold, retaining only elements with strong regulatory relationships.

[0117] Furthermore, to eliminate the inconsistency in regulatory intensity between different cell type networks due to differences in cell number and fluctuations in gene expression levels, and to enable direct comparison between networks with uniform quantitative significance for regulatory intensity, the gene regulation network for a given cell type can be normalized. Here, the gene regulation network is represented as an adjacency matrix, where elements represent the regulatory intensity between genes. Accordingly, after performing a second calculation based on the second gene expression matrix and the inter-cell regulatory relationship matrix to obtain the gene regulation network for the given cell type, the adjacency matrix of the gene regulation network is normalized to eliminate differences in network size between different cell types.

[0118] Furthermore, after the adjacency matrix of the gene regulatory network is normalized, the normalized adjacency matrix can be output, visualized, and stored, enabling the reuse of cell type-specific gene regulatory networks in various scenarios.

[0119] In practical applications, the process of inferring gene regulatory networks can be found in [reference needed]. Figure 6 As shown, Figure 6 The diagram in section A illustrates the process of constructing a matrix of intercellular regulatory relationships. This process involves extracting cells from organs and tissues, and then performing batch sequencing on a portion of these cells to obtain a batch gene expression matrix. Then, the global gene regulation relationship is inferred through the batch gene expression matrix to obtain the global gene regulation adjacency matrix. The other part underwent single-cell sequencing to obtain the second gene expression matrix. Then, the global gene regulation matrix and the second gene expression matrix are combined. Construct a bilateral self-representation model and solve the pseudo-inverse to obtain the intercellular regulatory relationship matrix. . Figure 6 The diagram in section B illustrates the process of constructing a gene regulation network for a given cell type. This process first involves dimensionality reduction and clustering of cells to identify different cell types, and then using these cell types as the basis for further analysis. ( Taking this cell type as an example, we extract the intercellular regulatory relationships within that cell type, i.e., cell type Extracting the intercellular regulatory relationship sub-matrix Then from the second gene expression matrix Extracting cell types Gene expression matrix Further, cell types were constructed based on cell homeostasis characteristics. Combined with gene expression submatrix Submatrix of intercellular regulatory relationships Construct a two-sided self-representation model and solve the pseudo-inverse matrix to obtain the adjacency matrix of the gene regulatory network for cell type A. It should be noted that, Figure 6 The colors in the matrix correspond to the values ​​of different elements. The shade of the color indicates the magnitude of the value, with darker colors representing larger values.

[0120] Furthermore, as a specific implementation of the above method, embodiments of this application provide a gene regulatory network inference device, such as... Figure 7 As shown, the device includes: an acquisition unit 61, a construction unit 62, a first calculation unit 63, and a second calculation unit 64.

[0121] The acquisition unit 61 is used to acquire a first gene expression matrix and a second gene expression matrix, wherein the first gene expression matrix is ​​a gene expression matrix of batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix of single-cell RNA sequencing data from the same tissue.

[0122] The construction unit 62 is used to construct a global gene regulation adjacency matrix by using the first gene expression matrix as input, employing different gene regulation inference algorithms to predict regulatory relationships.

[0123] The first computing unit 63 is used to embed the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model, so as to perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulation relationship matrix.

[0124] The second computing unit 64 is used to embed the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix into a two-sided self-representation model based on the set cell type, so as to perform a second solution calculation through the two-sided self-representation model to obtain the gene regulatory network of the set cell type.

[0125] The gene regulation network inference device provided in this invention, compared with the current method of inferring gene regulation models using gene expression correlation and depth models, obtains a first gene expression matrix and a second gene expression matrix. The first gene expression matrix is ​​a gene expression matrix from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix from single-cell RNA sequencing data from the same tissue. Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix. The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to perform a first solution calculation through the bilateral self-representation model to obtain the intercellular regulation relationship matrix. Based on the set cell type, a second solution calculation is performed according to the second gene expression matrix and the intercellular regulation relationship matrix to obtain the gene regulation network of the set cell type. The entire process uses the global gene regulation adjacency matrix as a prior constraint and solves the intercellular regulatory relationship matrix through the matrix factorization framework provided by the bilateral self-representation model. The prior constraint of this global regulation adjacency matrix can limit abnormal fluctuations in intercellular regulation, while the matrix factorization framework can separate the dynamic changes and homeostatic core structure of gene expression. This allows the intercellular regulatory relationship matrix to accurately capture the dynamic homeostatic mechanisms in gene expression, enabling cells to maintain the stability of the overall regulatory structure even with temporal changes or microenvironmental differences when regulating gene expression. This provides a more accurate causal relationship inference for intercellular regulation, and the corresponding solved gene regulation network has high accuracy and strong biological explanatory power.

