A multi-omics regulatory network-oriented key node identification method and system

Abstract

The application discloses a kind of multi-omics regulatory network-oriented key node identification method and system, it is related to multi-omics data processing technical field, method includes: to data is standardized, constructs multi-omics regulatory network;Screening candidate key node;Determine disease syndrome related node, and search the cross-layer regulatory path between candidate key node and disease syndrome related node, calculate cross-layer regulatory contribution;According to the basic score of candidate key node and cross-layer regulatory contribution form initial key node score, carry out hierarchical conservation disturbance to multi-omics regulatory network, and according to the stability of ranking after disturbance, initial key node score is corrected, determine key node and its verification priority.

Description

Technical Field

[0001] This invention relates to the field of multi-omics data processing technology, specifically to a method and system for identifying key nodes in multi-omics regulatory networks. Background Technology

[0002] With the development of high-throughput detection technologies, the same sample can generate data at different levels, such as transcriptomics, proteomics, metabolomics, and phenotypic states. Different omics data reflect different states in biological processes. By establishing correlations and regulatory relationships between omics features, multi-omics regulatory networks can be formed, and key nodes closely related to the target phenotypic state can be further identified from these networks. The results of key node identification can be used to narrow down the scope of subsequent analysis and provide a data foundation for the analysis of regulatory mechanisms and the screening of experimental validation targets.

[0003] Current methods for identifying key nodes in multi-omics regulatory networks typically rely heavily on network topology metrics such as node degree, betweenness centrality, and proximity centrality, or on ranking nodes based on the degree of omics differences. While these methods reflect the importance of a node within a local network structure, for multi-omics networks with multi-layered structures, a node's high connectivity does not necessarily mean it can continuously influence downstream nodes related to the target phenotype along credible cross-layer regulatory relationships. For example, a gene node may have many connections within its own gene layer, but its corresponding gene-to-protein and protein-to-metabolite regulatory chains lack reliable support. In such cases, relying solely on local network attributes can easily increase the node's identification priority. Therefore, the original scheme further introduces disease-related nodes, constrained cross-layer paths, and cross-layer regulatory contributions to evaluate the ability of candidate nodes to transmit regulatory influences across multi-omics levels.

[0004] Furthermore, multi-omics networks often contain multiple similar paths sharing partial connections. Directly summing the contributions from different paths can lead to duplicate calculations of the same regulatory evidence due to varying path combinations. Simultaneously, multi-omics networks are susceptible to detection errors, statistical estimation errors, and uncertainties in connectivity. Some nodes may achieve high rankings based solely on a few specific connections, and their rankings can fluctuate significantly once local network relationships change reasonably. Therefore, minimizing the impact of duplicate paths and accidental network structures while considering the effectiveness of cross-omics regulation, and obtaining highly stable key node identification results, presents a crucial technical challenge in processing multi-omics regulatory network data. Summary of the Invention

[0005] To address the above situation and overcome the shortcomings of existing technologies, the present invention adopts the following technical solution: The present invention provides a method for identifying key nodes in multi-omics regulatory networks, which includes the following steps:

[0006] Step S1: Standardize the multi-omics data to obtain multi-omics standardized data;

[0007] Step S2: Construction of multi-omics regulatory network, identifying gene nodes, protein nodes, metabolite nodes and phenotype nodes, and establishing intra-layer association edges and cross-layer regulatory edges between nodes to obtain the multi-omics regulatory network;

[0008] Step S3: Initial screening of candidate key nodes. Based on the basic network attributes, omics difference strength and phenotypic association strength of each node in the multi-omics regulatory network, a basic score is given to each node, and nodes whose basic scores meet the preset initial screening conditions are determined as candidate key nodes, thus obtaining a set of candidate key nodes.

[0009] Step S4: Cross-level regulation at the omics level. Based on the multi-omics regulation network, determine the set of disease-related nodes. Starting from the candidate key nodes in the candidate key node set, search for cross-level regulation paths between the candidate key nodes and the set of disease-related nodes. Based on the confidence of the regulation edges, path length, and number of omics levels crossed in the cross-level regulation path, combined with the overlap between the cross-level regulation paths and the disease-related association strength of the disease-related nodes, calculate the cross-level regulation contribution of the candidate key nodes to obtain the cross-level regulation contribution data.

