Intercellular communication analysis method and device based on graph convolutional network

Abstract

The invention discloses an intercellular communication analysis method and device based on a graph convolutional network, and the method comprises the steps: constructing an initial feature matrix of ligands and receptors according to the digital features of the ligands and the receptors; the method comprises the following steps: collecting an initial ligand-receptor interaction pair of protein, and constructing a ligand-receptor interaction adjacency matrix; the method comprises the following steps: constructing a ligand-receptor interaction prediction network model on the basis of a graph convolutional network, and performing prediction processing after training to obtain predicted ligand-receptor interaction pairs with interaction; the method comprises the following steps: identifying and filtering single-cell RNA sequencing data to obtain a ligand-receptor interaction pair meeting set requirements; cell communication strength is obtained through cell communication score calculation and three-point estimation method measurement; and according to the cell communication intensity, constructing a cell communication heat map and carrying out visual analysis. According to the system and the method, a relatively accurate cell communication analysis result can be obtained while the cost is reduced, and various visualizations are provided.

Description

Technical Field

[0001] The present invention relates to the field of intercellular communication technology, and in particular to an intercellular communication analysis method and device based on a graph convolutional network. Background Art

[0002] Cell-cell communication is essential for the development and homeostasis of multicellular organisms and plays a key role in many biological processes. For example, tumor-associated macrophages and carcinogenesis-associated fibroblasts play important roles in disease progression in the tumor microenvironment. Therefore, a comprehensive analysis of cell-cell communication during disease development and progression can improve our understanding of disease mechanisms and help identify new therapeutic strategies.

[0003] Cell-cell communication is often facilitated by interactions between different proteins, such as receptor-receptor interactions, extracellular matrix-receptor interactions, and ligand-receptor interactions. Specifically, the communication between two cells can be quantified by examining all the ligand-receptor interactions involved in regulating this process.

[0004] At present, the study of cell-cell communication by traditional experimental methods is time-consuming, labor-intensive and costly. The progress of single-cell RNA sequencing data has led to the development of various computational methods. The process of analyzing and inferring cell-cell communication by computational methods mainly includes two main steps: screening ligand-receptor interactions and measuring the strength of cell-cell communication mediated by ligand-receptor interactions. By establishing a comprehensive and reliable ligand-receptor interaction database, on this basis, ligand-receptor interaction resources can be used to measure the strength of cell-cell communication. A lot of research work is devoted to building high-quality LRI databases. CellPhoneDB uses a permutation method to calculate the ligand-receptor interaction score based on ligand and receptor expression, and uses the resulting score to evaluate the specificity of ligand-receptor interaction. iTALK first identifies highly expressed or differentially expressed genes, and then uses its ligand-receptor database to match and associate these genes, thereby detecting significant ligand-receptor interactions. In addition, CellDialog takes protein features as input, uses feature selection methods based on tree boosting and mixed effect models to reduce the dimensionality of these features, and finally uses the KTBoost algorithm to predict ligand-receptor interactions. With the development of these ligand-receptor interaction databases, many cell-cell communication inference methods have been developed using these databases. For example, CellPhoneDB identifies cell-cell communication by detecting ligand-receptor interactions that are highly enriched between cell types, and NATMI builds complex network models to evaluate cell-cell communication mediated by ligand-receptor interactions. Graph convolutional networks can capture the relationship and features between nodes and their neighbors by performing convolution operations in the graph structure, identify ligand-receptor interactions with higher accuracy, and thus build a more comprehensive and reliable ligand-receptor interaction database. Further, cell-cell communication analysis can be performed based on this ligand-receptor interaction resource.

[0005] Therefore, how to design a method based on graph convolutional networks to analyze cell-cell communication mediated by ligand-receptor interactions, while reducing costs and realizing visualization of cell-cell communication analysis, has become an urgent problem to be solved. Summary of the invention

[0006] To this end, the present invention provides a method and device for analyzing intercellular communication based on a graph convolutional network, which can obtain relatively accurate cell-cell communication analysis results while reducing costs and provide a variety of visualizations.

