A method for identifying key regulators of tumor immune crosstalk

Abstract

The application discloses a method for identifying a key regulator of tumor immune interaction, comprising the following steps: S1, integrating single-cell sequencing, and constructing a tumor-immune cell interaction identification model of a multi-view attention network to screen a tumor-immune cell interaction pair with the greatest influence on prognosis; S2, according to the tumor-immune cell interaction pair with the greatest influence on prognosis, obtaining gene mutation, copy number variation and gene expression characteristics of the key regulator based on a cell interaction key regulator identification algorithm based on multi-omics feature fusion; S3, constructing a key regulator function exploration and regulation network; and S4, based on the key regulator function exploration and regulation network, designing an immune therapy response correlation of the key regulator of tumor immune interaction. The application integrates large-scale multi-omics data, systematically analyzes the correlation between the key regulator of tumor immune interaction and its functional module and immune response characteristics, clinical phenotypes and immune therapy response, and explores the potential of the key regulator as a prediction marker.

Description

Technical Field

[0001] This invention relates to the fields of attention mechanisms, neural networks, and bioinformatics, and more specifically, to a method for identifying key regulators of tumor immune interactions. Background Technology

[0002] The "game" between malignant tumors and the immune system runs throughout the entire process of tumor development and progression. Complex molecular changes in the tumor-immune interaction process determine the patient's response to treatment and overall survival. In recent years, tumor immunotherapy, represented by immune checkpoint blockade (ICB), has brought revolutionary breakthroughs to the treatment of malignant tumors. However, due to the heterogeneity of the tumor microenvironment caused by the dynamic changes in tumor-immune interactions, the application of immunotherapy still faces a series of challenges, such as low response rates and severe immune-related adverse reactions. Unfortunately, our current understanding of the regulatory factors of tumor-immune interactions is still far from complete, hindering our understanding of the molecular basis of immunotherapy responses and restricting the widespread application of immunotherapy. Therefore, there is an urgent need for effective computational methods and experimental analyses to participate in promoting research on the regulatory mechanisms of tumor-immune interactions.

[0003] With the advent of the biomedical big data era, the continuous accumulation of data from cancer genomics, transcriptomics, proteomics, single-cell sequencing, and clinical phenotyping has created a data foundation for characterizing tumor immune features based on multi-omics features. Theoretical models of tumor-immune interactions, such as the "Cancer-Immunity Cycle" and the "Cancer Immunogram," demonstrate that tumor-immune interactions are a complex whole involving multiple cells, multiple processes, and multiple factors, requiring comprehensive analysis of multi-omics features. Single-cell sequencing, in particular, is widely used in tumor-immune interaction research, enabling the study of molecular features at a finer cellular resolution. Furthermore, increasing evidence suggests that tumor-derived molecular alterations, including specific gene mutations, copy number variations, differential gene expression, and epigenetic changes, regulate tumor-immune interaction processes from different perspectives, leading to aggressive tumor progression or treatment resistance. Therefore, computational analysis integrating multi-omics data can more comprehensively reveal the molecular characteristics of the tumor microenvironment and help explore the regulatory mechanisms of tumor-immune interactions from different dimensions, which is essential for a deeper understanding of the molecular basis of immunotherapy responses.

[0004] The explosive growth of biomedical big data and the specific scientific questions it addresses have placed higher demands and challenges on computational methods. Machine learning algorithms, represented by deep learning, provide solid methodological support and guarantees for the efficient and accurate mining and analysis of massive multi-omics data. Professor Han Leng's team at the University of Texas constructed biomarkers for immune-related adverse events (irAEs) based on transcriptomic and genomic data from immunotherapy tumor samples. Professor Gao Lin's team at Xi'an University of Electronic Science and Technology and Professor Bo Xiaochen's team at the Academy of Military Medical Sciences systematically evaluated the efficacy of multi-omics fusion methods based on machine learning and deep learning in cancer subtyping, clustering, and clinical relevance analysis, clarifying the advantages and applicability of different models. Applying single-cell multi-omics data, Professor Wang Jianxin's team at Central South University and Professor Li Min's team developed the SSRE cell type identification algorithm based on similarity learning. These studies suggest that applying deep learning and machine learning models can effectively address the task of integrating and analyzing large-sample, complex, and high-dimensional multi-omics data. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying key regulators of tumor immune interactions, thereby overcoming the deficiencies of the prior art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for identifying key regulators of tumor immune interactions, comprising the following steps:

[0008] S1. Integrate single-cell sequencing and construct a tumor-immune cell interaction recognition model based on a multi-view attention network to screen tumor-immune cell interaction pairs that have the greatest impact on prognosis;

[0009] S2. Based on the screening of tumor-immune cell interaction pairs with the greatest impact on prognosis, the key regulators of cell interaction are identified using a multi-omics feature fusion algorithm to obtain the gene mutations, copy number variations, and gene expression characteristics of the key regulators.

[0010] S3. Constructing a key regulatory sub-function and regulatory network;

[0011] S4. Based on the exploration of key regulator functions and regulatory networks, design the immunotherapy response association of key regulators of tumor immune interaction.

[0012] Further, step S1 specifically includes:

[0013] S11. Collect single-cell transcriptome data of tumors;

[0014] S12. Perform batch correction, filtering and standardization on the single-cell transcriptome data to identify tumor cells and immune cell subsets;

[0015] S13. Collect ligand-receptor gene pairs and screen for ligand and receptor genes expressed in each cell type;

[0016] S14. Construct a cell communication network under the view of each pair of cell type interactions. The interaction strength between cells i and j is defined as:

[0017]

[0018] In the formula, This indicates the expression of the ligand gene L in sender cell i. The expression of receptor gene R in receiver cell j is represented by the summation of K and LR pairs for expression in cells i and j.

[0019] S15. Filter the edges of the cell communication network based on random perturbation experiments, and delete the edges between cells of the same type and between cells from different patients.

[0020] S16. Using cell type unique encoding or cell-specific gene expression as feature values, the feature matrix X of the node is obtained. A GNN model is constructed to obtain the embeddings of each node. The node embeddings of each view are integrated through an attention-based strategy to obtain the final embeddings (E):

[0021]

[0022] In the formula, A v W is the cell adjacency matrix of the current view. v Let B be the node weight coefficient matrix of the current view. v Let C be the error matrix. v This represents the attention weight, reflecting the contribution of each pair of cells to the interaction;

[0023] S17. Construct a fully connected network to predict the survival of cancer patients, and use the Cox loss function to measure the accuracy of prognosis prediction:

[0024]

[0025] In the formula, T i E i They represent patient x i Survival time and survival status The loss function represents the survival time predicted by the neural network model, calculated for all patients who experienced the endpoint event (E). i =1) sum of losses;

[0026] After training and optimization of the S18 and GNN models, the weights C of the multi-view attention network are used.v The survival impact of all cell interaction views was ranked, and tumor-immune cell interaction pairs with the greatest prognostic impact were screened.

[0027] Further, step S2 specifically includes:

[0028] S21. For the tumor-immune cell interaction pairs that have the greatest prognostic impact, ligand and receptor genes that are specifically highly expressed in cells are screened and sorted based on differential expression analysis and random permutation test.

[0029] S22. Construct a neural network model for predicting the gene expression of ligand and receptor genes by fusing multi-omics features;

[0030] S23. Based on the attention mechanism or SHAP method, calculate the attention weight or shapley value of different features in the input matrix, and screen the top 10 features in terms of contribution as gene mutations, copy number variations and gene expression features of key regulators of ligand or receptor genes.

