A method for identifying traditional chinese medicine targets for chronic hepatitis b based on a graph neural network

Abstract

This invention relates to the field of hepatitis B pharmacology, and discloses a method for identifying targets of traditional Chinese medicine (TCM) for chronic hepatitis B based on graph neural networks. First, the active ingredients and potential targets of at least two TCM formulas are extracted, and "TCM-component-target" networks are constructed respectively. From a TCM systems pharmacology database network, the degree centrality, betweenness centrality, and eigenvector centrality of node i in the data network are used as identification indicators for key nodes. Topological analysis is used to screen key nodes in the network, and then the importance of node i is comprehensively ranked using the PageRank algorithm. The core components and targets that rank highly in all TCM formulas are calculated. By constructing a modular network, the common components and target groups in the TCM formulas are calculated. Simultaneously, GO function and KEGG pathway enrichment analyses are performed to calculate the biological functions and signaling pathways associated with key targets. This invention utilizes molecular docking technology to verify the interaction between core targets and small drug molecules.

Description

Technical Field

[0001] This invention relates to the field of hepatitis B pharmacology, specifically to a method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on graph neural networks. Background Technology

[0002] Hepatitis B is a systemic infectious disease caused by the hepatitis B virus (HBV), and its main clinical manifestation is liver damage. Currently, the clinical treatment of hepatitis B mainly uses Western medicine regimens, such as antiviral drugs like entecavir and tenofovir disoproxil fumarate. While these drugs can slow disease progression, they carry the risk of drug resistance, and the long-term prognosis is not ideal.

[0003] Traditional Chinese medicine (TCM) has been passed down and developed in China for over 23 centuries. Its core concept lies in preventing and treating diseases by maintaining the body's internal balance—that is, the harmony of Yin and Yang. Chinese herbal medicine occupies a central position in this system, enhancing organ function and restoring the body's internal balance through compound prescriptions. As a valuable resource for drug development, the unique value of TCM compound prescriptions lies in their synergistic effect: the therapeutic effect produced by combining multiple medicinal materials is far more than a simple sum of the chemical components and physical properties of a single herb.

[0004] Xiao Chai Hu Granules originate from Zhang Zhongjing's *Treatise on Febrile Diseases*, and its formula consists of "Bupleurum chinense (half a catty), Scutellaria baicalensis (three liang), Ginseng (three liang), Pinellia ternata (half a liter, washed), Glycyrrhiza uralensis (prepared, three liang), Zingiber officinale (sliced, three liang), and jujubes (twelve pieces, broken)." This formula has significant effects in anti-liver damage, anti-inflammation, anti-tumor, and immunomodulation. Studies by Wang et al. have found that Xiao Chai Hu Tang can improve liver fibrosis through multi-target and multi-pathway mechanisms. Sedum sarmentosum, first recorded in *Compendium of Materia Medica Supplement*, is renowned for its effectiveness in detoxifying snake venom. According to the *Guidelines for the Diagnosis and Treatment of Chronic Hepatitis B with Traditional Chinese Medicine*, Sedum sarmentosum granules are significantly effective in protecting the liver and lowering enzyme levels.

[0005] Although a large number of clinical studies have been conducted on the treatment of chronic hepatitis B with Xiaochaihu granules and Sedum sarmentosum granules, their mechanism of action still lacks a systematic scientific explanation.

[0006] Therefore, to address the above problems, a method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on graph neural networks is needed. Summary of the Invention

[0007] The purpose of this invention is to provide a method for identifying targets of traditional Chinese medicine (TCM) for chronic hepatitis B based on graph neural networks. This invention first extracts the active ingredients and potential targets of two TCM formulas, constructing "TCM-component-target" networks respectively, and revealing network connectivity characteristics through node degree values. Then, degree centrality, eigenvector centrality, and PageRank topology analysis are used to screen key nodes in the network, identifying core components and targets that rank highly in both formulas. By constructing modular networks, the common components and target groups of the two therapies are further revealed. Simultaneously, GO function and KEGG pathway enrichment analyses are performed to elucidate the biological functions and signaling pathways associated with key targets. Finally, molecular docking technology is used to verify the interaction between core targets and small drug molecules.

[0008] This invention is implemented as follows:

[0009] This invention provides a method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on graph neural networks, which is specifically implemented according to the following steps:

[0010] S1: Extract the active ingredients and potential targets from at least two traditional Chinese medicine formulas, and construct "traditional Chinese medicine-ingredient-target" networks respectively; specifically, follow these steps:

[0011] S1.1: The active ingredients of traditional Chinese medicine formulas were obtained from the TCMSP database, along with their ADME parameters. Among them, oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18 were used to screen for active ingredients.