[0010] Step S5: Key node identification. Based on the candidate key node set and the cross-layer regulation contribution data, calculate the key node score of the candidate key nodes, and perform perturbation verification on the multi-omics regulation network. Correct the key node score based on the ranking stability of the candidate key nodes after perturbation verification. Determine the candidate key nodes whose corrected key node scores meet the preset identification conditions as key nodes, and obtain the key node identification result.

[0011] Furthermore, in step S1, the multi-omics data standardization processing is used to uniformly organize multi-omics data from different sources, with different detection scales and different data structures to form basic data that can be used for subsequent network construction. Specifically, it involves obtaining transcriptome data, proteome data, metabolome data and phenotypic state data corresponding to the target sample, performing sample identification unification, data quality processing, batch effect correction and feature identification mapping on different omics data, and organizing different omics data according to sample number, detection object and feature identification to obtain multi-omics standardized data.

[0012] The multi-omics standardized data specifically includes transcriptomic standardized data, proteomic standardized data, metabolomic standardized data, and phenotypic state standardized data; wherein, in one specific embodiment, the phenotypic state standardized data includes pulmonary function indicators, inflammatory response indicators, oxidative stress indicators, and syndrome quantification indicators corresponding to COPD lung-kidney qi deficiency syndrome samples; thereby, multi-omics standardized data is obtained.

[0013] Further, in step S2, the construction of the multi-omics regulatory network is used to establish node association relationships between different omics levels based on the multi-omics standardized data, forming a network structure that can characterize multi-omics regulatory relationships. Specifically, gene nodes, protein nodes, metabolite nodes, and phenotype nodes are determined based on the multi-omics standardized data, and intra-level association edges and cross-level regulatory edges are established between nodes based on the correlation relationships between omics features at the same level, the mapping relationships between different omics features, and the known regulatory relationships between different omics levels, thus obtaining the multi-omics regulatory network.

[0014] The multi-omics regulatory network specifically includes omics node data and node connection data; the omics node data includes gene nodes, protein nodes, metabolite nodes and phenotype nodes; the node connection data includes intra-layer association edges, gene-to-protein cross-layer mapping edges, protein-to-metabolite functional association edges, and metabolite-to-phenotype state phenotypic association edges; thus, the multi-omics regulatory network is obtained.

[0015] Further, in step S3, the preliminary screening of candidate key nodes is used to screen candidate nodes with further computational value from the multi-omics regulation network, thereby reducing the node range for subsequent cross-layer regulation calculations. Specifically, based on the basic network attributes, omics difference strength, and phenotypic association strength of each node in the multi-omics regulation network, a basic score is given to each node, and nodes whose basic scores meet the preset preliminary screening conditions are determined as candidate key nodes, thus obtaining a set of candidate key nodes.

[0016] The candidate key node set specifically includes candidate node identifiers, candidate node omics level, candidate node basic network attributes, and candidate node state association data; the candidate node state association data includes candidate node omics difference strength and candidate node phenotypic association strength; wherein, the basic network attributes include node degree, betweenness centrality, proximity centrality, and eigenvector centrality; thus, the candidate key node set is obtained.

[0017] Further, in step S4, the omics-level cross-layer regulation is used to calculate the effective contribution of candidate key nodes in transmitting regulatory influence to target disease-related nodes across different omics levels. Specifically, it involves determining the set of disease-related nodes based on the multi-omics regulatory network, and starting from the candidate key nodes in the candidate key node set, searching for cross-layer regulatory paths between the candidate key nodes and the set of disease-related nodes. Based on the confidence of the regulatory edges, path length, and number of omics levels crossed in the cross-layer regulatory paths, combined with the overlap between the cross-layer regulatory paths and the disease-related association strength of the disease-related nodes, the cross-layer regulatory contribution of the candidate key nodes is calculated to obtain cross-layer regulatory contribution data, including the following steps:

[0018] Step S41: Determine disease-related nodes. This step is used to determine the target nodes for cross-layer regulatory contribution calculation. Specifically, it involves screening nodes that meet the disease-related conditions based on the association between multi-omics nodes and phenotypic states, and recording the corresponding disease-related strength to obtain a set of disease-related nodes.

[0019] Step S42: Constrained cross-layer path search, used to obtain the connection path with actual regulatory significance between candidate key nodes and disease-related nodes. Specifically, starting from the candidate key node, cross-layer regulatory paths are searched according to the omics level direction, path length limit and node non-repetition condition to obtain a set of candidate cross-layer regulatory paths.