[0031] Furthermore, in step S23:

[0032] Given the feature matrix X and gene expression vector q of a sample, what is the attention probability α of the i-th feature vector? i for:

[0033]

[0034] In the formula, z is the attention variable, z = i, which means the i-th feature vector is currently selected, s(x i ,q) is the attention scoring function;

[0035] The SHAP method, based on the concept of cooperative game theory, calculates the Shapley value to measure the marginal contribution of each feature to the model. It is the Shapley value of the i-th feature in set S:

[0036]

[0037] In the formula, N is the set of all features, v(S) is the value of set S, defined as the loss function value of the prediction model built based on set S, and |S| represents the size of set S.

[0038] Furthermore, step S23 specifically includes the following steps:

[0039] S31. Based on the gene information contained in the SNV site and the CNA fragment, key regulators are unified at the gene level, and functional enrichment analysis is performed on the identified key regulators.

[0040] S32. Integrate single-cell and bulk multi-omics data of tumor samples from public data resources to observe the cell expression specificity of key regulators and the association between regulators and the expression of ligand receptor genes and cell marker genes they regulate.

[0041] S33. Verify the function of regulators at the in vitro cellular level, construct a somatic mutant cell line transfection model, co-culture it with peripheral blood mononuclear cells, use Transwell migration chamber experiments to verify the effect of genomic variation on the migration ability of immune cells, and use multicolor immunohistochemical staining experiments to verify the changes in ligand receptor gene expression under the influence of genomic variation.

[0042] S34. Construct a ligand-receptor-regulatory network, integrate protein interaction information from the STRING database, TRRUST, and the prior gene function network of the GRNdb transcriptional regulatory network, and apply network module analysis and restart random walk algorithm to identify the gene cascade pathways and regulatory modules that regulators influence ligand-receptor genes.

[0043] Furthermore, step S4 specifically includes the following steps:

[0044] S41. Collect clinical phenotypic information and multi-omics data of tumor samples treated with immune checkpoint inhibitors, including bulk-level genomic, transcriptomic and single-cell-level transcriptomic data.

[0045] S42. Based on single-sample gene set enrichment analysis and AUCell algorithm, the expression activities of key regulatory functional modules in tissues and cells were calculated, and the expression activities of regulators and functional modules were calculated as related features of anti-tumor immune response process. These related features of anti-tumor immune response process include the correlation between tumor mutation burden, immune checkpoint gene expression, IFN-γ signaling, antigen presentation pathway, cell lysis activity, and immune infiltration ratio.

[0046] S43. Based on the mutation status, expression level and functional module expression activity of key regulators, the clinical characteristics of different groups were compared using the chi-square test and Wilcoxon rank-sum test, including age, gender, tumor TNM stage and immunotherapy response rate.

[0047] S44. Kaplan-Meier survival curves and log-rank tests were used to calculate the prognostic differences between groups, and univariate Cox regression and C-index methods were used to evaluate the prognostic efficacy of the regulators in immunotherapy.

[0048] S45. Calculate AUC and Precision index to assess the predictive efficacy of regulators and their functional modules for immunotherapy response and immune-related adverse reactions.

[0049] Compared with the prior art, the advantages of the present invention are as follows:

[0050] 1. This invention constructs a multi-view attention network model, treating each pair of cell communication's directed weighted graph as a single view. Based on attention weights, it ranks the prognostic influence of all cell interaction views, selecting the tumor-immune cell interaction pairs with the greatest prognostic impact. This invention not only innovatively proposes ranking the prognostic influence of cell interactions to identify the most critical cell interactions affecting patient prognosis, but also introduces a multi-view attention network, representing an innovation in computational methods.

[0051] 2. This invention designs an interpretable deep learning model that integrates multiple features to systematically identify key regulators of tumor-immune interactions from the perspectives of SNV, CNA, and gene expression. This invention will help to explore the regulatory mechanisms of tumor-immune interactions from different dimensions, and is also of great significance for a deeper understanding of the molecular basis of immunotherapy responses.