[0007] In order to achieve the above object, the present invention provides the following technical solution: a method for analyzing intercellular communication based on a graph convolutional network, comprising:

[0008] Download the ligand and receptor sequence information of the protein from the set database; process the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; construct the initial feature matrix of the ligand and receptor according to the digital features of the ligand and receptor; collect the initial ligand-receptor interaction pairs of the protein and construct the ligand-receptor interaction adjacency matrix;

[0009] Based on the graph convolutional network, a ligand-receptor interaction prediction network model is constructed; the ligand-receptor interaction prediction network model is trained through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; the trained ligand-receptor interaction prediction network model is used for prediction processing to obtain predicted ligand-receptor interaction pairs with interaction;

[0010] Identifying and filtering the collected initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet set requirements;

[0011] Performing cell-cell communication score calculation on the ligand-receptor interaction pairs that meet the set requirements to obtain expression threshold scores, expression product scores and cell expression scores of cell-cell communication of several ligand receptors between cell types; adding several corresponding scores of communication between cell types to obtain expression threshold results, expression product results and cell expression results of cell-cell communication; and obtaining the intensity of cell-cell communication by measuring the expression threshold results, expression product results and cell expression results of cell-cell communication using a three-point estimation method;

[0012] According to the intensity of the cell-to-cell communication, a cell-to-cell communication heat map is constructed; the most active ligand-receptor interaction pairs in the communication between two cell types are analyzed through the cell-to-cell communication heat map, and the cell-to-cell communication visualization analysis is completed.

[0013] As a preferred solution of a cell-to-cell communication analysis method based on a graph convolutional network, the ligand-receptor interaction prediction network model comprises two graph convolutional layers and three fully connected layers, and the convolutional layers are provided with skip connections;

[0014] The output expression of the convolutional layer with skip connection is:

[0015]

[0016] Where X is the initial feature matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is the activation function commonly used in deep learning.

[0017] As a preferred embodiment of a method for analyzing intercellular communication based on graph convolutional networks, the calculation formulas for the expression threshold score, expression product score and cell expression score of the cell-cell communication of the ligand receptor are:

[0018]

[0019]

[0020] In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M b are the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; N A and N B are the number of cells in cell type A and cell type B, respectively.

[0021] As a preferred embodiment of a method for analyzing intercellular communication based on graph convolutional networks, the expression threshold results, expression product results, and cell expression results of cell-to-cell communication are expressed as follows:

[0022]

[0023] Where m is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

[0024] As a preferred solution of a cell-to-cell communication analysis method based on a graph convolutional network, the expression of the cell-to-cell communication intensity is:

[0025]

[0026] In the formula, R max , R min and R med They are respectively the maximum, minimum and median values ​​of the expression threshold results of cell-cell communication, the expression product results and the standard cell-cell communication results after normalization of the cell expression results.

[0027] The present invention also provides a cell-to-cell communication analysis device based on a graph convolutional network, based on the above cell-to-cell communication analysis method based on a graph convolutional network, comprising:

[0028] The ligand and receptor initial feature matrix and ligand-receptor interaction adjacency matrix construction module is used to download the ligand and receptor sequence information of the protein from the set database; process the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; construct the ligand and receptor initial feature matrix according to the digital features of the ligand and receptor; collect the initial ligand-receptor interaction pairs of the protein and construct the ligand-receptor interaction adjacency matrix;

[0029] A ligand-receptor interaction prediction network model construction and processing module is used to construct a ligand-receptor interaction prediction network model based on a graph convolutional network; the ligand-receptor interaction prediction network model is trained through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; the trained ligand-receptor interaction prediction network model is used to perform prediction processing to obtain a predicted ligand-receptor interaction pair with interaction;

[0030] A ligand-receptor interaction pair identification and filtering module, used to identify and filter the initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs collected through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet set requirements;

[0031] A cell-to-cell communication strength acquisition module is used to calculate the cell-to-cell communication score for the ligand-receptor interaction pairs that meet the set requirements, and obtain the expression threshold score, expression product score and cell expression score of the cell-to-cell communication of several ligand receptors between cell types; add several corresponding scores of communication between cell types to obtain the expression threshold result, expression product result and cell expression result of the cell-to-cell communication; according to the expression threshold result, the expression product result and the cell expression result of the cell-to-cell communication, measure and obtain the cell-to-cell communication strength by the three-point estimation method;

[0032] The cell-cell communication heat map construction and analysis module is used to construct a cell-cell communication heat map according to the cell-cell communication intensity; analyze the most active ligand-receptor interaction pairs in the communication between two cell types through the cell-cell communication heat map, and complete the cell-cell communication visualization analysis.