[0020] Step S43: Single path credibility contribution calculation, used to distinguish the effective regulation capabilities corresponding to different cross-layer regulation paths. Specifically, the basic credibility of the path is determined based on the credibility of each regulation edge in the path, and attenuation processing is performed based on the path length and the number of omics layers crossed to obtain the effective contribution value of the path.

[0021] Step S44: Redundancy removal of overlapping path contributions, used to reduce the duplicate contributions caused by similar paths sharing a large number of regulatory edges. Specifically, it calculates the degree of edge overlap between cross-layer regulatory paths pointing to the same disease-related node, and reduces the effective contribution of subsequent paths according to the degree of path overlap to obtain redundancy-removed path contribution data.

[0022] Step S45: Disease and syndrome target contribution aggregation, used to determine the overall cross-layer impact of candidate key nodes on the set of disease and syndrome related nodes. Specifically, it involves bounded aggregation of the redundancy removal path contribution corresponding to the same disease and syndrome related node, and combining the disease and syndrome association strength to integrate different disease and syndrome related nodes to obtain the cross-layer regulation contribution value of candidate key nodes.

[0023] Step S46: Generation of cross-layer regulation results, which is used to form the data required for subsequent key node identification. Specifically, it records the disease-related nodes that can be reached by the candidate key nodes, the corresponding cross-layer regulation paths, the confidence of the path edges, the path level span, and the cross-layer regulation contribution value to obtain cross-layer regulation contribution data.

[0024] The cross-layer regulation contribution data specifically includes the candidate key node identifier, the omics level to which the candidate key node belongs, the number of disease-related nodes reachable by the candidate key node, and the cross-layer regulation contribution value of the candidate key node; the cross-layer regulation contribution data also includes the cross-layer regulation path corresponding to the candidate key node, the edge credibility of the cross-layer regulation path, and the hierarchical span of the cross-layer regulation path.

[0025] Further, in step S5, the key node identification is used to determine the final key node identification result based on the cross-layer regulation contribution and network stability of the candidate key nodes. Specifically, it involves calculating the key node score of the candidate key nodes based on the candidate key node set and the cross-layer regulation contribution data, performing perturbation verification on the multi-omics regulation network, correcting the key node score based on the ranking stability of the candidate key nodes after perturbation verification, and identifying the candidate key nodes whose corrected key node scores meet the preset identification conditions as key nodes to obtain the key node identification result. This includes the following steps:

[0026] Step S51: Initial score of candidate nodes, used to determine the basic ranking of each candidate key node without network disturbance. Specifically, it is to combine the basic score of candidate key nodes and the cross-layer regulation contribution value according to a unified scale to obtain the initial key node score and initial ranking of candidate nodes.

[0027] Step S52: Hierarchical conservation network perturbation, used to generate network change samples while maintaining the basic structure of the original multi-omics network. Specifically, it involves perturbing the confidence of the control edges with a limited amplitude, and performing degree-preserving permutation on some node connections according to edge type and connection omics hierarchy to obtain multiple perturbation control networks.

[0028] Step S53: Repeated scoring of perturbation network is used to observe the identification results of candidate key nodes under the condition of local network change. Specifically, the cross-layer control contribution and key node score of candidate key nodes are recalculated in each perturbation control network to obtain the perturbation ranking sequence of candidate nodes.

[0029] Step S54: Ranking robustness calculation, used to quantify the sensitivity of candidate key nodes to network disturbances. Specifically, it calculates the robustness score based on the average deviation of the candidate node's disturbance ranking relative to the initial ranking and the retention ratio within the allowed ranking range, thus obtaining the candidate node robustness data.

[0030] Step S55: Key node score correction, used to reduce the identification priority of high-scoring nodes that rely on accidental network structure. Specifically, the initial key node score is reduced according to the robustness score of the candidate node, and the key node set and ranking are determined according to the corrected key node score.

[0031] Step S56: Key node verification information is generated to determine the subsequent verification order of key nodes. Specifically, the verification priority is determined based on the cross-layer regulation contribution and robustness score of key nodes, and the corresponding main cross-layer regulation paths are extracted to obtain the key node identification results.

[0032] The key node identification results specifically include a set of key nodes, a ranking of key nodes, the omics level to which the key nodes belong, and the cross-layer regulatory contribution value of the key nodes; the key node identification results also include a key node robustness score, the main cross-layer regulatory path corresponding to the key node, and the key node verification priority; wherein, the key node verification priority is determined jointly based on the key node cross-layer regulatory contribution value and the robustness score; thus, the key node identification results are obtained.