[0052] 3. This invention constructs a ligand-receptor-regulator network, which integrates protein interaction networks and transcription factor regulatory networks, thereby uncovering the gene cascade pathways and regulatory modules through which regulators influence ligand-receptor genes.

[0053] 4. This invention integrates large-scale multi-omics data to systematically analyze the association between key regulators of tumor-immune interactions and their functional modules with immune response characteristics, clinical phenotypes and immunotherapy responses, and explores their potential as predictive biomarkers. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart of the method for identifying key regulators of tumor immune interactions according to the present invention.

[0056] Figure 2 This is a flowchart of the tumor-immune cell interaction recognition model of the multi-view attention network in this invention. Detailed Implementation

[0057] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0058] See Figures 1-2As shown, this embodiment discloses a method for identifying key regulators of tumor immune interactions, including the following steps:

[0059] Step S1: Integrate single-cell sequencing and construct a tumor-immune cell interaction recognition model using a multi-view attention network to screen tumor-immune cell interaction pairs that have the greatest impact on prognosis.

[0060] Step S2: Based on the selection of tumor-immune cell interaction pairs with the greatest impact on prognosis, the key regulators of cell interaction are identified using a multi-omics feature fusion algorithm to obtain the gene mutations, copy number variations, and gene expression characteristics of the key regulators.

[0061] Step S3: Construct a key regulatory sub-function exploration and regulation network.

[0062] Step S4: Based on the exploration of key regulator functions and regulatory networks, design the immunotherapy response association of key regulators of tumor immune interaction.

[0063] Step S1 in this embodiment specifically includes:

[0064] Step S11: Construction of a tumor-immune cell interaction recognition model integrating single-cell sequencing and multi-view attention network: Integrate databases such as scTIME, TISCH, CancerSCEM, and GEO to collect single-cell transcriptome (RNA-seq) data and clinical phenotype information from cancers such as melanoma, non-small cell lung cancer, breast cancer, and kidney cancer.

[0065] Step S12: Batch correct, filter, and standardize the single-cell transcriptome data to identify tumor cell and immune cell subsets.

[0066] Step S13: Integrate ligand-receptor gene pairs from databases such as CellPhoneDB, CellChatDB, and connectomeDB2020, and screen for ligand and receptor genes expressed in each cell type.

[0067] Step S14: Construct a cell communication network under the view of each pair of cell type interactions. The interaction strength between cells i and j is defined as:

[0068]

[0069] In the formula, This indicates the expression of the ligand gene L in sender cell i. The expression of receptor gene R in receiver cell j is represented by the summation of K and LR pairs for expression in cells i and j.

[0070] Step S15: Filter the edges of the cell communication network based on random perturbation experiments, and delete edges between cells of the same type and between cells from different patients.

[0071] Step S16: Using cell type one-hot encoding or cell-specific gene expression as feature values, obtain the feature matrix X of the node. Construct a GNN model to obtain the embeddings of each node. The node embeddings of each view are integrated through an attention-based strategy to obtain the final embeddings (E):

[0072]

[0073] In the formula, A v W is the adjacency matrix of the cells in the current view. v Let B be the node weight coefficient matrix of the current view. v Let C be the error matrix. v This represents the attention weight, reflecting the contribution of each pair of cells to the interaction.

[0074] Step S17: Construct a fully connected network to predict the survival of cancer patients, and use the Cox loss function to measure the accuracy of the prognosis prediction.

[0075]

[0076] In the formula, T i E i They represent patient x i Survival time and survival status The loss function represents the survival time predicted by the neural network model, calculated for all patients who experienced the endpoint event (E). i =1) sum of losses;

[0077] Step S18: After training and optimizing the GNN model, based on the weights C of the multi-view attention network... v The survival impact of all cell interaction views was ranked, and tumor-immune cell interaction pairs with the greatest prognostic impact were screened.