[0126] In specific application scenarios, the acquisition unit is specifically used for:

[0127] RNA molecules from the same tissue were sequenced using different sequencing dimensions to obtain batch RNA sequencing data and single-cell RNA sequencing data.

[0128] Gene name matching and quality control were performed on the batch RNA sequencing data and single-cell RNA sequencing data to obtain a first gene expression matrix and a second gene expression matrix with the same row dimensions.

[0129] In specific application scenarios, the building unit is specifically used for:

[0130] Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict the regulatory relationship of the first gene expression matrix to obtain the global regulatory relationship between genes. The global regulatory relationship is used to characterize whether there is a regulatory edge between genes in the first gene expression matrix.

[0131] The matrix elements are assigned values ​​based on the global regulatory relationships between the genes to obtain a global gene regulation adjacency matrix. The matrix element assignment is performed on gene pairs in the global gene regulation adjacency matrix that meet the set conditions. The set conditions are that at least two gene regulation inference algorithms predict the existence of regulatory edges between genes.

[0132] In specific application scenarios, the first computing unit is specifically used for:

[0133] The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to output a first fusion representation matrix that encodes the global regulatory relationship structure and single-cell expression heterogeneity through the bilateral self-representation model;

[0134] The pseudo-inverse of the first fusion characterization matrix is ​​solved to obtain the intercellular regulatory relationship matrix.

[0135] In specific application scenarios, the device further includes:

[0136] The extraction unit is used to extract the gene expression sub-matrix of the set cell type from the second gene expression matrix and the sub-matrix of the intercellular regulation relationship matrix, respectively, based on the set cell type, before embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulation relationship matrix into the bilateral self-representation model to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulation network of the set cell type.

[0137] Correspondingly, the second computing unit is specifically used to embed the gene expression sub-matrix of the set cell type and the intercellular regulatory relationship sub-matrix of the set cell type into the bilateral self-representation model, so as to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the set cell type.

[0138] The second computing unit is further used for:

[0139] The gene expression sub-matrix of the specified cell type and the inter-cell regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to output a second fusion representation matrix encoding single-cell expression heterogeneity and inter-cell regulatory relationship constraints through the two-sided self-representation model;

[0140] The pseudo-inverse of the second fusion characterization matrix is ​​solved to obtain the gene regulatory network for the specified cell type.

[0141] In specific application scenarios, the device further includes:

[0142] The elimination unit is used to apply sparsity constraints to the intercellular regulatory relationship matrix after embedding the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model to obtain the intercellular regulatory relationship matrix through the first solution calculation of the bilateral self-representation model, so as to eliminate elements in the intercellular regulatory relationship matrix whose regulatory intensity is lower than a set threshold, and retain elements with strong regulatory relationship.

[0143] In specific application scenarios, the gene regulatory network is represented in the form of an adjacency matrix, where the elements of the adjacency matrix represent the regulatory strength between genes. The device further includes:

[0144] The normalization processing unit is used to normalize the adjacency matrix of the gene regulation network after embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulatory relationship matrix into the bilateral self-representation model to obtain the gene regulation network of the set cell type through the second solution calculation of the bilateral self-representation model, so as to eliminate the difference in network size between different cell types.

[0145] Based on the above, Figures 1-5 Accordingly, this application embodiment also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method. Figures 1-5 The method for inferring the gene regulatory network shown.

[0146] Based on this understanding, the technical solution of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, or portable hard drive), and includes several instructions to cause a computer device (such as a personal computer, server, or network device) to execute the methods described in the various implementation scenarios of this application.

