Construction method and application of multi-scale and multi-level regulation and control network

Abstract

The invention discloses a construction method and application of a multi-scale and multi-level regulation and control network, and relates to the field of biological information.The construction method comprises the following steps that firstly, 20 databases about interaction among protein, transcription factors, enhancer super enhancers, ncRNA and metabolites are collected; 2, performing quality control on each database, and annotating a specific disease physiological state to each regulation and control relationship; 3, integrating molecular regulation and control relationships among different levels according to biological logic; 4, a bidirectional random walk algorithm is adopted to cooperate with a multi-level biological network to predict and mine potential drug treatment targets; the regulation and control network constructed by the method has a multi-level structure and a specific disease physiological state, the dynamic regulation and control network conforming to the real disease physiological state can be constructed, and the method can be applied to mining potential drug targets and exploring pharmacological mechanisms.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method for constructing a multi-scale and multi-level regulatory network and its application. Background Art

[0002] Network pharmacology explores the key targets of drugs to exert therapeutic effects through biological networks, providing new ideas for the analysis of pharmacological mechanisms. Constructing biological networks is the core step in network pharmacology research, and the quality of biological networks directly affects the reliability of the results. Since current drugs mainly exert therapeutic effects by targeting proteins, the biological networks used in network pharmacology research focus on protein-protein interaction (PPI) networks. With the development of high-throughput sequencing technology, more precise molecular regulatory landscapes in the human body have been revealed, which cover a rich range of intracellular roles such as enhancers, transcription factors (TF), non-coding RNA (ncRNA) and proteins. Studies have shown that functional RNA and DNA may also become targets for new drugs. Although proteins are important substances for performing biological functions and PPI networks are indeed an important component of biological networks, complex biological regulatory networks cannot be restored by PPI networks alone. In addition, biological networks are multi-scale dynamic changes, and have different regulatory relationships in specific cells, tissues, organs, and disease states. It can be seen that constructing a multi-scale and multi-level regulatory network that is close to the real pathophysiological environment is of great significance for drug target prediction and mechanism analysis.

[0003] There are currently some biological network databases and regulatory network construction algorithms, but none of them simultaneously solves the problem of having multi-level regulatory molecules and restoring specific pathophysiological states. The STRING database and the BioGRID database are currently the most commonly used databases in network pharmacology research, but both only cover proteomics regulatory information. NetworkAnalyst and OmicsNet integrate multi-omics data such as proteomics and transcriptomics, but do not reflect the dynamic changes of biological networks under different pathophysiological states. The WGCNA and scHumanNet algorithms use bulk transcriptome and single-cell transcriptome data to construct biological networks specific to different tissues and cell types, respectively. Although they reflect the dynamic characteristics of the regulatory network, they ignore the multi-level biological regulatory relationships. In order to construct a multi-scale, multi-level regulatory network that is close to the real pathophysiological environment, the present invention proposes a multi-scale, multi-level regulatory network construction method. Summary of the invention

[0004] In order to address the deficiencies of the prior art, the purpose of the present invention is to provide a method for constructing a multi-scale and multi-level regulatory network and its application, which can solve the two defects of the biological regulatory network currently used in network pharmacology research that does not have a multi-level structure and a specific pathophysiological state, and realize the construction of a dynamic regulatory network that conforms to the real pathophysiological state.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for constructing a multi-scale and multi-level regulatory network comprises the following steps:

[0007] In step 1, 20 databases on the interactions between proteins, transcription factors, enhancers, super enhancers, nc RNAs, and metabolites were collected;

[0008] In step 2, each database was quality controlled and specific pathophysiological states were annotated for each regulatory relationship, resulting in a total of 1,133 different pathophysiological states, including 669 cell line-specific states, 281 tissue-specific states, and 183 disease-specific states;

[0009] The specific steps are as follows:

[0010] (a) Examine data sources, including data based on literature mining, experimental verification, and computational prediction;

[0011] (b) formulating various filtering rules for the data from literature mining, experimental verification and computational prediction sources, wherein the data from experimental verification sources are filtered according to high-throughput and low-throughput methods;

[0012] (c) Deletion of non-human biomolecular interaction information;

[0013] (d) annotation of pathophysiological states of molecular interactions in cell lines, tissues, and disease-specific conditions based on experimental or computational context;