[0033] This invention provides a key node identification system for multi-omics regulatory networks, including a multi-omics data standardization processing module, a multi-omics regulatory network construction module, a candidate key node initial screening module, an omics-level cross-layer regulation module, and a key node identification module;

[0034] The multi-omics data standardization processing module is used to uniformly organize multi-omics data from different sources, with different detection scales and different data structures. Through multi-omics data standardization processing, multi-omics standardized data is obtained, and the multi-omics standardized data is transmitted to the multi-omics regulatory network construction module.

[0035] The multi-omics regulation network construction module is used to establish node association relationships between different omics levels. Through the construction of the multi-omics regulation network, a multi-omics regulation network is obtained, and the multi-omics regulation network is transmitted to the candidate key node screening module and the omics level cross-level regulation module.

[0036] The candidate key node screening module is used to screen candidate nodes with further computational value from the multi-omics regulation network. Through the candidate key node screening, a set of candidate key nodes is obtained, and the set of candidate key nodes is transmitted to the omics-level cross-layer regulation module.

[0037] The omics-level cross-layer regulation module is used to calculate the effective contribution of candidate key nodes to the transmission of regulatory influence to disease-related nodes. Through omics-level cross-layer regulation, cross-layer regulation contribution data is obtained, and the cross-layer regulation contribution data is transmitted to the key node identification module.

[0038] The critical node identification module is used to determine the final critical node based on the cross-layer regulation contribution and network stability of the candidate critical nodes, and to obtain the critical node identification result through critical node identification.

[0039] The beneficial effects achieved by the present invention using the above solution are as follows:

[0040] (1) In the existing multi-omics regulatory network key node identification, there is a technical problem that the selection is mainly based on the topological importance of nodes or the degree of difference of a single omics, which makes it difficult to simultaneously reflect the regulatory influence of nodes across omics levels to the disease status and the stability of the identification results. This scheme creatively adopts a key node identification method that combines multi-omics regulatory network construction, candidate node screening, cross-level regulatory contribution calculation and network perturbation verification. This expands the evaluation of node importance from local network characteristics to cross-level regulatory ability and network robustness, thereby improving the correlation and reliability between the key node identification results and the target disease status.

[0041] (2) In response to the technical problems in the existing evaluation of cross-level regulatory relationships, such as the repeated accumulation of multiple highly overlapping paths, the overestimation of the contribution of excessively long indirect paths, and the failure to effectively distinguish the importance of target nodes of different diseases, this scheme creatively adopts the calculation of single-path contribution based on the credibility of regulatory edge, path length and number of cross-levels, and combines the redundancy removal of path overlap and the bounded contribution aggregation method of disease association strength weighting, so that the evidence of repeated regulation and weak indirect regulation relationship are reasonably suppressed, and the ability of cross-level regulation contribution to represent the real and effective regulatory chain is improved.

[0042] (3) In response to the technical problem that high scores of nodes may depend on a small number of specific connections in the existing key node ranking, and the ranking fluctuates significantly after slight changes in the network structure, resulting in insufficient stability of the screening results, this scheme creatively adopts hierarchical conservation network perturbation, perturbation network repeated scoring and ranking robustness correction methods. Under the condition of maintaining the basic structure of omics hierarchy and connection scale, it simulates the changes in edge credibility and local connection relationship, and uses perturbation ranking stability to correct the key node scores, thereby reducing the impact of accidental network structure on the final identification results and improving the robustness of key node screening and subsequent verification order. Attached Figure Description

[0043] Figure 1 A flowchart illustrating a key node identification method for multi-omics regulatory networks provided by this invention;

[0044] Figure 2 This is a schematic diagram of a key node identification system for multi-omics regulatory networks provided by the present invention;

[0045] Figure 3 This is a flowchart illustrating the process of step S4, omics-based cross-level regulation.

[0046] Figure 4 This is a flowchart illustrating the process of identifying key nodes in step S5.

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0048] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0049] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0050] Example 1, see Figure 1 This invention provides a method for identifying key nodes in multi-omics regulatory networks, which includes the following steps:

[0051] Step S1: Standardize the multi-omics data to obtain multi-omics standardized data;

[0052] Step S2: Construction of multi-omics regulatory network, identifying gene nodes, protein nodes, metabolite nodes and phenotype nodes, and establishing intra-layer association edges and cross-layer regulatory edges between nodes to obtain the multi-omics regulatory network;

[0053] Step S3: Initial screening of candidate key nodes. Based on the basic network attributes, omics difference strength and phenotypic association strength of each node in the multi-omics regulatory network, a basic score is given to each node, and nodes whose basic scores meet the preset initial screening conditions are determined as candidate key nodes, thus obtaining a set of candidate key nodes.