[0033] As a preferred solution of an intercellular communication analysis device based on a graph convolutional network, in the ligand-receptor interaction prediction network model construction and processing module, the ligand-receptor interaction prediction network model comprises two graph convolutional layers and three fully connected layers, and the convolutional layers are provided with skip connections;

[0034] The output expression of the convolutional layer with skip connection is:

[0035]

[0036] Where X is the initial feature matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is the activation function commonly used in deep learning.

[0037] As a preferred solution of the intercellular communication analysis device based on graph convolutional network, in the cell-to-cell communication intensity acquisition module, the calculation formulas of the expression threshold score, expression product score and cell expression score of the cell-to-cell communication of the ligand receptor are respectively:

[0038]

[0039]

[0040] In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M b are the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; NA and N B are the number of cells in cell type A and cell type B, respectively.

[0041] As a preferred solution of a cell-to-cell communication analysis device based on a graph convolutional network, in the cell-to-cell communication intensity acquisition module, the expression threshold result, expression product result and cell expression result of the cell-to-cell communication are respectively:

[0042]

[0043] Where n is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

[0044] As a preferred solution of a cell-to-cell communication analysis device based on a graph convolutional network, in the cell-to-cell communication strength acquisition module, the expression of the cell-to-cell communication strength is:

[0045]

[0046] In the formula, R max , R min and R med They are respectively the maximum, minimum and median values ​​of the expression threshold results of cell-cell communication, the expression product results and the standard cell-cell communication results after normalization of the cell expression results.

[0047] The present invention has the following advantages: downloading protein ligand and receptor sequence information from a set database; processing the protein ligand and receptor sequence information through iFeature to obtain digital features of ligands and receptors; constructing ligand and receptor initial feature matrices based on the digital features of ligands and receptors; collecting initial ligand-receptor interaction pairs of proteins, and constructing a ligand-receptor interaction adjacency matrix; constructing a ligand-receptor interaction prediction network model based on a graph convolutional network; training the ligand-receptor interaction prediction network model through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; performing prediction processing through the trained ligand-receptor interaction prediction network model to obtain predicted ligand-receptor interaction pairs with interactions; collecting the ligand-receptor interaction pairs through single-cell RNA sequencing data The initial ligand-receptor interaction pair and the predicted ligand-receptor interaction pair are identified and filtered to obtain the ligand-receptor interaction pair that meets the set requirements; the ligand-receptor interaction pair that meets the set requirements is subjected to cell-cell communication score calculation to obtain the expression threshold score, expression product score and cell expression score of the cell-cell communication of several ligand receptors between cell types; several corresponding scores of communication between cell types are added to obtain the expression threshold result, expression product result and cell expression result of cell-cell communication; according to the expression threshold result, expression product result and cell expression result of cell-cell communication, the intensity of cell-cell communication is measured by the three-point estimation method; according to the intensity of cell-cell communication, a cell-cell communication heat map is constructed; the most active ligand-receptor interaction pairs in the communication between two cell types are analyzed by the cell-cell communication heat map, and the visualization analysis of cell-cell communication is completed. The present invention measures the intensity of cell-cell communication mediated by these ligand-receptor interactions through the identified high-confidence ligand-receptor interaction pairs and single-cell RNA sequencing data, and further performs cell-cell communication analysis. This paper uses graph convolutional networks for the first time, using the multi-dimensional features of ligands and receptors as initial embeddings to obtain a more complete, reliable and well-organized ligand-receptor interaction database for single-cell RNA sequencing data. This paper uses filtered ligand-receptor interaction resources and a three-point evaluation method to calculate the cell-cell communication strength, and obtains relatively accurate cell-cell communication analysis results compared to other popular tools, and provides a variety of visualizations. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0049] The structures, proportions, sizes, etc. illustrated in this specification are only used to match the contents disclosed in the specification so as to facilitate understanding and reading by persons familiar with the technology. They are not used to limit the conditions under which the present invention can be implemented, and therefore have no substantial technical significance. Any structural modification, change in proportion or adjustment of size shall still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0050] Figure 1 This is a schematic diagram of a flow chart of a method for analyzing intercellular communication based on a graph convolutional network provided in Example 1 of the present invention;