[0012] S1.2: Predict component targets using the STITCH database, unify gene names using the UniProt database, and form a traditional Chinese medicine-component-target network; as shown in the following formula:

[0013] J(A,B)=

[0014] Where J(A,B) is the similarity index of Chinese medicine components that highly overlap with disease targets, A is the set of predicted targets of active Chinese medicine components, B is the set of disease-related targets, |A∩B| is the number of intersection targets, and |A∪B| is the number of union targets.

[0015] S2: Extract data from the pharmacology database network of traditional Chinese medicine systems, focusing on nodes within the network. Degree centrality, betweenness centrality, and eigenvector centrality data are used as identification indicators for key nodes. Topology analysis methods are used to screen key nodes in the network, and then the PageRank algorithm is applied to further refine the node ranking. The importance of each factor should be ranked comprehensively; the specific steps are as follows:

[0016] S2.1: Based on the data of traditional Chinese medicine-components-targets, construct a heterogeneous graph G=(V,E,R), where: V is a set of nodes, including traditional Chinese medicine nodes V H Component node V C and target node V T E is the set of edges, representing the relationships between nodes (e.g., traditional Chinese medicine-ingredient relationship, ingredient-target relationship), and R is the set of relationship types (e.g., "contains", "acts on").

[0017] S2.2: Initialize a feature vector for each node. If features are unavailable, use one-hot encoding or learn an embedding.

[0018] For a Chinese medicine node v∈VH, the feature vector hv(0) may include the properties of the Chinese medicine such as its nature, flavor and meridian tropism.

[0019] For a component node v∈VC, the feature vector hv(0) includes chemical descriptors (such as molecular weight, LogP value).

[0020] For a target node v∈VT, the feature vector hv(0) includes gene expression data or protein sequence features;

[0021] The feature matrix is ​​as follows:

[0022] ;

[0023] Where d is the feature dimension;

[0024] Within the framework of the adjacency matrix, nodes The degree centrality of a vector is calculated by the sum of the elements of its corresponding row vectors, as follows:

[0025] ;

[0026] in, Let N represent the degree centrality of a node, and let N represent the total number of nodes in the network.

[0027] node The betweenness centrality is expressed as follows:

[0028] ;

[0029] in, Represents a set of network nodes. For nodes The total number of shortest paths between them. This indicates that these paths pass through the nodes. The quantity, when hour, ;like ,but .

[0030] node The eigenvector centrality values ​​are as follows:

[0031] ;

[0032] Among them, define nodes For vectors The One portion, Let G be the adjacency matrix. If we take the eigenvalues ​​of the adjacency matrix, then there exists a unique solution when we take the largest eigenvalue of the adjacency matrix. And all of its components are positive.

[0033] By employing a modularity index to evaluate the tightness of connections between similar nodes, the nodes are analyzed. Importance is ranked comprehensively; modularity is an important indicator for measuring the significance of connections between nodes of the same type, and its value ranges from negative to positive: when the value approaches 1, it indicates that the internal connections of the module are highly compact; if the value approaches 0, it means that the nodes are randomly connected, as shown in the following formula:

[0034] ;

[0035] in, Represents the total number of edges in the network. For the community Number of internal edges This represents the sum of the degree values ​​of the nodes within the community. For resolution parameters, resolution parameters To balance the weight ratio of internal and external connections within a community, different weights are selected. The values ​​are analyzed and the ranking results are integrated for the same network.

[0036] S2.3: Use a multi-layer graph convolutional network (GCN) to process heterogeneous graphs. For each layer l=0,1,…,K−1 (K is the total number of layers), update the node features as follows:

[0037] ;

[0038] in, $ is the node feature matrix of the $th layer;

[0039] It is an adjacency matrix With the addition of self-loops, I is the identity matrix, used to ensure that the node's own information is included;

[0040] yes The degree matrix (diagonal matrix), where It is the first The trainable weight matrix of the layer;

[0041] It is an activation function;

[0042] It is a normalization operation used to avoid gradient vanishing or exploding.

[0043] S2.4: For target nodes, calculate their importance score using a multilayer perceptron (MLP) as the readout function, as shown in the following equation:

[0044] ;

[0045] in, The target node v is in the last layer. eigenvectors.

[0046] It is a node The importance score indicates the strength of the association between the target and chronic hepatitis B;

[0047] It is a multilayer perceptron;

[0048] S2.5: If supervised data is available (such as known key targets of chronic hepatitis B), train the model using the cross-entropy loss function; as shown in the following formula:

[0049] ;

[0050] in: It is the actual label of node v. Indicates key target points, Indicates non-critical target points; It is a predicted probability; It is the number of target nodes;

[0051] S2.6: Based on the score Sort all target nodes in descending order and select the node with the highest score as the core target.