[0054] Step S4: Cross-level regulation at the omics level. Based on the multi-omics regulation network, determine the set of disease-related nodes. Starting from the candidate key nodes in the candidate key node set, search for cross-level regulation paths between the candidate key nodes and the set of disease-related nodes. Based on the confidence of the regulation edges, path length, and number of omics levels crossed in the cross-level regulation path, combined with the overlap between the cross-level regulation paths and the disease-related association strength of the disease-related nodes, calculate the cross-level regulation contribution of the candidate key nodes to obtain the cross-level regulation contribution data.

[0055] Step S5: Key node identification. Based on the candidate key node set and the cross-layer regulation contribution data, calculate the key node score of the candidate key nodes, and perform perturbation verification on the multi-omics regulation network. Correct the key node score based on the ranking stability of the candidate key nodes after perturbation verification. Determine the candidate key nodes whose corrected key node scores meet the preset identification conditions as key nodes, and obtain the key node identification result.

[0056] By performing the above operations, this solution addresses the technical problem in existing multi-omics regulatory network key node identification methods. These methods primarily rely on node topological importance or the degree of difference in a single omics layer, making it difficult to simultaneously reflect the transmission of regulatory influence across omics levels to disease states and the stability of identification results. This approach creatively employs a key node identification method that combines multi-omics regulatory network construction, initial screening of candidate nodes, calculation of cross-layer regulatory contributions, and network perturbation verification. This expands the evaluation of node importance from local network characteristics to cross-layer regulatory capabilities and network robustness, improving the correlation and reliability between key node identification results and target disease states. For example, in multi-omics data on COPD lung and kidney qi deficiency syndrome, it can further distinguish between nodes with high connectivity only within a single omics layer and nodes capable of forming stable regulatory pathways to disease-related phenotypes through genes, proteins, and metabolites, thereby reducing the possibility of locally highly connected nodes being misidentified as key nodes.

[0057] Example 2, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S1, the multi-omics data standardization processing is used to uniformly organize multi-omics data from different sources, different detection scales, and different data structures to form basic data that can be used for subsequent network construction. Specifically, it involves obtaining transcriptome data, proteome data, metabolome data, and phenotypic state data corresponding to the target sample, performing sample identification unification, data quality processing, batch effect correction, and feature identification mapping on different omics data, and organizing different omics data according to sample number, detection object, and feature identification to obtain multi-omics standardized data.

[0058] The multi-omics standardized data specifically includes transcriptomic standardized data, proteomic standardized data, metabolomic standardized data, and phenotypic state standardized data; wherein, in one specific embodiment, the phenotypic state standardized data includes pulmonary function indicators, inflammatory response indicators, oxidative stress indicators, and syndrome quantification indicators corresponding to COPD lung-kidney qi deficiency syndrome samples; thereby, multi-omics standardized data is obtained.

[0059] Example 3, see Figure 1 , Figure 2This embodiment is based on the above embodiment. In step S2, the construction of the multi-omics regulatory network is used to establish node association relationships between different omics levels based on the multi-omics standardized data, forming a network structure that can characterize multi-omics regulatory relationships. Specifically, gene nodes, protein nodes, metabolite nodes and phenotype nodes are determined based on the multi-omics standardized data, and the same-level association edges and cross-level regulatory edges between nodes are established based on the correlation between omics features at the same level, the mapping relationship between different omics features and the known regulatory relationships between different omics levels, so as to obtain the multi-omics regulatory network.

[0060] The multi-omics regulatory network specifically includes omics node data and node connection data; the omics node data includes gene nodes, protein nodes, metabolite nodes and phenotype nodes; the node connection data includes intra-layer association edges, gene-to-protein cross-layer mapping edges, protein-to-metabolite functional association edges, and metabolite-to-phenotype state phenotypic association edges; thus, the multi-omics regulatory network is obtained.

[0061] Example 4, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S3, the preliminary screening of candidate key nodes is used to screen candidate nodes with further computational value from the multi-omics regulation network, thereby reducing the node range for subsequent cross-layer regulation calculations. Specifically, based on the basic network attributes, omics difference strength, and phenotypic association strength of each node in the multi-omics regulation network, a basic score is given to each node, and nodes whose basic scores meet the preset preliminary screening conditions are determined as candidate key nodes, thus obtaining a set of candidate key nodes.