[0147] Based on the above, Figures 1-5 The method shown, and Figure 7 To achieve the above objectives, this application also provides a physical device for inferring gene regulatory networks, as illustrated in the virtual device embodiment. Specifically, this device can be a computer, smartphone, tablet, smartwatch, server, or network device, etc. The physical device includes a storage medium and a processor; the storage medium stores a computer program; the processor executes the computer program to achieve the above-described... Figures 1-5 The method for inferring the gene regulatory network shown.

[0148] Optionally, the physical device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0149] In an exemplary embodiment, see Figure 8 The aforementioned physical device includes a communication bus, a processor, a memory, and a communication interface. It may also include an input / output interface and a display device. The various functional units can communicate with each other via the bus. The memory stores a computer program, and the processor executes the program stored in the memory to perform the gene regulatory network inference method described in the above embodiments.

[0150] Those skilled in the art will understand that the inferred physical device structure of a gene regulatory network provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0151] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device for inferring the aforementioned gene regulatory network, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0152] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented in hardware. By applying the technical solution of this application, compared with the existing methods, this application uses a global gene regulation adjacency matrix as a prior constraint condition, and solves the intercellular regulatory relationship matrix through the matrix decomposition framework provided by the bilateral self-representation model. The prior constraint of this global regulation adjacency matrix can limit the abnormal fluctuations of intercellular regulation, while the matrix decomposition framework can separate the dynamic changes and homeostatic core structure of gene expression. This allows the intercellular regulatory relationship matrix to accurately capture the dynamic homeostatic mechanism in gene expression, so that even if there are time changes or microenvironment differences when cells regulate gene expression, they can still maintain the stability of the overall regulatory structure, providing a more accurate causal relationship inference for intercellular regulation. The corresponding solved gene regulation network has high accuracy and strong biological explanatory power.

[0153] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing this application. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or can be modified to be located in one or more apparatuses different from this embodiment. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0154] The serial numbers in this application are for descriptive purposes only and do not represent the superiority or inferiority of any particular implementation scenario. The above disclosures are merely a few specific implementation scenarios of this application; however, this application is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A method for inferring gene regulatory networks, characterized in that, include: Obtain a first gene expression matrix and a second gene expression matrix, wherein the first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data from the same tissue; Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict regulatory relationships and construct a global gene regulation adjacency matrix. The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to perform the first solution calculation through the bilateral self-representation model, thereby obtaining the intercellular regulatory relationship matrix; Based on the defined cell type, the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix are embedded into the bilateral self-representation model, so as to perform a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network for the defined cell type. The bilateral self-representation model uses the following formula: in, This is the second gene expression matrix. This is a global gene regulation adjacency matrix. The intercellular regulatory relationship matrix is ​​to be solved. In the second solution calculation, a second fusion representation matrix encoding the single-cell expression heterogeneity and intercellular regulatory association constraints is output through a bilateral self-representation model; the pseudo-inverse solution of the second fusion representation matrix is ​​performed to obtain the gene regulatory network for the specified cell type. The pseudo-inverse of the second fusion representation matrix is ​​obtained using the following formula: in, To define the gene regulatory network for cell types, To define the gene expression matrix for cell types, To define the intercellular regulatory relationships submatrix for cell types.

2. The method according to claim 1, characterized in that, The process of obtaining the first gene expression matrix and the second gene expression matrix includes: RNA molecules from the same tissue were sequenced using different sequencing dimensions to obtain batch RNA sequencing data and single-cell RNA sequencing data. Gene name matching and quality control were performed on the batch RNA sequencing data and single-cell RNA sequencing data to obtain a first gene expression matrix and a second gene expression matrix with the same row dimensions.

3. The method according to claim 1, characterized in that, The step of using the first gene expression matrix as input, employing different gene regulation inference algorithms to predict regulatory relationships, and constructing a global gene regulation adjacency matrix includes: Using the first gene expression matrix as input, different gene regulation inference algorithms are used to predict the regulatory relationship of the first gene expression matrix to obtain the global regulatory relationship between genes. The global regulatory relationship is used to characterize whether there is a regulatory edge between genes in the first gene expression matrix. The matrix elements are assigned values ​​based on the global regulatory relationships between the genes to obtain a global gene regulation adjacency matrix. The matrix element assignment is performed on gene pairs in the global gene regulation adjacency matrix that meet the set conditions. The set conditions are that at least two gene regulation inference algorithms predict the existence of regulatory edges between genes.