[0014] (e) For proteins and mRNAs, the BioMart tool was used to uniformly convert gene IDs to Ensembl IDs, and all network files were organized into a unified format;

[0015] Step 3: Integrate the molecular regulatory relationships between different levels according to biological logic. The biological regulatory relationships include: the core process in the central dogma (DNA→RNA→protein), regulatory elements and mechanisms that affect the precise regulation of gene expression;

[0016] Regulatory elements include: enhancers, super enhancers, transcription factors, metabolites, and ncRNAs;

[0017] The multi-level regulatory network includes biological processes such as pre-transcriptional regulation, transcriptional regulation, pre-translational regulation and translational regulation; the elements of pre-transcriptional regulation include enhancers and super enhancers; the elements of transcriptional regulation include transcription factors; the elements of pre-translational regulation include ncRNA; the elements of translational regulation include proteins and metabolites;

[0018] Step 4: Use a bidirectional random walk algorithm in conjunction with a multi-level biological network to predict and mine potential drug treatment targets.

[0019] In the aforementioned method for constructing a multi-scale and multi-level regulatory network, the 20 databases on the interactions between proteins, transcription factors, enhancers, super enhancers, ncRNAs and metabolites in step one include: protein-protein interaction databases include: Bioplex, HuRI, String, BioGRID, DIP, MINT, InnateDB, PhosphoSitePlus and INSIDER; transcription factor-gene interaction databases include GRNdb, hTFtarget, KnockTF, TRRUST and IntACT; ncRNA-gene interaction databases include miRTarBase, ENCORI and ncFans; enhancer-gene interaction databases include Endb; super enhancer-gene interaction databases include SEdb; metabolite-protein interaction databases include KEGG.

[0020] In the aforementioned method for constructing a multi-scale and multi-level regulatory network, in step 2, literature mining only retains molecular interactions that have been manually verified by experts, and retains PPIs in BioGRID that have been verified to be physically interacting for many times; in the experimental verification, the high-throughput experiment uses the representative database KnockTF to provide the differential expression of transcriptomes before and after different TF knockouts, and sets |Log2FC|≥3 and P value <0.01 as thresholds to identify TF-regulated genes; in the experimental verification, the low-throughput experiment uses the representative database: INSIDER, and sets the filtering condition: high-quality co-crystal structure; computational prediction is filtered by setting a threshold, when the representative database used is: String, the filtering condition is set to a probability score of >900, and when the representative database is used, the filtering conditions are set: Pearson correlation coefficient >0.9 and P value <0.005 to filter the interactions between ncRNAs predicted by the WGCNA algorithm in ncFans.

[0021] In the aforementioned method for constructing a multi-scale and multi-level regulatory network, the random walk algorithm formula (1) in step 4 is shown as follows:

[0022] P t+1 =(1-λ)MP t +λP0 (Formula 1)

[0023] Among them, P 0 represents the initial probability vector of the random walk of the drug or disease, M is the transfer matrix. If the total number of nodes in the network is N, then M is obtained by normalizing the N×N neighbor matrix. The transfer matrix reflects the topological characteristics of the network. The walk probability λ∈(0,1) represents the probability of random walk to the next node, P t is the probability vector of the network nodes at time t; in order to prevent the random walk from iterating endlessly, ε=1e-6, when P t and P t+1 When the difference is less than ε, the random walk is considered to have reached a convergence state.

[0024] The aforementioned method for constructing a multi-scale and multi-level regulatory network is applied to predict therapeutic targets, and the process includes:

[0025] First, the drug target is set as the seed node of random walk to obtain the diffusion characteristics of the drug in the network, simulating the pharmacological phenomenon of drug diffusion in the human body through the molecular interaction network;

[0026] Then, the disease gene is set as the seed node of random walk to obtain the diffusion characteristics of the disease in the network and simulate the damage of the disease to the biological network;

[0027] The random walk algorithm formula (1) is shown as follows:

[0028] P t+1 =(1-λ)MP t +λP 0 (Formula 1)

[0029] Among them, P 0 represents the initial probability vector of the random walk of the drug or disease, M is the transfer matrix. If the total number of nodes in the network is N, then M is obtained by normalizing the N×N neighbor matrix. The transfer matrix reflects the topological characteristics of the network. The walk probability λ∈(0,1) represents the probability of random walk to the next node, P t is the probability vector of the network nodes at time t; in order to prevent the random walk from iterating endlessly, ε=1e-6, when P t and P t+1 When the difference is less than ε, the random walk is considered to have reached a convergence state;