[0062] The candidate key node set specifically includes candidate node identifiers, candidate node omics level, candidate node basic network attributes, and candidate node state association data; the candidate node state association data includes candidate node omics difference strength and candidate node phenotypic association strength; wherein, the basic network attributes include node degree, betweenness centrality, proximity centrality, and eigenvector centrality; thus, the candidate key node set is obtained.

[0063] Example 5, see Figure 1 , Figure 2 and Figure 3This embodiment is based on the above embodiment. In step S4, the omics-level cross-layer regulation is used to calculate the effective contribution of candidate key nodes to transmit regulatory influence to target disease-related nodes across different omics levels. Specifically, it involves determining the set of disease-related nodes based on the multi-omics regulatory network, and starting from the candidate key nodes in the candidate key node set, searching for cross-layer regulatory paths between the candidate key nodes and the set of disease-related nodes. Based on the confidence of the regulatory edges, path length, and number of omics levels crossed in the cross-layer regulatory paths, combined with the overlap between the cross-layer regulatory paths and the disease-related association strength of the disease-related nodes, the cross-layer regulatory contribution of the candidate key nodes is calculated to obtain the cross-layer regulatory contribution data, including the following steps:

[0064] Step S41: Determine disease-related nodes. This step is used to determine the target nodes for cross-layer regulatory contribution calculation. Specifically, it involves screening nodes that meet the disease-related conditions based on the association between multi-omics nodes and phenotypic states, and recording the corresponding disease-related strength to obtain a set of disease-related nodes.

[0065] In practice, the standardized phenotypic state data obtained in step S1 is used as the target state reference. Correlation analysis is performed on gene nodes, protein nodes, and metabolite nodes in the multi-omics regulatory network with each phenotypic state index. For continuous phenotypic indices, Spearman correlation coefficient is preferred to calculate the correlation between node feature values ​​and phenotypic indices. The Benjamini-Hochberg method is used to perform multiple tests to correct the significance probability obtained from the association test in the same batch, so as to reduce the accidental significant association caused by the large number of multi-omics nodes.

[0066] For COPD lung and kidney qi deficiency syndrome samples, phenotypic state indicators were divided into four categories according to lung function state, inflammatory response state, oxidative stress state, and syndrome quantification state. Within each category, the association strength between nodes and corresponding phenotypic indicators was calculated, and the association results of indicators within the same category were taken as an arithmetic mean. Then, the four phenotypic states were combined using an equal-weighted approach. The disease-syndrome association strength of node t can be expressed as:

[0067] ;

[0068] in, represents the disease-symptom association strength of node t; g represents the phenotypic state category; This indicates the number of indicators included in the g-th phenotypic state; This represents the Spearman correlation coefficient between node t and the j-th phenotypic index in class g; This represents the false detection rate of the correlation test after multiple comparison correction;

[0069] Both the correlation coefficient and the false detection rate are dimensionless quantities, therefore Also a dimensionless quantity, its value ranges from 0 to 1. The larger the value, the more stable and obvious the statistical correspondence between the changes in that node and the target disease state.

[0070] Preferably, the The condition for disease association is that the node has an effective association with at least one of the four phenotypic states with an FDR of not more than 0.05. The purpose of adopting the above dual conditions is to avoid including nodes in the disease-related node set based solely on a single extremely weak but statistically significant association result, and to avoid forming a high comprehensive value after simple averaging of multiple insignificant associations.

[0071] Therefore, nodes that meet the disease-symptom association criteria, along with their node identifiers, their respective omics levels, and the strength of their disease-symptom association, will be included. Record these data to create a set of disease-related nodes;

[0072] Step S42: Constrained cross-layer path search, used to obtain the connection path with actual regulatory significance between candidate key nodes and disease-related nodes. Specifically, starting from the candidate key node, cross-layer regulatory paths are searched according to the omics level direction, path length limit and node non-repetition condition to obtain a set of candidate cross-layer regulatory paths.

[0073] In practice, the multi-omics regulatory network obtained in step S2 is used to build an adjacency list according to the omics level of the nodes and the type of connection edges. Each candidate key node obtained in step S3 is used as the starting node to perform a constrained graph search. For cross-layer mapping edges from genes to proteins, functional association edges from proteins to metabolites, and phenotypic association edges from metabolites to phenotypic states, the search is performed according to the cross-layer regulatory direction determined during network construction. Returning to the upstream omics level already visited along the cross-layer regulatory edge is not allowed. For intra-layer association edges, connection expansion within the current omics level is allowed.