[0078] Step S2 in this embodiment specifically includes:

[0079] Step S21: For the tumor-immune cell interaction pairs that have the greatest prognostic impact, screen and sort the ligand and receptor genes that are specifically highly expressed in the cells based on differential expression analysis and random permutation test.

[0080] Step S22: Construct a neural network model that integrates multiple omics features such as SNV spectrum, CNA spectrum, and gene expression spectrum to predict the gene expression of ligand and receptor genes.

[0081] Step S23: Based on the attention mechanism or SHAP method model, calculate the attention weight or shapley value of different features in the input matrix, and screen the top 10 features in terms of contribution as key molecules that have a significant regulatory effect on ligand or receptor genes, namely gene mutations, copy number variations and gene expression characteristics of key regulators.

[0082] In step S23 of this embodiment:

[0083] Given the feature matrix X and gene expression vector q of a sample, what is the attention probability α of the i-th feature vector? i for:

[0084]

[0085] In the formula, z is the attention variable, z = i, which means the i-th feature vector is currently selected, s(x i ,q) is the attention scoring function. Common attention scoring functions include additive models, dot product models, scaling point models, and bilinear models.

[0086] On the other hand, the SHAP method, based on the idea of ​​cooperative game theory, calculates the Shapley value to measure the marginal contribution of each feature to the model. It is the Shapley value of the i-th feature in set S:

[0087]

[0088] In the formula, N is the set of all features, v(S) is the value of set S, which in this embodiment is defined as the loss function value of the prediction model built based on set S, and |S| represents the size of set S.

[0089] Step S23 in this embodiment specifically includes the following steps:

[0090] Step S31: Based on the gene information contained in the SNV site and the CNA fragment, key regulators are unified at the gene level. Functional enrichment analysis is performed on the identified key regulators, including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis, to observe their participation in cell communication-related biological processes.

[0091] Step S32: Integrate single-cell and bulk multi-omics data of tumor samples from public data resources such as TCGA, GEO, cBioPortal, and published studies to observe the cell expression specificity of key regulators and the association between regulators and the expression of ligand receptor genes and cell marker genes they regulate.

[0092] Step S33: In collaboration with the wet experiment research team, the function of regulators was verified at the in vitro cellular level. A somatic mutant cell line transfection model was constructed and co-cultured with peripheral blood mononuclear cells. The Transwell migration chamber experiment was used to verify the effect of genomic variation on the migration ability of immune cells. Multicolor immunohistochemical staining (mIHC) was used to verify the changes in ligand receptor gene expression under the influence of genomic variation.

[0093] Step S34: Construct a ligand-receptor-regulatory network, integrate protein interaction information from the STRING database, TRRUST, and the prior gene function network of the GRNdb transcriptional regulatory network, and apply network module analysis and restart random walk algorithm to identify the gene cascade pathways and regulatory modules that regulators affect ligand-receptor genes.

[0094] Step S4 in this embodiment specifically includes the following steps:

[0095] Step S41: Collect clinical phenotypic information and multi-omics data of tumor samples treated with immune checkpoint inhibitors, including bulk-level genomic, transcriptomic, and single-cell-level transcriptomic data.

[0096] Step S42: Based on single-sample gene set enrichment analysis and the AUCell algorithm, calculate the expression activity of key regulatory functional modules in tissues and cells, respectively, and calculate the correlation between the expression activity of regulators and functional modules and the anti-tumor immune response process. The correlation between the anti-tumor immune response process and the tumor mutation burden, immune checkpoint gene expression, IFN-γ signaling, antigen presentation pathway, cell lysis activity, and immune infiltration ratio.

[0097] Step S43: Based on the mutation status, expression level, and functional module expression activity of key regulators, the clinical characteristics of different groups are compared using the chi-square test and Wilcoxon rank-sum test, including age, gender, tumor TNM stage, and immunotherapy response rate.

[0098] Step S44: Calculate the prognostic differences between groups using Kaplan-Meier survival curves and log-rank tests, and evaluate the prognostic efficacy of the regulator's immunotherapy using univariate Cox regression and C-index methods.