[0052] These core targets will be combined with topology analysis results, for example, by taking the intersection of the rankings of the two or by weighted fusion, to improve reliability.

[0053] S3: Calculate the core components and targets that rank highly in traditional Chinese medicine formulas. By constructing a modular network, calculate the common components and target groups in traditional Chinese medicine formulas. Simultaneously conduct GO function and KEGG pathway enrichment analysis to calculate the biological functions and signaling pathways associated with key targets.

[0054] Follow these steps in detail:

[0055] S3.1: GO function and KEGG pathway enrichment analysis, firstly performing a hypergeometric distribution test, as shown in the following equation:

[0056] ;

[0057] Where FDR is the control for false positive rate, N is the total number of background genes, K is the number of genes associated with a specific GO / KEGG entry, n is the size of the study gene set, and k is the number of genes in the study gene set associated with the entry.

[0058] S3.2: Then calculate the enrichment score as follows:

[0059] ;

[0060] ;

[0061] S3.3: Next, perform key target function mapping, resulting in the target-function correlation matrix, as shown below:

[0062] .

[0063] S4: Verify the interaction between the core target and the drug molecule using molecular docking. The specific steps are as follows:

[0064] S4.1: First, the target protein structure is preprocessed. The target protein crystal structure is obtained from the PDB database, water molecules and ligands are removed, hydrogen atoms are added, and energy minimization is performed using OpenBabel, as shown in the following formula;

[0065] ;

[0066] S4.2: Ligand small molecule structure optimization. First, the ligand molecule structure is drawn, and then the geometry is optimized using Gaussian 09, B3LYP / 6-31G* basis set, as shown in the following equation:

[0067] ;

[0068] S4.3: To perform molecular docking calculations, first define the docking search space, then perform a conformational search and calculate the binding energy, as shown in the following equation:

[0069] ;

[0070] .

[0071] Furthermore, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on a graph neural network as described above.

[0072] Furthermore, the present invention provides a computer-storable medium storing a computer program, wherein when the program is executed, it sequentially executes any one of the above-described methods for identifying the target of traditional Chinese medicine for chronic hepatitis B based on a graph neural network.

[0073] Compared with the prior art, the beneficial effects of the present invention are:

[0074] 1. This invention systematically elucidates the commonly used components and targets in the treatment of chronic hepatitis B, integrating network science and clinical data to systematically evaluate the efficacy of chronic hepatitis B treatment, thus providing a scientific analysis of traditional Chinese medicine formulas. Simultaneously, it elevates pharmacological effects to the molecular level, laying a solid foundation for drug mechanism research. This invention provides a new pathway for further exploring the therapeutic potential of traditional Chinese medicine formulas.

[0075] 2. This invention, by integrating network pharmacology and topology analysis methods, reveals the mechanism of action of traditional Chinese medicine at the system level. Integrating network science and clinical data will effectively bridge the gap between traditional Chinese medicine and modern medicine, providing a scientific basis for evaluating the efficacy of Chinese medicine prescriptions. Attached Figure Description

[0076] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0077] Figure 1 This is a flowchart of the method of the present invention;

[0078] Figure 2 This is a network diagram of the "Chinese medicine-component-target" of the two Chinese medicines of the present invention;

[0079] Figure 3 This is a protein interaction network diagram of the two traditional Chinese medicine targets of the present invention;

[0080] Figure 4 This is a graph showing the enrichment analysis results of GO and KEGG in this invention;

[0081] Figure 5 This is a visualization of the optimal docking sites of the present invention, drawn using PyMOL; wherein, Figure 5In the diagram, (A) is the ESR2-isorhamnine docking conformation, (B) is the CYP19A1-isorhamnine docking conformation, (C) is the PTPN1-isorhamnine docking conformation, (D) is the ESR2-quercetin docking conformation, (E) is the CYP19A1-quercetin docking conformation, and (F) is the PTPN1-quercetin docking conformation. Detailed Implementation

[0082] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to describe selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0083] Please see Figures 1-5 This invention provides a method for identifying targets of traditional Chinese medicine for chronic hepatitis B based on graph neural networks, which is specifically implemented according to the following steps:

[0084] S1: Extract the active ingredients and potential targets from at least two traditional Chinese medicine formulas, and construct "traditional Chinese medicine-ingredient-target" networks respectively; specifically, follow these steps:

[0085] S1.1: The active ingredients of traditional Chinese medicine formulas were obtained from the TCMSP database, along with their ADME parameters. Among them, oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18 were used to screen for active ingredients.