[0074] During the path search process, the node identifiers that have been passed are recorded synchronously for the current search path. When the node to be expanded already exists in the current path, the branch expansion is stopped to avoid forming a loop. At the same time, it is stipulated that a valid cross-layer control path must contain at least one cross-layer control edge, and the current path is recorded when the disease-related node determined in step S41 is reached for the first time, and the path is no longer expanded downward from the target node.

[0075] Preferably, the maximum number of edges in a single candidate cross-layer regulatory path is set to 6. This parameter is determined based on the hierarchical structure of the gene layer, protein layer, metabolite layer, and phenotype layer in this embodiment, so that candidate nodes can enter the effective cross-layer regulatory chain through the necessary intra-layer association relationship, while restricting excessively long indirect association paths from entering subsequent contribution calculations. In actual data processing, when the number of network layers changes, the upper limit of the path length can be predetermined according to the maximum effective regulatory layer depth of the target multi-omics network.

[0076] For each searched path, record the starting candidate node, the nodes along the route, the target disease-related nodes, the connecting edges contained in the path, and the types of connecting edges, and count the number of path edges. and the number of cross-layer control edges. ;in Indicates path length. Both represent the number of actual omics level transitions, and are dimensionless count parameters.

[0077] Thus, the set of candidate cross-layer control paths corresponding to each candidate key node is obtained;

[0078] Step S43: Single path credibility contribution calculation, used to distinguish the effective regulation capabilities corresponding to different cross-layer regulation paths. Specifically, the basic credibility of the path is determined based on the credibility of each regulation edge in the path, and attenuation processing is performed based on the path length and the number of omics layers crossed to obtain the effective contribution value of the path.

[0079] In practice, the first step is to uniformly convert the regulatory edges from different sources in the multi-omics regulatory network into edge confidence levels ranging from 0 to 1. For edge associations or phenotypic edges established based on sample data statistics, the absolute correlation coefficient between corresponding features and the significance after multiple tests are used together as the basis for confidence. For example, the confidence of the edge is determined by multiplying the absolute correlation coefficient by 1 and subtracting the corresponding FDR value. For regulatory edges from known gene-protein mapping relationships or protein-metabolite function databases, the original confidence score of the database is linearly mapped to 0 to 1 according to the effective score range specified by the database to obtain the confidence of the corresponding edge.

[0080] Therefore, although the various edges have different origins, they are all calculated uniformly based on the dimensionless relative reliability, avoiding the direct mixing of correlation coefficients, significance probabilities, and database scores with inconsistent dimensions.

[0081] For including Connecting edge, crossing The effective contribution value of path P at the sub-omics level. The preferred method is to calculate according to the following formula:

[0082] ;

[0083] in, Indicates the effective contribution value of the path. This represents the edge confidence level of the edge connecting edge e in path P; This represents the geometric mean of the confidence levels of all control edges within the path. This represents the control edge with the lowest credibility in the path; Indicates the path length half-life scale; Indicates the scale of cross-layer regulation half-life;

[0084] Preferably, Choosing 3 means that, all other things being equal, after adding 3 more connecting edges to the path compared to a direct connection, the path contribution caused by indirect transmission is reduced to half of the original. Taking 2 indicates that when the regulatory effect cumulatively crosses 2 omics levels, the uncertainty brought about by cross-detection platform and cross-biological regulatory stage reduces the corresponding contribution to one-half of the original.

[0085] Using the geometric mean of path credibility instead of directly multiplying the credibility of control edges can avoid an exponential decrease in credibility simply due to an increase in the number of path edges; further introducing the credibility of the lowest edge of the path is to reflect the limitation of weak connections in the continuous control chain on the interpretability of the entire path; then, attenuation is implemented according to the path length and the number of levels, thereby distinguishing between the two types of uncertainty: whether the edge itself is credible and whether the control relationship is too indirect.

[0086] , The corresponding ratio, The corresponding ratios are all dimensionless quantities, therefore the effective contribution value of the path is... A dimensionless index in the range of 0 to 1. The larger the value, the higher the degree of effective regulatory support for the cross-level regulatory path;

[0087] Therefore, the effective contribution value of each path in the candidate cross-layer regulation path set is obtained;

[0088] Step S44: Redundancy removal of overlapping path contributions, used to reduce the duplicate contributions caused by similar paths sharing a large number of regulatory edges. Specifically, it calculates the degree of edge overlap between cross-layer regulatory paths pointing to the same disease-related node, and reduces the effective contribution of subsequent paths according to the degree of path overlap to obtain redundancy-removed path contribution data.