[0030] Finally, the diffusion characteristics of drugs and diseases are used to find the nodes with the largest number of visits. These nodes are the genes most affected by drugs and diseases and are the key targets for drugs to exert their therapeutic effects. drug Indicates the probability of each node being affected by the drug target in a network with N nodes; P diseaseIt represents the probability of each node being affected by the pathogenic gene in a network with N nodes; Indicates the degree to which node k is affected by the drug; represents the degree to which node k is affected by the disease; Indicates the degree to which node k is affected by both drugs and diseases.

[0031] The aforementioned method for constructing a multi-scale and multi-level regulatory network is applied to the key targets that play a role in the combined treatment of non-small cell lung cancer (NSCLC) with sirolimus and trametinib, and the process includes:

[0032] Step 1: Construct a lung-specific multi-level regulatory network:

[0033] Specific networks related to lung cancer cell lines, lung tissues, and lung disease states were retrieved from the 1,133 specific networks collected, and protein-protein interaction networks from HuRI data, TF-gene interaction networks from the GRNdb database, and ncRNA-gene interaction networks from the ncFans database were obtained; biological regulatory relationships at different levels were merged using the above-mentioned biologically logical network integration strategy, and finally a lung-specific multi-level biological network with 20,370 nodes and 200,191 edges was obtained;

[0034] Step 2: Use the bidirectional random walk algorithm to explore potential therapeutic targets:

[0035] Five drug targets from the DrugBank database and 156 NSCLC-related genes from the DisGeNET database were used as seed nodes for random walks in the lung-specific multi-level biological network to capture drug diffusion characteristics. drug and disease spread characteristics P disease ; The key parameters are set as follows: walk probability = 0.2, number of iterations = 1000, convergence threshold = 1e-6;

[0036] Next, through Calculate the score of each node in the lung-specific multi-level biological network, Score k The probability that the drug and the disease visit node k at the same time. The higher the probability, the more likely node k is to be the key target for treatment.

[0037] Step 3: Result analysis:

[0038] The core subnetwork consists of drug targets and their neighboring nodes; all scores of 20,370 nodes in the lung-specific multi-level biological network were obtained, and the rankings are as follows: Rank 1 is gene MAP2K1, type is protein; Rank 2 is gene MAP2K1, type is protein; Rank 3 is gene MTOR, type is protein; Rank 4 is gene SMARCA4, type is TF; Rank 5 is gene E2F6, type is TF; Rank 6 is gene GTF2F1, type is TF; Rank 7 is gene HCFC1, type is TF; Rank 8 is gene ESRRA, type is TF.

[0039] The present invention is beneficial in that:

[0040] The regulatory network of the present invention has a multi-level structure and a specific pathophysiological state, and can realize the construction of a dynamic regulatory network that conforms to the real pathophysiological state;

[0041] The present invention can be used to construct a multi-level biological regulatory network, covering a variety of biological molecules such as proteins, RNA and metabolites, which is conducive to the discovery of potential drug targets other than proteins;

[0042] The present invention can be used to construct different cell line, tissue, and disease-specific biological regulatory networks, which can meet the analysis needs of researchers with different research backgrounds and conduct pharmacological mechanism analysis in a state closer to the actual pathophysiological state. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is the technical roadmap of the present invention;

[0044] Figure 2 is a quality control flow chart of the regulation relationship of the present invention;

[0045] Figure 3 is the biological regulation logic diagram of the present invention;

[0046] Figure 4 It is a schematic diagram of the therapeutic target prediction process of the bidirectional random walk of the present invention;

[0047] Figure 5 It is a schematic diagram of the results of the present invention predicting the key targets that play a role in the combined treatment of non-small cell lung cancer (NSCLC) with sirolimus and trametinib. DETAILED DESCRIPTION

[0049] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] like Figure 1As shown, the present invention is a data-driven regulatory network construction method, which collects multiple biomolecular regulatory relationships through high-quality databases, annotates different pathophysiological states according to the conditions when the data is produced, and finally constructs a multi-level specific regulatory network according to biological logic. The present invention collects 20 databases on the interactions between proteins, transcription factors, enhancers, super enhancers, nc RNA and metabolites. The protein-protein interaction databases include: Bioplex, HuRI, String, BioGRID, DIP, MINT, InnateDB, PhosphoSitePlus and INSIDER. The transcription factor-gene interaction databases include GRNdb, hTFtarget, KnockTF, TRRUST and IntACT. The ncRNA-gene interaction databases include miRTarBase, ENCORI and ncFans. The enhancer-gene interaction database includes Endb. The super enhancer-gene interaction database includes SEdb. The metabolite-protein interaction database includes KEGG.