[0099] Step S45: Calculate AUC and Precision index to assess the predictive efficacy of regulators and their functional modules for immunotherapy response and immune-related adverse reactions.

[0100] This invention constructs a multi-view attention network model, treating each pair of cell communication's directed weighted graph as a single view. Based on attention weights, it ranks the prognostic influence of all cell interaction views, selecting the tumor-immune cell interaction pairs with the greatest prognostic impact. This invention not only innovatively proposes ranking the prognostic influence of cell interactions to identify the most critical cell interactions affecting patient prognosis, but also introduces a multi-view attention network, representing an innovation in computational methods.

[0101] This invention designs an interpretable deep learning model that integrates multiple features to systematically identify key regulators of tumor-immune interactions from the perspectives of SNVs, CNAs, and gene expression. This invention will help to uncover the regulatory mechanisms of tumor-immune interactions from different dimensions and is of great significance for a deeper understanding of the molecular basis of immunotherapy responses.

[0102] This invention constructs a ligand-receptor-regulator network, which integrates protein interaction networks and transcription factor regulatory networks, thereby uncovering the gene cascade pathways and regulatory modules through which regulators influence ligand-receptor genes.

[0103] This invention integrates large-scale multi-omics data to systematically analyze the association between key regulators of tumor-immune interactions and their functional modules with immune response characteristics, clinical phenotypes and immunotherapy responses, and explores their potential as predictive biomarkers.

[0104] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner may make various modifications or alterations within the scope of the appended claims, as long as they do not exceed the protection scope described in the claims of the present invention, they shall be within the protection scope of the present invention.

Claims

1. A method for identifying key regulators of tumor immune interactions, characterized in that, Includes the following steps: S1. Integrate single-cell sequencing and construct a tumor-immune cell interaction recognition model with a multi-view attention network to screen tumor-immune cell interaction pairs that have the greatest impact on prognosis; S2. Based on the screening of tumor-immune cell interaction pairs with the greatest impact on prognosis, the key regulators of cell interaction are identified using a multi-omics feature fusion algorithm to obtain the gene mutations, copy number variations, and gene expression characteristics of the key regulators. S3. Constructing a key regulatory sub-function and regulatory network; S4. Based on the exploration of key regulator functions and regulatory networks, design the immunotherapy response association of key regulators of tumor immune interaction; Step S1 specifically includes: S11. Collect single-cell transcriptome data of tumors; S12. Perform batch correction, filtering and standardization on the single-cell transcriptome data to identify tumor cells and immune cell subsets; S13. Collect ligand-receptor gene pairs and screen for ligand and receptor genes expressed in each cell type; S14. Construct a cell communication network under the view of each pair of cell type interactions. The interaction strength between cells i and j is defined as: In the formula, This indicates the expression of the ligand gene L in sender cell i. The expression of receptor gene R in receiver cell j is represented by the summation of K and LR pairs for expression in cells i and j. S15. Filter the edges of the cell communication network based on random perturbation experiments, and delete the edges between cells of the same type and between cells from different patients. S16. Using cell type unique encoding or cell-specific gene expression as feature values, the feature matrix X of the node is obtained. A GNN model is constructed to obtain the embeddings of each node. The node embeddings of each view are integrated through an attention-based strategy to obtain the final embeddings (E): In the formula, Create an adjacency matrix for the cells in the current view. This is the node weight coefficient matrix for the current view. Here is the error matrix. This represents the attention weight, reflecting the contribution of each pair of cells to the interaction; S17. Construct a fully connected network to predict the survival of cancer patients, and use the Cox loss function to measure the accuracy of prognosis prediction: In the formula, , Representing patients Survival time and survival status This represents the survival time predicted by the neural network model; the loss function is calculated for all patients who experienced the endpoint event. The sum of losses; After training and optimization of the S18 and GNN models, the weights of the multi-view attention network are used... The survival impact of all cell interaction views was ranked, and tumor-immune cell interaction pairs with the greatest prognostic impact were screened.