[0051] Figure 2 This is a schematic diagram of a specific implementation process of a method for analyzing intercellular communication based on a graph convolutional network provided in Example 1 of the present invention;

[0052] Figure 3 A schematic diagram of the communication intensity between cell types in human melanoma tissue in a method for analyzing intercellular communication based on a graph convolutional network provided in Example 1 of the present invention;

[0053] Figure 4 A schematic diagram of the three most active ligand-receptor pairs mediating the communication between melanoma cancer cells and other cell types in a cell-to-cell communication analysis method based on a graph convolutional network provided in Example 1 of the present invention;

[0054] Figure 5 This is a schematic diagram of the architecture of an intercellular communication analysis device based on a graph convolutional network provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0055] The following is a description of the implementation of the present invention by specific embodiments. People familiar with the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Example 1

[0057] See also Figure 1and Figure 2 Embodiment 1 of the present invention provides a method for analyzing intercellular communication based on a graph convolutional network, comprising the following steps:

[0058] S1. Download the ligand and receptor sequence information of the protein from the set database; process the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; construct the ligand and receptor initial feature matrix according to the digital features of the ligand and receptor; collect the initial ligand-receptor interaction pairs of the protein, and construct the ligand-receptor interaction adjacency matrix;

[0059] S2. Based on the graph convolutional network, a ligand-receptor interaction prediction network model is constructed; the ligand-receptor interaction prediction network model is trained by using the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; prediction processing is performed by using the trained ligand-receptor interaction prediction network model to obtain predicted ligand-receptor interaction pairs with interaction;

[0060] S3, identifying and filtering the collected initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet the set requirements;

[0061] S4, performing cell-cell communication score calculation on the ligand-receptor interaction pairs that meet the set requirements, obtaining expression threshold scores, expression product scores and cell expression scores of cell-cell communication of several ligand receptors between cell types; adding several corresponding scores of communication between cell types to obtain expression threshold results, expression product results and cell expression results of cell-cell communication; according to the expression threshold results, expression product results and cell expression results of cell-cell communication, measuring and obtaining the intensity of cell-cell communication by three-point estimation method;

[0062] S5. Construct a cell-cell communication heat map based on the intensity of the cell-cell communication; analyze the most active ligand-receptor interaction pairs in the communication between two cell types through the cell-cell communication heat map to complete the visualization analysis of the cell-cell communication.

[0063] In this embodiment, in step S1, the ligand and receptor sequence information of the protein is downloaded from the set database; the ligand and receptor sequence information of the protein is processed by iFeature to obtain the digital features of the ligand and receptor; according to the digital features of the ligand and receptor, the initial feature matrix of the ligand and receptor is constructed; the initial ligand-receptor interaction pairs of the protein are collected, and the ligand-receptor interaction adjacency matrix is ​​constructed;

[0064] Specifically, the sequence information of proteins (ligands and receptors) is downloaded from the UniProt database; then the sequence information of ligands and receptors is processed using iFeature to obtain the digital features of ligands and receptors and construct the initial feature matrix of ligands and receptors; the collected ligand-receptor interaction pairs are organized into the form of a ligand-receptor interaction matrix.

[0065] In this embodiment, in step S2, a ligand-receptor interaction prediction network model is constructed based on a graph convolutional network; the ligand-receptor interaction prediction network model is trained through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; the trained ligand-receptor interaction prediction network model is used to perform prediction processing to obtain a predicted ligand-receptor interaction pair with interaction;

[0066] Specifically, the ligand-receptor interaction prediction network model includes a network of two graph convolutional layers and three fully connected layers (all layers except the last fully connected layer use ReLU as the activation function), and skip connections are added in the convolutional layers to avoid problems such as gradient disappearance.

[0067] The initial feature matrix X of ligand and receptor and the ligand-receptor interaction adjacency matrix A are taken as the input of the network for training and prediction.

[0068] The output expression of each convolutional layer with skip connection is:

[0069]

[0070] Where X is the initial feature matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is the activation function commonly used in deep learning.

[0071] The Sigmoid activation function is used after the last fully connected layer so that the output is a number between (0, 1). When the value is greater than 0.55, we believe that the corresponding ligand-receptor interaction exists, otherwise it is believed that there is no interaction. The expression of the fully connected layer is the same as the skip connection expression, and the Sigmoid expression is:

[0072]

[0073] Where e is the base of the natural logarithm function.