[0051] The present invention performs quality control on each database and annotates specific pathophysiological states for each regulatory relationship. The specific steps are as follows:

[0052] (a) The present invention carefully examines the data sources, including data based on literature mining, experimental verification and computational prediction.

[0053] (b) Different filtering rules were formulated for data from different sources, and data from experimental sources were filtered according to high-throughput and low-throughput methods. Figure 2 As shown in the quality control flowchart of regulatory relationships, literature mining only retains molecular interactions that have been manually verified by experts, and retains PPIs in BioGRID that have been verified to be physically interacting many times; in experimental verification, high-throughput experiments use the representative database KnockTF to provide differential expression of transcriptomes before and after different TF knockouts, and set |Log2FC|≥3 and P value <0.01 as thresholds to identify TF-regulated genes; in the experimental verification, low-throughput experiments use the representative database: INSIDER, and set the filtering condition: high-quality co-crystal structure; computational predictions are filtered by setting thresholds. When the representative database used is: String, the filtering condition is set to a probability score of >900. When the representative database is used, the filtering conditions are set: Pearson correlation coefficient >0.9 and P value <0.005 to filter the interactions between ncRNAs predicted by the WGCNA algorithm in ncFans.

[0054] (c) Delete the biomolecular interaction information of non-human origin.

[0055] (d) Annotation of pathophysiological states of molecular interactions in cell lines, tissues, and disease-specific conditions based on experimental or computational context.

[0056] (e) For proteins and mRNAs, the BioMart tool was used to uniformly convert gene IDs to Ensembl IDs, and all network files were organized into a unified format.

[0057] A total of 1,133 different pathophysiological states were annotated by this method, including 669 cell line-specific states, 281 tissue-specific states, and 183 disease-specific states.

[0058] The present invention integrates the molecular regulatory relationships between different levels according to biological logic. When describing biological regulatory relationships, in addition to the core process in the central dogma (DNA→RNA→protein), there are also a series of regulatory elements and mechanisms that affect the precise regulation of gene expression. After adding regulatory elements such as enhancers, super enhancers, transcription factors, metabolites and ncRNAs, the regulatory relationship can be expressed as follows:

[0059] (a) Gene expression level. Enhancers are specific sequences on DNA that can remotely regulate the transcription level of genes. They increase the transcriptional activity of genes by binding to transcription factors. Super enhancers are highly active regulatory regions composed of multiple enhancers, which are usually associated with important genes that determine cell characteristics and identity. They can greatly increase the transcription level of specific genes and are crucial for the regulation of cell fate.

[0060] (b) Transcriptional regulation level. Transcription factors are proteins that can bind to specific sequences of DNA (such as promoters or enhancers), and they regulate the initiation of gene transcription. By binding to enhancers and promoters, transcription factors can enhance or inhibit the activity of RNA polymerase in the gene promoter region, thereby regulating the production of mRNA. ncRNA is an RNA molecule that is not translated into protein. They are involved in the multi-level regulation of gene expression, especially in post-transcriptional regulation and the regulation of chromatin structure. Important ncRNAs include miRNA and lncRNA, which regulate gene expression, inhibit transcription, or affect RNA stability by interacting with mRNA or proteins.

[0061] (c) Metabolic regulation level. Metabolites are small molecule compounds produced in organisms that can also regulate gene expression. Metabolites can indirectly affect gene transcription by directly binding to transcription factors or other regulatory proteins, changing the activity of these proteins. Fluctuations in metabolite concentrations can reflect the metabolic state within cells, thereby regulating the expression of related genes to adapt to metabolic needs.

[0062] The final multi-level regulatory network includes biological processes such as pre-transcriptional regulation (enhancers, super enhancers), transcriptional regulation (transcription factors), pre-translational regulation (ncRNA) and translational regulation (proteins, metabolites). The specific biological regulatory logic is as follows: Figure 3 shown.