[0086] S1.2: Predict component targets using the STITCH database, unify gene names using the UniProt database, and form a traditional Chinese medicine-component-target network; as shown in the following formula:

[0087] J(A,B)= ;

[0088] Where J(A,B) is the similarity index of Chinese medicine components that highly overlap with disease targets, A is the set of predicted targets of active Chinese medicine components, B is the set of disease-related targets, |A∩B| is the number of intersection targets, and |A∪B| is the number of union targets.

[0089] S2: Extract data from the pharmacology database network of traditional Chinese medicine systems, focusing on nodes within the network. Degree centrality, betweenness centrality, and eigenvector centrality data are used as identification indicators for key nodes. Topology analysis methods are used to screen key nodes in the network, and then the PageRank algorithm is applied to further refine the node ranking. The importance of each factor should be ranked comprehensively; the specific steps are as follows:

[0090] S2.1: Based on the data of traditional Chinese medicine-components-targets, construct a heterogeneous graph G=(V,E,R), where: V is a set of nodes, including traditional Chinese medicine nodes V H Component node V C and target node V T E is the set of edges, representing the relationships between nodes (e.g., traditional Chinese medicine-ingredient relationship, ingredient-target relationship), and R is the set of relationship types (e.g., "contains", "acts on").

[0091] S2.2: Initialize a feature vector for each node. If features are unavailable, use one-hot encoding or learn an embedding.

[0092] For a Chinese medicine node v∈VH, the feature vector hv(0) may include the properties of the Chinese medicine such as its nature, flavor and meridian tropism.

[0093] For a component node v∈VC, the feature vector hv(0) includes chemical descriptors (such as molecular weight, LogP value).

[0094] For a target node v∈VT, the feature vector hv(0) includes gene expression data or protein sequence features;

[0095] The feature matrix is ​​as follows:

[0096] ;

[0097] Where d is the feature dimension;

[0098] Within the framework of the adjacency matrix, nodes The degree centrality of a vector is calculated by the sum of the elements of its corresponding row vectors, as follows:

[0099] ;

[0100] in, Let N represent the degree centrality of a node, and let N represent the total number of nodes in the network.

[0101] node The betweenness centrality is expressed as follows:

[0102] ;

[0103] in, Represents a set of network nodes. For nodes The total number of shortest paths between them. This indicates that these paths pass through the nodes. The quantity, when hour, ;like ,but .

[0104] node The eigenvector centrality values ​​are as follows:

[0105] ;

[0106] Among them, define nodes For vectors The One portion, Let G be the adjacency matrix. If we take the eigenvalues ​​of the adjacency matrix, then there exists a unique solution when we take the largest eigenvalue of the adjacency matrix. And all of its components are positive.

[0107] By employing a modularity index to evaluate the tightness of connections between similar nodes, the nodes are analyzed. Importance is ranked comprehensively; modularity is an important indicator for measuring the significance of connections between nodes of the same type, and its value ranges from negative to positive: when the value approaches 1, it indicates that the internal connections of the module are highly compact; if the value approaches 0, it means that the nodes are randomly connected, as shown in the following formula:

[0108] ;

[0109] in, Represents the total number of edges in the network. For the community Number of internal edges This represents the sum of the degree values ​​of the nodes within the community. For resolution parameters, resolution parameters To balance the weight ratio of internal and external connections within a community, different weights are selected. The values ​​are analyzed and the ranking results are integrated for the same network.

[0110] S2.3: Use a multi-layer graph convolutional network (GCN) to process heterogeneous graphs. For each layer l=0,1,…,K−1 (K is the total number of layers), update the node features as follows:

[0111] ;

[0112] in, $ is the node feature matrix of the l-th layer;

[0113] It is an adjacency matrix Adding self-loops (where I is the identity matrix) ensures that the node's own information is included;

[0114] yes The degree matrix (diagonal matrix), where It is the first The trainable weight matrix of the layer;

[0115] It is an activation function;

[0116] It is a normalization operation used to avoid gradient vanishing or exploding.

[0117] S2.4: For target nodes, calculate their importance score using a multilayer perceptron (MLP) as the readout function, as shown in the following equation:

[0118] ;

[0119] in, The target node v is in the last layer. eigenvectors.

[0120] It is a node The importance score indicates the strength of the association between the target and chronic hepatitis B;

[0121] MLP stands for Multilayer Perceptron;

[0122] S2.5: If supervised data is available (such as known key targets of chronic hepatitis B), train the model using the cross-entropy loss function; as shown in the following formula:

[0123] ;

[0124] in: It is a node The true label, Indicates key target points, Indicates something is not critical; It is a predicted probability; It is the number of target nodes;

[0125] S2.6: Based on the score Sort all target nodes in descending order and select the node with the highest score as the core target.