[0074] In this embodiment, in step S3, the collected initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs are identified and filtered through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet the set requirements;

[0075] Specifically, the collected initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs are identified and filtered through single-cell RNA sequencing data. When the ligand or receptor in the ligand-receptor interaction pair is not expressed in a certain cell or the expression value is lower than the average level, it is considered that the ligand-receptor pair does not mediate the communication between the corresponding cells and is filtered.

[0076] In this embodiment, in step S4, the cell-cell communication score is calculated for the ligand-receptor interaction pairs that meet the set requirements to obtain the expression threshold score, expression product score and cell expression score of the cell-cell communication of several ligand receptors between the cell types; several corresponding scores of communication between cell types are added to obtain the expression threshold result, expression product result and cell expression result of the cell-cell communication; according to the expression threshold result, the expression product result and the cell expression result of the cell-cell communication, the cell-cell communication intensity is measured by the three-point estimation method;

[0077] Specifically, the calculation formulas for the expression threshold score, expression product score and cell expression score of the cell-cell communication of the ligand receptor are:

[0078]

[0079] In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M b are the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; N A and N Bare the number of cells in cell type A and cell type B, respectively.

[0080] After calculating the expression threshold score, expression product score, and cell expression score of cell-cell communication of all ligand receptors between all cell types, all corresponding scores of communication between cell types are added together to obtain the final expression threshold result, expression product result, and cell expression result. Specifically, assuming that for cell type A and cell type B, the expressions of the above three results are as follows:

[0081]

[0082] Where n is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

[0083] The three results obtained above were normalized by applying the minimum-maximum scaling method to obtain the standardized cell-cell communication results R1, R2 and R3; the final cell-cell communication strength was measured using the three-point estimation method:

[0084]

[0085] In the formula, R max , R min and R med They are the maximum, minimum and median scores in R1, R2 and R3 respectively.

[0086] In this embodiment, in step S5, a cell-cell communication heat map is constructed according to the cell-cell communication intensity; the most active ligand-receptor interaction pairs in the communication between two cell types are analyzed through the cell-cell communication heat map to complete the visualization analysis of cell-cell communication.

[0087] Specifically, a cell-cell communication heat map was constructed using the final cell-cell communication intensities obtained in step S4, and the most active ligand-receptor interaction pairs in pairwise cell type communication were analyzed.

[0088] like Figure 3 As shown, the communication intensity between cell types in human melanoma tissue is displayed, and the darker the color, the higher the communication intensity between the corresponding cell types.

[0089] like Figure 4 As shown, the three most active ligand-receptor pairs that mediate communication between melanoma cancer cells and other cell types are shown. The darker the color, the more active it is. For example, for Melanoma-T, when melanoma cancer cells act as efferent cells and T cells act as afferent cells, the three most active ligand-receptor pairs that mediate communication between the two are LGALS1_PTPRC, B2M_HLA-F, and B2M_CD3G, where the underscores represent the ligands and receptors, respectively.

[0090] In summary, the present invention downloads the ligand and receptor sequence information of the protein from a set database; processes the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; constructs the ligand and receptor initial feature matrix according to the digital features of the ligand and receptor; collects the initial ligand-receptor interaction pairs of the protein, and constructs the ligand-receptor interaction adjacency matrix; constructs a ligand-receptor interaction prediction network model based on a graph convolutional network; trains the ligand-receptor interaction prediction network model through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; performs prediction processing through the trained ligand-receptor interaction prediction network model to obtain predicted ligand-receptor interaction pairs with interaction; collects all the ligands and receptors through single-cell RNA sequencing data. The initial ligand-receptor interaction pair and the predicted ligand-receptor interaction pair are identified and filtered to obtain the ligand-receptor interaction pair that meets the set requirements; the ligand-receptor interaction pair that meets the set requirements is subjected to cell-cell communication score calculation to obtain the expression threshold score, expression product score and cell expression score of the cell-cell communication of several ligand receptors between cell types; several corresponding scores of communication between cell types are added to obtain the expression threshold result, expression product result and cell expression result of cell-cell communication; according to the expression threshold result, expression product result and cell expression result of cell-cell communication, the intensity of cell-cell communication is measured by the three-point estimation method; according to the intensity of cell-cell communication, a cell-cell communication heat map is constructed; the most active ligand-receptor interaction pairs in the communication between two cell types are analyzed by the cell-cell communication heat map, and the visualization analysis of cell-cell communication is completed. The present invention measures the intensity of cell-cell communication mediated by these ligand-receptor interactions through the identified high-confidence ligand-receptor interaction pairs and single-cell RNA sequencing data, and further performs cell-cell communication analysis. This paper uses graph convolutional networks for the first time, using the multi-dimensional features of ligands and receptors as initial embeddings to obtain a more complete, reliable and well-organized ligand-receptor interaction database for single-cell RNA sequencing data. This paper uses filtered ligand-receptor interaction resources and a three-point evaluation method to calculate the cell-cell communication strength, and obtains relatively accurate cell-cell communication analysis results compared to other popular tools, and provides a variety of visualizations.