4. The method according to claim 1, characterized in that, The step of embedding the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model, and performing a first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix, includes: The global gene regulation adjacency matrix and the second gene expression matrix are embedded into a bilateral self-representation model to output a first fusion representation matrix that encodes the global regulatory relationship structure and single-cell expression heterogeneity through the bilateral self-representation model; The pseudo-inverse of the first fusion characterization matrix is ​​solved to obtain the intercellular regulatory relationship matrix.

5. The method according to any one of claims 1-4, characterized in that, Before embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the intercellular regulatory relationship matrix into a bilateral self-representation model, and performing a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network for the given cell type, the method further includes: Based on the defined cell type, gene expression sub-matrixes for the defined cell type are extracted from the second gene expression matrix, and inter-cell regulatory relationship sub-matrixes for the defined cell type are extracted from the inter-cell regulatory relationship matrix. Accordingly, the gene expression sub-matrix of the specified cell type and the intercellular regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to perform a second solution calculation through the two-sided self-representation model to obtain the gene regulatory network of the specified cell type. The step of embedding the gene expression sub-matrix of the specified cell type and the intercellular regulatory relationship sub-matrix of the specified cell type into a bilateral self-representation model, and then performing a second solution calculation through the bilateral self-representation model to obtain the gene regulatory network of the specified cell type, includes: The gene expression sub-matrix of the specified cell type and the inter-cell regulatory relationship sub-matrix of the specified cell type are embedded into a two-sided self-representation model, so as to output a second fusion representation matrix encoding single-cell expression heterogeneity and inter-cell regulatory relationship constraints through the two-sided self-representation model; The pseudo-inverse of the second fusion characterization matrix is ​​solved to obtain the gene regulatory network for the specified cell type.

6. The method according to any one of claims 1-4, characterized in that, After embedding the global gene regulation adjacency matrix and the second gene expression matrix into a bilateral self-representation model to perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix, the method further includes: A sparsity constraint is applied to the intercellular regulatory relationship matrix to remove elements with regulatory intensity below a set threshold, while retaining elements with strong regulatory relationships.

7. The method according to any one of claims 1-4, characterized in that, The gene regulation network is represented as an adjacency matrix, where each element represents the regulatory strength between genes. After embedding the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix into a bilateral self-representation model for a second solution calculation to obtain the gene regulation network for the specified cell type, the method further includes: The adjacency matrix of the gene regulatory network is normalized to eliminate differences in network size between different cell types.

8. A device for inferring gene regulatory networks, characterized in that, include: The acquisition unit is used to acquire a first gene expression matrix and a second gene expression matrix, wherein the first gene expression matrix is ​​a gene expression matrix derived from batch RNA sequencing data from the same tissue, and the second gene expression matrix is ​​a gene expression matrix derived from single-cell RNA sequencing data from the same tissue; The construction unit is used to construct a global gene regulation adjacency matrix by using the first gene expression matrix as input and different gene regulation inference algorithms to predict the regulatory relationship. The first computing unit is used to embed the global gene regulation adjacency matrix and the second gene expression matrix into the bilateral self-representation model, so as to perform the first solution calculation through the bilateral self-representation model to obtain the intercellular regulatory relationship matrix. The second computing unit is used to embed the sub-matrix corresponding to the second gene expression matrix and the sub-matrix corresponding to the inter-cell regulatory relationship matrix into a two-sided self-representation model based on the set cell type, so as to perform a second solution calculation through the two-sided self-representation model to obtain the gene regulation network of the set cell type. The bilateral self-representation model uses the following formula: in, This is the second gene expression matrix. This is a global gene regulation adjacency matrix. The intercellular regulatory relationship matrix is ​​to be solved. In the second solution calculation, a second fusion representation matrix encoding the single-cell expression heterogeneity and intercellular regulatory association constraints is output through a bilateral self-representation model; the pseudo-inverse solution of the second fusion representation matrix is ​​performed to obtain the gene regulatory network for the specified cell type. The pseudo-inverse of the second fusion representation matrix is ​​obtained using the following formula: in, To define the gene regulatory network for cell types, To define the gene expression matrix for cell types, To define the intercellular regulatory relationships submatrix for cell types.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images