[0126] These core targets will be combined with topology analysis results, for example, by taking the intersection of the rankings of the two or by weighted fusion, to improve reliability.

[0127] S3: Calculate the core components and targets that rank highly in traditional Chinese medicine formulas. By constructing a modular network, calculate the common components and target groups in traditional Chinese medicine formulas. Simultaneously conduct GO function and KEGG pathway enrichment analysis to calculate the biological functions and signaling pathways associated with key targets.

[0128] Follow these steps in detail:

[0129] S3.1: GO function and KEGG pathway enrichment analysis, firstly performing a hypergeometric distribution test, as shown in the following equation:

[0130] ;

[0131] Where FDR is the control for false positive rate, N is the total number of background genes, K is the number of genes associated with a specific GO / KEGG entry, n is the size of the study gene set, and k is the number of genes in the study gene set associated with the entry.

[0132] S3.2: Then calculate the enrichment score as follows:

[0133] ;

[0134] ;

[0135] S3.3: Next, perform key target function mapping, resulting in the target-function correlation matrix, as shown below:

[0136] .

[0137] S4: Verify the interaction between the core target and the drug molecule using molecular docking. The specific steps are as follows:

[0138] S4.1: First, the target protein structure is preprocessed. The target protein crystal structure is obtained from the PDB database, water molecules and ligands are removed, hydrogen atoms are added, and energy minimization is performed using OpenBabel, as shown in the following formula;

[0139] ;

[0140] S4.2: Ligand small molecule structure optimization. First, the ligand molecule structure is drawn, and then the geometry is optimized using Gaussian 09, B3LYP / 6-31G* basis set, as shown in the following equation:

[0141] ;

[0142] S4.3: To perform molecular docking calculations, first define the docking search space, then perform a conformational search and calculate the binding energy, as shown in the following equation:

[0143] ;

[0144] .

[0145] In this embodiment, step S1 of the present invention constructs a "traditional Chinese medicine-component-target" network. This network includes the active ingredients of traditional Chinese medicine formulas and the corresponding potential target information. At the same time, by calculating the similarity index J(A,B) of traditional Chinese medicine components, traditional Chinese medicine components that highly overlap with disease targets are screened out, providing basic data for subsequent key node screening.

[0146] Step S2: Based on the "Traditional Chinese Medicine-Component-Target" network data constructed in Step S1, a heterogeneous graph G=(V,E,R) is built, clarifying the relationships between different types of nodes. Simultaneously, using the component and target information determined in Step S1, feature vectors are initialized for each node, such as the properties and meridian tropism of traditional Chinese medicine nodes, chemical descriptors for component nodes, and gene expression data or protein sequence features for target nodes.

[0147] Step S2 processes the heterogeneous graph using a multilayer graph convolutional network (GCN) to update node features. A multilayer perceptron (MLP) is then used to calculate the importance score (sv) of the target nodes. Based on this score, all target nodes are sorted in descending order, and the node with the highest score is selected as the core target. Furthermore, topological analysis methods such as degree centrality, betweenness centrality, and eigenvector centrality are used to screen key network nodes. The PageRank algorithm is then used to comprehensively rank node importance, and the modularity index is used to assess the connection tightness between similar nodes, further enhancing the overall node importance ranking.

[0148] Step S3 utilizes the core targets identified in Step S2 to calculate the top-ranking core components and targets in the traditional Chinese medicine formula. By constructing a modular network, based on the component and target information determined in Steps S1 and S2, the common components and target groups in the traditional Chinese medicine formula are calculated. These common components and target groups, along with the core component and target information, will serve as input data for GO function and KEGG pathway enrichment analysis to calculate the biological functions and signaling pathways associated with key targets.

[0149] Step S3 uses GO function and KEGG pathway enrichment analysis to calculate the biological functions and signaling pathways associated with key targets, and obtains the target-function association matrix Mij, clarifying the association between each target and different functions. This helps to identify which targets play key roles in biological functions and signaling pathways.

[0150] Step S4, based on the key targets identified in Step S3, retrieves the target protein crystal structure from the PDB database and performs target protein structure preprocessing, including removing water molecules and ligands, adding hydrogen atoms, and using OpenBabel energy minimization. Simultaneously, for drug small molecules interacting with the key targets, ligand small molecule structures are optimized, ligand molecular structures are plotted, and geometry optimization is performed using Gaussian 09, B3LYP / 6-31G* basis sets. Finally, based on the processed target protein structure and the optimized ligand small molecule structure, molecular docking calculations are performed to verify the interaction relationship between the core target and the drug small molecule. Ultimately, this achieves accurate identification and interaction verification of traditional Chinese medicine targets for chronic hepatitis B.