[0091] It should be noted that the method of the embodiment of the present disclosure can be performed by a single device, such as a computer or a server. The method of the present embodiment can also be applied in a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present disclosure, and the multiple devices will interact with each other to complete the described method.

[0092] It should be noted that the above describes some embodiments of the present disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0093] Example 2

[0094] See also Figure 5 Embodiment 2 of the present invention further provides a cell-to-cell communication analysis device based on a graph convolutional network, comprising:

[0095] The ligand and receptor initial feature matrix and ligand-receptor interaction adjacency matrix construction module 001 is used to download the ligand and receptor sequence information of the protein from the set database; process the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; construct the ligand and receptor initial feature matrix according to the digital features of the ligand and receptor; collect the initial ligand-receptor interaction pairs of the protein, and construct the ligand-receptor interaction adjacency matrix;

[0096] The ligand-receptor interaction prediction network model construction and processing module 002 is used to construct a ligand-receptor interaction prediction network model based on a graph convolutional network; train the ligand-receptor interaction prediction network model through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; perform prediction processing through the trained ligand-receptor interaction prediction network model to obtain a predicted ligand-receptor interaction pair with interaction;

[0097] The ligand-receptor interaction pair identification and filtering module 003 is used to identify and filter the initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs collected through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet the set requirements;

[0098] The cell-cell communication strength acquisition module 004 is used to calculate the cell-cell communication score for the ligand-receptor interaction pairs that meet the set requirements, and obtain the expression threshold score, expression product score and cell expression score of the cell-cell communication of several ligand receptors between cell types; add several corresponding scores of communication between cell types to obtain the expression threshold result, expression product result and cell expression result of the cell-cell communication; according to the expression threshold result, the expression product result and the cell expression result of the cell-cell communication, measure and obtain the cell-cell communication strength by the three-point estimation method;

[0099] The cell-cell communication heat map construction and analysis module 005 is used to construct a cell-cell communication heat map according to the cell-cell communication intensity; analyze the most active ligand-receptor interaction pairs in the communication between two cell types through the cell-cell communication heat map, and complete the cell-cell communication visualization analysis.

[0100] In this embodiment, in the ligand-receptor interaction prediction network model construction and processing module 002, the ligand-receptor interaction prediction network model includes two graph convolution layers and three fully connected layers, and the convolution layers are provided with skip connections;

[0101] The output expression of the convolutional layer with skip connection is:

[0102]

[0103] Where X is the initial feature matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is the activation function commonly used in deep learning.

[0104] In this embodiment, in the cell-to-cell communication intensity acquisition module 004, the calculation formulas for the expression threshold score, expression product score and cell expression score of the cell-to-cell communication of the ligand receptor are respectively:

[0105]

[0106] In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M bare the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; N A and N B are the number of cells in cell type A and cell type B, respectively.

[0107] In this embodiment, in the cell-to-cell communication intensity acquisition module 004, the expression threshold result, expression product result and cell expression result of the cell-to-cell communication are respectively expressed as:

[0108]

[0109] Where m is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

[0110] In this embodiment, in the cell-to-cell communication strength acquisition module 004, the expression of the cell-to-cell communication strength is:

[0111]

[0112] In the formula, R max , R min and R med They are respectively the maximum, minimum and median values ​​of the expression threshold results of cell-cell communication, the expression product results and the standard cell-cell communication results after normalization of the cell expression results.