[0063] Faced with complex regulatory networks constructed by massive multi-omics data, basic network topology attributes such as degree, network centrality and network density are no longer able to capture the biological significance behind the network. This paper proposes a bidirectional random walk algorithm for mining potential drug therapeutic targets in conjunction with multi-level biological networks.

[0064] When a drug produces a therapeutic effect, it may affect disease-related genes in the biological network through direct targeting or indirect intervention. The bidirectional random walk algorithm is based on the diffusion characteristics of drugs and diseases in the biological network in this biological logic step. The random walk algorithm formula (1) is shown as follows:

[0065] P t+1 =(1-λ)MP t +λP 0 (Formula 1)

[0066] Among them, P 0 represents the initial probability vector of the random walk of the drug or disease, M is the transfer matrix (if the total number of nodes in the network is N, then M is obtained by normalizing the N×N proximity matrix, and the transfer matrix reflects the topological characteristics of the network), the walk probability λ∈(0,1) represents the probability of random walk to the next node, P t is the probability vector of the network nodes at time t. In order to prevent the random walk from iterating endlessly, ε=1e-6. t and P t+1 When the difference is less than ε, the random walk is considered to have reached a convergence state.

[0067] The therapeutic target prediction process based on bidirectional random walk is as follows: First, the drug target is set as the seed node of the random walk to obtain the diffusion characteristics of the drug in the network, simulating the pharmacological phenomenon of the drug diffusing in the human body through the molecular interaction network. Then, the disease gene is set as the seed node of the random walk to obtain the diffusion characteristics of the disease in the network, simulating the damage of the disease to the biological network. Finally, the diffusion characteristics of drugs and diseases are used to find the nodes with the largest number of visits. These nodes are the genes most affected by drugs and diseases, and may be the key targets for drugs to exert their therapeutic effects.

[0068] like Figure 4 As shown, P drug Indicates the probability that each node is affected by the drug target in a network with N nodes. diseaseIt represents the probability that each node is affected by the pathogenic gene in a network with N nodes. Indicates the degree to which node k is affected by the drug. Indicates the degree to which node k is affected by the disease. k Indicates the degree to which node k is affected by both drugs and diseases.

[0069] The multi-scale and multi-level regulatory network construction method of the present invention is applied to predict therapeutic targets, and the prediction accuracy is verified by reproducing the key targets that play a role in the combined treatment of non-small cell lung cancer (NSCLC) with sirolimus and trametinib:

[0070] Step 1: Construct a lung-specific multi-level regulatory network

[0071] Specific networks related to lung cancer cell lines, lung tissues and lung disease states were retrieved from the 1,133 specific networks collected by the present invention, and protein-protein interaction networks from HuRI data, TF-gene interaction networks from GRNdb database, and ncRNA-gene interaction networks from ncFans database were obtained. The biological regulatory relationships at different levels were merged using the above-mentioned network integration strategy that conforms to biological logic, and finally a lung-specific multi-level biological network with 20,370 nodes and 200,191 edges was obtained.

[0072] Step 2: Use the bidirectional random walk algorithm to explore potential therapeutic targets

[0073] The present invention uses 5 drug targets from the DrugBank database and 156 NSCLC-related genes from the DisGeNET database as seed nodes for random walks in a lung-specific multi-level biological network to capture drug diffusion characteristics. drug and disease spread characteristics P disease The key parameters are set as follows: walk probability = 0.2, number of iterations = 1000, convergence threshold = 1e-6. Calculate the score of each node in the lung-specific multi-level biological network, Score k The probability that drugs and diseases visit node k at the same time. The higher the probability, the more likely node k is to be the key target for therapeutic effects.

[0074] Step 3: Result Analysis

[0075] like Figure 5The core subnetwork composed of the displayed drug targets and their neighboring nodes, the present invention obtains all the scores of 20,370 nodes in the lung-specific multi-level biological network through the above method, among which the top 8 are as follows Figure 5 As shown in the figure, all are included in the core subnetwork. The top three are known drug targets approved by the FDA. In addition, five key transcription factors were found. These transcription factors play an important role in the occurrence and development of cancer and are likely to be potential therapeutic targets.