[0151] In this embodiment, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on a graph neural network as described above.

[0152] In this embodiment, the present invention provides a computer-storable medium storing a computer program, wherein when the program is executed, it sequentially executes any one of the above-described methods for identifying the target of traditional Chinese medicine for chronic hepatitis B based on a graph neural network.

[0153] In this embodiment, the present invention constructed traditional Chinese medicine-component-target networks for Xiao Chai Hu granules and Chui Pen Cao granules, respectively, such as... Figure 2 The "TCM-component-target" networks of two TCMs are shown. (A) Network map of Xiao Chai Hu granules, (B) Network map of Sedum sarmentosum granules. In the Xiao Chai Hu granule network, the degree values ​​of quercetin, kaempferol, isorhamnetin, and baicalin are 101, 101, 84, and 49, respectively; the degree values ​​of key targets CYP19A1, ESR1, and CA12 are 23, 22, and 13, respectively. In the Sedum sarmentosum granule network, the degree values ​​of quercetin, luteolin, and isorhamnetin are 101, 97, and 84, respectively; the degree value of the core target CYP19A1 is 5, and the degree values ​​of targets such as CA7, ESR2, ABCC1, CA12, CA4, CYP1B1, ADORA1, and ABCCG2 are all 4.

[0154] In this embodiment, the ranking comparison of different key node identification methods is carried out. The top 20 nodes are selected as key components and targets using different centrality methods. The results are summarized in Tables 1 to 4.

[0155] Table 1. Ranking of different identification methods for components of Xiaochaihu granules

[0156]

[0157] Table 2 Ranking of different identification methods for Xiaochaihu granule targets

[0158] Table 3. Ranking of different identification methods for the granular components of Sedum sarmentosum.

[0159]

[0160] Table 4. Ranking of different identification methods for Sedum sarmentosum granular target points

[0161]

[0162] In this embodiment, as shown in Table 1, quercetin, kaempferol, isorhamnetin, and baicalein ranked among the top in all centrality methods; lotus leaf alkaloid and 24-ethylcholest-4-en-3-one ranked highly in degree centrality, betweenness centrality, PageRank, and overall degree value. Table 2 shows that CYP19A1 and ESR2 consistently ranked highly in all methods, and PTPN1 was also identified in multiple methods. Comprehensive analysis shows that the key components of Xiao Chai Hu granules include quercetin, kaempferol, isorhamnetin, baicalein, nuciferine, and 24-ethylcholest-4-en-3-one, with key potential targets being CYP19A1, ESR2, and PTPN1. The core components of Sedum sarmentosum granules are quercetin, luteolin, and isorhamnetin, with major targets including CYP19A1, CA7, ESR2, ABCC1, CA12, CA4, CYP1B1, ADORA1, and ABCCG2. Xiao Chai Hu granules and Sedum sarmentosum granules show significant overlap in their mechanisms of action. Quercetin and isorhamnetin were identified as key shared components in both therapies, while CYP19A1, ESR2, and the CA series targets are important potential targets in both treatments. This suggests that both therapies may exert their therapeutic effects on chronic hepatitis B through similar pathways.

[0163] In this embodiment, a modular network was constructed and mapped to a protein-protein interaction network, as shown in the following figure. Figure 3 As shown. Protein-protein interaction networks of two traditional Chinese medicine targets. (A) PPI network of Xiaochaihu granules, (B) PPI network of Sedum sarmentosum granules. Figure 3The target network of Xiao Chai Hu granules exhibits ten modules, while that of Sedum sarmentosum granules is divided into two modules. Notably, targets such as ADORA3, PTGS1, ABCB1, CYP1B1, CBR1, HSD17B1, CA4, ADORA1, CA12, SHBG, ABCG2, ESR1, ESR2, and CYP19A1 are grouped into the same module in both Xiao Chai Hu granules and Sedum sarmentosum granules, suggesting that the two therapies have similar mechanisms of action. Furthermore, key targets in Xiao Chai Hu granules (such as EGFR, PPARG, ESR1, PTGS2, AKT1, and SRC) highly overlap with the core targets of Sedum sarmentosum granules (ESR1, AR, PTGS2, SRC, EGFR, and AKT1), further confirming the strong similarity of their mechanisms of action.