[0113] It should be noted that the information interaction, execution process and other contents between the modules of the above-mentioned system are based on the same concept as the method embodiment in Example 1 of the present application, and the technical effects they bring are the same as those of the method embodiment of the present application. For specific contents, please refer to the description in the method embodiment shown above in the present application, and will not be repeated here.

[0114] Example 3

[0115] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium, in which a program code for a cell-to-cell communication analysis method based on a graph convolutional network is stored. The program code includes instructions for executing embodiment 1 or any possible implementation thereof.

[0116] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)).

[0117] Example 4

[0118] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0119] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute a cell-to-cell communication analysis method based on a graph convolutional network of Example 1 or any possible implementation thereof.

[0120] Specifically, the processor can be implemented by hardware or by software. When implemented by hardware, the processor can be a logic circuit, an integrated circuit, etc.; when implemented by software, the processor can be a general-purpose processor implemented by reading software codes stored in a memory. The memory can be integrated in the processor or can be located outside the processor and exist independently.

[0121] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable systems. The computer instructions can be stored in a computer-readable storage medium, or transmitted from a computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center.

[0122] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing system, they can be concentrated on a single computing system, or distributed on a network composed of multiple computing systems, and optionally, they can be implemented by a program code executable by a computing system, so that they can be stored in a storage system and executed by the computing system, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0123] Although the present invention has been described in detail above by general description and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made on the basis of the present invention. Therefore, these modifications or improvements made on the basis of not departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for analyzing intercellular communication based on graph convolutional networks, characterized in that: include: Download protein ligand and receptor sequence information from the set database; Processing the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; constructing the initial feature matrix of the ligand and receptor according to the digital features of the ligand and receptor; Collect initial ligand-receptor interaction pairs of proteins and construct a ligand-receptor interaction adjacency matrix; Based on the graph convolutional network, a ligand-receptor interaction prediction network model is constructed; the ligand-receptor interaction prediction network model is trained through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; the trained ligand-receptor interaction prediction network model is used for prediction processing to obtain predicted ligand-receptor interaction pairs with interaction; Identifying and filtering the collected initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet set requirements; Performing cell-cell communication score calculation on the ligand-receptor interaction pairs that meet the set requirements to obtain expression threshold scores, expression product scores and cell expression scores of cell-cell communication of several ligand receptors between cell types; adding several corresponding scores of communication between cell types to obtain expression threshold results, expression product results and cell expression results of cell-cell communication; and obtaining the intensity of cell-cell communication by measuring the expression threshold results, expression product results and cell expression results of cell-cell communication using a three-point estimation method; According to the intensity of the cell-to-cell communication, a cell-to-cell communication heat map is constructed; the most active ligand-receptor interaction pairs in the communication between two cell types are analyzed through the cell-to-cell communication heat map, and the cell-to-cell communication visualization analysis is completed.

2. According to claim 1, a method for analyzing intercellular communication based on graph convolutional networks is characterized in that: The ligand-receptor interaction prediction network model comprises two graph convolution layers and three fully connected layers, and the convolution layers are provided with skip connections; The output expression of the convolutional layer with skip connection is: Where X is the initial characteristic matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is a commonly used activation function in deep learning.

3. According to claim 2, a method for analyzing intercellular communication based on graph convolutional networks is characterized in that: The calculation formulas for the expression threshold score, expression product score and cell expression score of the cell-cell communication of the ligand receptor are: In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M b are the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; N A and N B are the number of cells in cell type A and cell type B, respectively.

4. The method for analyzing intercellular communication based on graph convolutional networks according to claim 3, characterized in that: The expressions of the expression threshold result, expression product result and cell expression result of the cell-cell communication are respectively: Where n is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

5. The method for analyzing intercellular communication based on graph convolutional networks according to claim 4, characterized in that: The expression of the cell-to-cell communication intensity is: In the formula, R max , R min and R med They are respectively the maximum, minimum and median values ​​of the expression threshold results of cell-cell communication, the expression product results and the standard cell-cell communication results after normalization of the cell expression results.