[0076] The present invention integrates a variety of biomolecular interaction data according to biological logic to obtain a multi-scale and multi-level biological regulatory network under different pathophysiological states. The present invention also proposes a bidirectional random walk algorithm for discovering potential therapeutic targets of drugs in multi-level specific biological networks, and provides application cases. The results show that the combination of multi-level specific biological networks and bidirectional random walk algorithms can restore known targets and recommend new candidate therapeutic targets. In short, the regulatory network constructed by this method has spatiotemporal dynamics, and can meet the needs of different research backgrounds such as tumors, metabolic diseases, cardiovascular and cerebrovascular diseases, and nervous system diseases in conjunction with the bidirectional random walk algorithm, which is beneficial for researchers to explore potential drug targets and pharmacological mechanisms in a more realistic pathophysiological environment.

[0077] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any form, and any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A method for constructing a multi-scale and multi-level regulatory network, characterized in that: The steps include: In step 1, 20 databases on the interactions between proteins, transcription factors, enhancers, super enhancers, nc RNAs, and metabolites were collected; In step 2, each database was quality controlled and specific pathophysiological states were annotated for each regulatory relationship, resulting in a total of 1,133 different pathophysiological states, including 669 cell line-specific states, 281 tissue-specific states, and 183 disease-specific states; The specific steps are as follows: (a) Examine data sources, including data based on literature mining, experimental verification, and computational prediction; (b) formulating various filtering rules for the data from literature mining, experimental verification and computational prediction sources, wherein the experimental verification source data is filtered according to high-throughput and low-throughput methods; (c) Deletion of non-human biomolecular interaction information; (d) annotation of pathophysiological states of molecular interactions in cell lines, tissues, and disease-specific conditions based on experimental or computational context; (e) For proteins and mRNAs, the BioMart tool was used to uniformly convert gene IDs to Ensembl IDs, and all network files were organized into a unified format; Step 3: Integrate the molecular regulatory relationships between different levels according to biological logic. The biological regulatory relationships include: the core process in the central dogma (DNA→RNA→protein), regulatory elements and mechanisms that affect the precise regulation of gene expression; The regulatory elements include: enhancers, super enhancers, transcription factors, metabolites and ncRNA; The multi-level regulatory network includes biological processes such as pre-transcriptional regulation, transcriptional regulation, pre-translational regulation and translational regulation; the elements of pre-transcriptional regulation include enhancers and super enhancers; the elements of transcriptional regulation include transcription factors; the elements of pre-translational regulation include ncRNA; the elements of translational regulation include proteins and metabolites; Step 4: Use a bidirectional random walk algorithm in conjunction with a multi-level biological network to predict and mine potential drug treatment targets.

2. The method for constructing a multi-scale and multi-level regulatory network according to claim 1, characterized in that: The 20 databases on the interactions between proteins, transcription factors, enhancers, super enhancers, ncRNAs and metabolites in step 1 include: protein-protein interaction databases include: Bioplex, HuRI, String, BioGRID, DIP, MINT, InnateDB, PhosphoSitePlus and INSIDER. Transcription factor-gene interaction databases include GRNdb, hTFtarget, KnockTF, TRRUST and IntACT; ncRNA-gene interaction databases include miRTarBase, ENCORI and ncFans; enhancer-gene interaction databases include Endb. Super enhancer-gene interaction databases include SEdb. Metabolite-protein interaction databases include KEGG.

3. The method for constructing a multi-scale and multi-level regulatory network according to claim 1, characterized in that: The literature mining in step 2 only retains the molecular interactions verified by experts, and retains the PPIs that have been verified multiple times in BioGRID to be physically interacting; the high-throughput experiment in the experimental verification uses the representative database KnockTF to provide the differential expression of transcriptomes before and after different TF knockouts, and sets |Log2FC|≥3 and P value <0.01 as thresholds to identify TF-regulated genes; The experimental verification and low-throughput experiments used the representative database: INSIDER, and set the filtering conditions: high-quality co-crystal structure; the computational prediction was filtered by setting a threshold. When the representative database used was: String, the filtering conditions were set to a degree of affinity score > 900. When the representative database was used, the filtering conditions were set: Pearson correlation coefficient > 0.9 and P value < 0.005 to filter the interactions between ncRNAs predicted by the WGCNA algorithm in ncFans.