[0164] In this embodiment, R language was used to perform GO functional enrichment analysis and KEGG pathway enrichment analysis on the two drugs, and the results were visualized. GO analysis covered three dimensions: biological processes (BP), cellular components (CC), and molecular functions (MF). Based on the criteria of p-value < 0.05 and number of genes > 3, a total of 291 significantly enriched functional items were selected. Table 5 lists the top 20 enriched items. Figure 4 (A) presents the visualization results of the first eight core functions. Analysis shows that at the biological process level, the two drugs mainly regulate rhythmic processes, organic anion transport, positive regulation of the MAPK cascade reaction, and hormone metabolism; cellular components involve structures such as the apical structure, cell protrusion membrane, axonal membrane, apical membrane, and basal membrane; molecular functions are mainly related to carbonic anhydrase activity, hydrolase activity, lyase activity, nuclear steroid receptor activity, carbonyl lyase activity, and iron ion binding. KEGG pathway enrichment analysis shows that the two drugs exert their effects through multiple signaling pathways, among which ovarian steroid production, chemical oncogenesis-receptor activation, nitrogen metabolism, and steroid hormone biosynthesis are particularly significant. The relevant analysis results are visualized in... Figure 4 (B).

[0165] Table 5. Enrichment results of the top 20 biological processes

[0166]

[0167] In this embodiment, based on the aforementioned network pharmacology, quercetin and isorhamnetin are key components of two traditional Chinese medicines, and CYP19A1, ESR2, and PTPN1 were identified as important potential targets for both therapies. First, the macromolecular structures of CYP19A1, ESR2, and PTPN1 were obtained from the PDB database, and the small molecule structures of quercetin and isorhamnetin were downloaded from the PubChem database. Pairwise molecular docking was then performed using Autodock software. Quercetin and isorhamnetin all possess multiple low-binding-energy docking sites with CYP19A1, ESR2, and PTPN1. Figure 5 The optimal docking conformation was visualized using Pymol, verifying the interaction between the core targets of quercetin and isorhamnetin and the small drug molecules. This ultimately achieved accurate identification and interaction verification of the targets of traditional Chinese medicine for chronic hepatitis B, as shown in Table 6.

[0168] Table 6. Validation of the minimum binding energy interaction between target proteins and small drug molecules.

[0169]

[0170] In summary, this invention, based on network topology analysis and following the principles of network pharmacology, systematically elucidates the commonly used components and targets in the treatment of chronic hepatitis B, and verifies the interaction between key targets and small drug molecules through molecular docking. The fusion of network science and clinical data allows for a systematic evaluation of the efficacy of chronic hepatitis B treatment, thereby bridging the gap between traditional medicine and modern science at the systemic level and providing a scientific explanation for traditional Chinese medicine formulas. Simultaneously, it elevates pharmacological effects to the molecular level, laying a solid foundation for drug mechanism research. This invention highlights the significant value of integrating network science and clinical data in revealing the mechanisms of traditional Chinese medicine in treating chronic hepatitis B, and provides a new pathway for further exploring the therapeutic potential of traditional Chinese medicine formulas.

[0171] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations will be apparent to those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on graph neural networks, characterized in that: Follow these steps: S1: Extract the active ingredients and potential targets of at least two traditional Chinese medicine formulas, and construct "traditional Chinese medicine-ingredient-target" networks respectively; S2: Extract data from the pharmacology database network of traditional Chinese medicine systems, focusing on nodes within the network. Degree centrality, betweenness centrality, and eigenvector centrality data are used as identification indicators for key nodes. Topology analysis methods are used to screen key nodes in the network, and then the PageRank algorithm is applied to further refine the node ranking. Ranking by importance; S3: Calculate the core components and targets that rank highly in traditional Chinese medicine formulas. By constructing a modular network, calculate the common components and target groups in traditional Chinese medicine formulas. Simultaneously conduct GO function and KEGG pathway enrichment analysis to calculate the biological functions and signaling pathways associated with key targets. S4: Verify the interaction between the core target and the drug molecule using molecular docking.

2. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S1, the following steps are specifically performed: S1.1: The active ingredients of traditional Chinese medicine formulas were obtained from the TCMSP database, along with their ADME parameters. Among them, oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥ 0.18 were used to screen for active ingredients. S1.2: Predict component targets using the STITCH database, unify gene names using the UniProt database, and form a traditional Chinese medicine-component-target network; as shown in the following formula: ； Where J(A,B) is the similarity index of Chinese medicine components that highly overlap with disease targets, A is the set of predicted targets of active Chinese medicine components, B is the set of disease-related targets, |A∩B| is the number of intersection targets, and |A∪B| is the number of union targets.

3. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S2, the following steps are specifically performed: S2.1: Based on the data of traditional Chinese medicine-components-targets, construct a heterogeneous graph G=(V,E,R), where: V is a set of nodes, including traditional Chinese medicine nodes V H Component node V C and target node V T E is the set of edges, representing the relationships between nodes; R is the set of relation types. S2.2: Initialize a feature vector for each node. If features are unavailable, use one-hot encoding or learn an embedding. For a Chinese herbal medicine node v∈VH, the feature vector hv(0) may include the properties of the Chinese herbal medicine; for a component node v∈VC, the feature vector hv(0) includes chemical descriptors. For a target node v∈VT, the feature vector hv(0) includes gene expression data or protein sequence features; The feature matrix is ​​as follows: ； Where d is the feature dimension; S2.3: Use a multi-layer graph convolutional network (GCN) to process heterogeneous graphs. For each layer l=0,1,…,K−1 (K is the total number of layers), update the node features as follows: ； in, $ is the node feature matrix of the $th layer; It is an adjacency matrix With the addition of self-loops, I is the identity matrix, used to ensure that the node's own information is included; yes The degree matrix, where It is the first The trainable weight matrix of the layer; It is an activation function; It is a normalization operation used to avoid gradient vanishing or exploding. S2.4: For target nodes, calculate their importance score using a multilayer perceptron (MLP) as the readout function, as shown in the following equation: ； in, It is the feature vector of the target node v in the last layer K; It is the importance score of node v, which indicates the strength of the association between the target and chronic hepatitis B; MLP stands for Multilayer Perceptron; S2.5: If supervised data is available, train the model using the cross-entropy loss function; as shown in the following formula: ； in: It is the actual label of node v. Indicates key target points, Indicates non-critical target points; It is a predicted probability; It is the number of target nodes; S2.6: Sort all target nodes in descending order according to the score sv, and select the node with the highest score as the core target.

4. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S2, within the framework of the adjacency matrix, nodes The degree centrality of a vector is calculated by the sum of the elements of its corresponding row vectors, as follows: ； in, Let N represent the degree centrality of a node, and let N represent the total number of nodes in the network.

5. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: node The betweenness centrality is expressed as follows: ； in, Represents a set of network nodes. For nodes The total number of shortest paths between them. This indicates that these paths pass through the nodes. The quantity, when hour, ;like ,but .

6. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: node The eigenvector centrality values ​​are as follows: ； Among them, define nodes For vectors The One portion, Let G be the adjacency matrix. If we take the eigenvalues ​​of the adjacency matrix, then there exists a unique solution when we take the largest eigenvalue of the adjacency matrix. And all of its components are positive.

7. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S1, the degree of connection between similar nodes is evaluated using a modularity index, and the nodes are then evaluated. The importance is ranked comprehensively, as shown in the following formula: ； in, Represents the total number of edges in the network. For the community Number of internal edges This represents the sum of the degree values ​​of the nodes within the community. For resolution parameters, resolution parameters To balance the weight ratio of internal and external connections within a community, different weights are selected. The values ​​are analyzed and the ranking results are integrated for the same network.

8. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S3, the following steps are performed in detail: S3.1: GO function and KEGG pathway enrichment analysis, firstly performing a hypergeometric distribution test, as shown in the following equation: ； Where FDR is the control for false positive rate, N is the total number of background genes, K is the number of genes associated with a specific GO / KEGG entry, n is the size of the study gene set, and k is the number of genes in the study gene set associated with the entry. S3.2: Then calculate the enrichment score as follows: ； ； S3.3: Next, perform key target function mapping, resulting in the target-function correlation matrix, as shown below: 。 9. The method for identifying traditional Chinese medicine targets for chronic hepatitis B based on graph neural networks according to claim 1, characterized in that: In step S4, the following steps are specifically performed: S4.1: First, the target protein structure is preprocessed. The target protein crystal structure is obtained from the PDB database, water molecules and ligands are removed, hydrogen atoms are added, and energy minimization is performed using OpenBabel, as shown in the following formula; ； S4.2: Ligand small molecule structure optimization. First, the ligand molecule structure is drawn, and then the geometry is optimized using Gaussian 09, B3LYP / 6-31G* basis set, as shown in the following equation: ； S4.3: To perform molecular docking calculations, first define the docking search space, then perform a conformational search and calculate the binding energy, as shown in the following equation: ； 。 10. A computer-storable medium storing a computer program, characterized in that: When the computer program in the storage medium is executed, it performs the method for identifying the target of traditional Chinese medicine for chronic hepatitis B based on graph neural networks, as described in any one of claims 1-9.

Images