6. A cell-to-cell communication analysis device based on a graph convolutional network, using a cell-to-cell communication analysis method based on a graph convolutional network according to any one of claims 1 to 5, characterized in that: include: The ligand and receptor initial feature matrix and ligand-receptor interaction adjacency matrix construction modules are used to download the ligand and receptor sequence information of proteins from the set database; Processing the ligand and receptor sequence information of the protein through iFeature to obtain the digital features of the ligand and receptor; constructing the initial feature matrix of the ligand and receptor according to the digital features of the ligand and receptor; Collect initial ligand-receptor interaction pairs of proteins and construct a ligand-receptor interaction adjacency matrix; A ligand-receptor interaction prediction network model construction and processing module is used to construct a ligand-receptor interaction prediction network model based on a graph convolutional network; the ligand-receptor interaction prediction network model is trained through the ligand and receptor initial feature matrix and the ligand-receptor interaction adjacency matrix to obtain the trained ligand-receptor interaction prediction network model; the trained ligand-receptor interaction prediction network model is used to perform prediction processing to obtain a predicted ligand-receptor interaction pair with interaction; A ligand-receptor interaction pair identification and filtering module, used to identify and filter the initial ligand-receptor interaction pairs and the predicted ligand-receptor interaction pairs collected through single-cell RNA sequencing data to obtain ligand-receptor interaction pairs that meet set requirements; A cell-to-cell communication strength acquisition module is used to calculate the cell-to-cell communication score for the ligand-receptor interaction pairs that meet the set requirements, and obtain the expression threshold score, expression product score and cell expression score of the cell-to-cell communication of several ligand receptors between cell types; add several corresponding scores of communication between cell types to obtain the expression threshold result, expression product result and cell expression result of the cell-to-cell communication; according to the expression threshold result, the expression product result and the cell expression result of the cell-to-cell communication, measure and obtain the cell-to-cell communication strength by the three-point estimation method; The cell-cell communication heat map construction and analysis module is used to construct a cell-cell communication heat map according to the cell-cell communication intensity; analyze the most active ligand-receptor interaction pairs in the communication between two cell types through the cell-cell communication heat map, and complete the cell-cell communication visualization analysis.

7. The intercellular communication analysis device based on graph convolutional network according to claim 6, characterized in that: In the ligand-receptor interaction prediction network model construction and processing module, the ligand-receptor interaction prediction network model comprises two graph convolution layers and three fully connected layers, and the convolution layers are provided with skip connections; The output expression of the convolutional layer with skip connection is: Where X is the initial characteristic matrix of ligand and receptor; A is the ligand-receptor interaction adjacency matrix; D is the degree matrix of the interaction adjacency matrix A; θ, b, b', W are all learnable parameters; ReLU is a commonly used activation function in deep learning.

8. The intercellular communication analysis device based on graph convolutional network according to claim 7, characterized in that: In the cell-to-cell communication intensity acquisition module, the calculation formulas for the expression threshold score, expression product score and cell expression score of the cell-to-cell communication of the ligand receptor are respectively: In the formula, scoring expression thresholds for cell-cell communication; score the expression products of cell-cell communication; Cell expression scores for cell-cell communication; A and B are cell types; a is ligand; b is receptor; M a,A is the average expression level of a in cell type A; M b,B is the average expression level of b in cell type B; M a and M b are the average expression levels of ligand a and receptor b in all cells respectively; σ a and σ b are the standard deviations of the expression levels of ligand a and receptor b in all cells, respectively. The result is 1 when both inequalities are satisfied, otherwise it is 0; N a,A is the number of cells whose expression value of ligand a in cell type A is greater than 0; N b,B is the number of cells whose expression value of receptor b in cell type B is greater than 0; N A and N B are the number of cells in cell type A and cell type B, respectively.

9. The intercellular communication analysis device based on graph convolutional network according to claim 8, characterized in that: In the cell-to-cell communication intensity acquisition module, the expression threshold result, expression product result and cell expression result of the cell-to-cell communication are respectively expressed as: Where n is the number of ligand-receptor interaction pairs involved in mediating cell-cell communication.

10. The intercellular communication analysis device based on graph convolutional network according to claim 9, characterized in that: In the cell-to-cell communication strength acquisition module, the expression of the cell-to-cell communication strength is: In the formula, R max , R min and R med They are respectively the maximum, minimum and median values ​​of the expression threshold results of cell-cell communication, the expression product results and the standard cell-cell communication results after normalization of the cell expression results.

Images