4. The method for constructing a multi-scale and multi-level regulatory network according to claim 1, characterized in that: The random walk algorithm formula (1) in step 4 is as follows: P t+1 =(1-λ)MP t +λP 0 (Formula 1) Among them, P 0 represents the initial probability vector of the random walk of the drug or disease, M is the transfer matrix. If the total number of nodes in the network is N, then M is obtained by normalizing the N×N neighbor matrix. The transfer matrix reflects the topological characteristics of the network. The walk probability λ∈(0,1) represents the probability of random walk to the next node, P t is the probability vector of the network nodes at time t; in order to prevent the random walk from iterating endlessly, ε=1e-6, when P t and P t+1 When the difference is less than ε, the random walk is considered to have reached a convergence state.

5. The application of the method for constructing a multi-scale and multi-level regulatory network according to claim 1, characterized in that: Applied to predict therapeutic targets, the process includes: First, the drug targets are set as seed nodes of random walks to obtain the diffusion characteristics of drugs in the network, simulating the pharmacological phenomenon of drugs diffusing in the human body through the molecular interaction network. Then, the disease gene is set as the seed node of random walk to obtain the diffusion characteristics of the disease in the network and simulate the damage of the disease to the biological network; The random walk algorithm formula (1) is shown as follows: P t+1 =(1-λ)MP t +λP 0 (Formula 1) Among them, P 0 represents the initial probability vector of the random walk of the drug or disease, M is the transfer matrix. If the total number of nodes in the network is N, then M is obtained by normalizing the N×N neighbor matrix. The transfer matrix reflects the topological characteristics of the network. The walk probability λ∈(0,1) represents the probability of random walk to the next node, P t is the probability vector of the network nodes at time t; in order to prevent the random walk from iterating endlessly, ε=1e-6, when P t and P t+1 When the difference is less than ε, the random walk is considered to have reached a convergence state; Finally, the diffusion characteristics of drugs and diseases are used to find the nodes with the largest number of visits. These nodes are the genes most affected by drugs and diseases and are the key targets for drugs to exert their therapeutic effects. drug Indicates the probability of each node being affected by the drug target in a network with N nodes; P disease It represents the probability of each node being affected by the pathogenic gene in a network with N nodes; Indicates the degree to which node k is affected by the drug; represents the degree to which node k is affected by the disease; Indicates the degree to which node k is affected by both drugs and diseases.

6. The application of the method for constructing a multi-scale and multi-level regulatory network according to claim 5, characterized in that: The key targets used in the combined treatment of non-small cell lung cancer (NSCLC) with sirolimus and trametinib include: Step 1: Construct a lung-specific multi-level regulatory network: Specific networks related to lung cancer cell lines, lung tissues, and lung disease states were retrieved from the 1,133 specific networks collected, and protein-protein interaction networks from HuRI data, TF-gene interaction networks from the GRNdb database, and ncRNA-gene interaction networks from the ncFans database were obtained; biological regulatory relationships at different levels were merged using the above-mentioned biologically logical network integration strategy, and finally a lung-specific multi-level biological network with 20,370 nodes and 200,191 edges was obtained; Step 2: Use the bidirectional random walk algorithm to explore potential therapeutic targets: Five drug targets from the DrugBank database and 156 NSCLC-related genes from the DisGeNET database were used as seed nodes for random walks in the lung-specific multi-level biological network to capture drug diffusion characteristics. drug and disease spread characteristics P disease ; The key parameters are set as follows: walk probability = 0.2, number of iterations = 1000, convergence threshold = 1e-6; Next, through Calculate the score of each node in the lung-specific multi-level biological network, Score k The probability that the drug and the disease visit node k at the same time. The higher the probability, the more likely node k is to be the key target for treatment. Step 3: Result analysis: The core subnetwork consists of drug targets and their neighboring nodes; all scores of 20,370 nodes in the lung-specific multi-level biological network were obtained, and the rankings are as follows: Rank 1 is gene MAP2K1, type is protein; Rank 2 is gene MAP2K1, type is protein; Rank 3 is gene MTOR, type is protein; Rank 4 is gene SMARCA4, type is TF; Rank 5 is gene E2F6, type is TF; Rank 6 is gene GTF2F1, type is TF; Rank 7 is gene HCFC1, type is TF; Rank 8 is gene ESRRA, type is TF.

Images