2. The method for identifying key regulators of tumor immune interactions according to claim 1, characterized in that, Step S2 specifically includes: S21. For the tumor-immune cell interaction pairs that have the greatest prognostic impact, ligand and receptor genes that are specifically highly expressed in cells are screened and sorted based on differential expression analysis and random permutation test. S22. Construct a neural network model for predicting the gene expression of ligand and receptor genes by fusing multi-omics features; S23. Based on the attention mechanism or SHAP method, calculate the attention weight or shapley value of different features in the input matrix, and screen the top 10 features in terms of contribution as gene mutations, copy number variations and gene expression features of key regulators of ligand or receptor genes.

3. The method for identifying key regulators of tumor immune interactions according to claim 2, characterized in that, In step S23: Given the feature matrix X and gene expression vector q of a sample, what is the attention probability of the i-th feature vector? for: In the formula, z is the attention variable. That is, the current selection is the i-th feature vector. For attention scoring functions; The SHAP method, based on the concept of cooperative game theory, calculates the Shapley value to measure the marginal contribution of each feature to the model. It is the Shapley value of the i-th feature in set S: In the formula, N is the set of all features, v(S) is the value of set S, defined as the loss function value of the prediction model built based on set S, and |S| represents the size of set S.

4. The method for identifying key regulators of tumor immune interactions according to claim 1, characterized in that, Step S3 specifically includes the following steps: S31. Based on the gene information contained in the SNV site and the CNA fragment, key regulators are unified at the gene level, and functional enrichment analysis is performed on the identified key regulators. S32. Integrate single-cell and bulk multi-omics data of tumor samples from public data resources to observe the cell expression specificity of key regulators and the association between regulators and the expression of ligand receptor genes and cell marker genes they regulate. S33. Verify the function of regulators at the in vitro cellular level, construct a somatic mutant cell line transfection model, co-culture it with peripheral blood mononuclear cells, use Transwell migration chamber experiments to verify the effect of genomic variation on the migration ability of immune cells, and use multicolor immunohistochemical staining experiments to verify the changes in ligand receptor gene expression under the influence of genomic variation. S34. Construct a ligand-receptor-regulatory network, integrate protein interaction information from the STRING database, TRRUST, and the prior gene function network of the GRNdb transcriptional regulatory network, and apply network module analysis and restart random walk algorithm to identify the gene cascade pathways and regulatory modules that regulators influence ligand-receptor genes.

5. The method for identifying key regulators of tumor immune interactions according to claim 1, characterized in that, Step S4 specifically includes the following steps: S41. Collect clinical phenotypic information and multi-omics data of tumor samples treated with immune checkpoint inhibitors, including bulk-level genomic, transcriptomic and single-cell-level transcriptomic data. S42. Based on single-sample gene set enrichment analysis and AUCell algorithm, the expression activities of key regulatory functional modules in tissues and cells were calculated, and the expression activities of regulators and functional modules were calculated as related features of anti-tumor immune response process. These related features of anti-tumor immune response process include the correlation between tumor mutation burden, immune checkpoint gene expression, IFN-γ signaling, antigen presentation pathway, cell lysis activity, and immune infiltration ratio. S43. Based on the mutation status, expression level and functional module expression activity of key regulators, the clinical characteristics of different groups were compared using the chi-square test and Wilcoxon rank-sum test, including age, gender, tumor TNM stage and immunotherapy response rate. S44. Kaplan-Meier survival curves and log-rank tests were used to calculate the prognostic differences between groups, and univariate Cox regression and C-index methods were used to evaluate the prognostic efficacy of the regulators in immunotherapy. S45. Calculate AUC and Precision index to assess the predictive efficacy of regulators and their functional modules for immunotherapy response and immune-related adverse reactions.

Images