[0089] In practice, for the same candidate key node, cross-layer control paths pointing to the same disease-related node are divided into a path group and sorted from high to low according to the effective contribution value of the path obtained in step S43. The first ranked path is used as the benchmark control path of the current target node, and its effective contribution value is fully retained. For the paths ranked later, the control edges shared between the path and the path that is already in front of it are compared in turn.

[0090] The shared control edge is determined based on the identifiers of the nodes at both ends of the edge and the edge type. That is, only when there is a control edge connecting the same two nodes and having the same connection type in two paths is the edge identified as a common control edge. For the j-th path, its redundancy removal path contribution value is... Calculate according to the following formula:

[0091] ;

[0092] in, This represents the effective contribution value of the original path for the j-th path; and Let i and j represent the sets of control edges corresponding to the i-th path and the j-th path, respectively; the ratio of the number of common edges to the number of edges of the shorter path is used to represent the degree of overlap in the path structure. This represents the overlap suppression coefficient, which ranges from 0 to 1.

[0093] Preferably, η is 0.8; when two paths have no common control edges, the overlap is 0, and the contribution of subsequent paths is not reduced; when most of the control edges of subsequent paths are covered by existing high-contribution paths, their new contribution is reduced accordingly; for the first-ranked path, it is stipulated that its ;

[0094] The reason for using the number of shorter path edges as the normalization benchmark for the degree of overlap is that it is necessary to determine how much structure in the subsequent path has been covered by the existing path, rather than to determine the overall similarity between the two paths. For example, when a 3-edge path is completely contained in another 5-edge path, its main regulatory evidence has actually been explained by the existing path, and a judgment result of near-complete overlap should be obtained in this case.

[0095] The number of edge sets, the edge overlap ratio, and All are dimensionless quantities, therefore after redundancy removal The result obtained in step S43 Maintain consistent dimensionless properties;

[0096] This yields the redundancy removal path contribution data for each candidate key node to nodes related to different diseases.

[0097] Step S45: Disease and syndrome target contribution aggregation, used to determine the overall cross-layer impact of candidate key nodes on the set of disease and syndrome related nodes. Specifically, it involves bounded aggregation of the redundancy removal path contribution corresponding to the same disease and syndrome related node, and combining the disease and syndrome association strength to integrate different disease and syndrome related nodes to obtain the cross-layer regulation contribution value of candidate key nodes.

[0098] In practice, for a candidate key node v, the contributions of all redundancy-free paths leading to the same disease-related node t are first aggregated in a bounded manner. This aggregation method does not directly add the path contributions, but limits each new path to supplementing the remaining contribution space that has not yet been explained by existing paths, thereby preventing a large number of low-contribution paths from accumulating and forming abnormal results that exceed a reasonable range.

[0099] Subsequently, the aggregated contributions corresponding to different disease-related nodes are weighted and synthesized according to the disease-related association strength obtained in step S41, and the cross-layer regulatory contribution value of candidate key node v is calculated. The preferred method is to calculate according to the following formula:

[0100] ;

[0101] Where T represents the set of disease-related nodes; This indicates the strength of the disease-related association at node t. This represents the set of redundant paths from candidate key node v to disease-related node t. This represents the redundancy removal path contribution value of the corresponding path; for disease-related nodes that cannot be reached by candidate key nodes, the corresponding square bracket item is set to 0;

[0102] In the formula The bounded path aggregation results within the range of 0 to 1 do not interpret path contribution as a strict probability of biological event occurrence. Instead, they represent the cumulative support of multiple cross-layer regulatory paths for the same target node with a finite contribution space. When an existing path has already formed a high degree of support, the contribution space that subsequent paths can increase automatically decreases, thereby avoiding the problem that the more paths there are, the larger the contribution value will necessarily be due to simple summation.

[0103] Further adoption Weighting is applied to give target nodes that are more closely related to the target disease state a higher influence in the overall contribution calculation of candidate key nodes; denominator Used to eliminate the impact of differences in the number of disease-related nodes and the total amount of associated weights on the final numerical scale;

[0104] because and All of them are dimensionless quantities, therefore Similarly, it is a dimensionless index within the range of 0 to 1; The higher the value, the more likely the candidate key node is to reach more disease-related nodes through a reliable cross-layer control path, and the higher the correlation between the nodes it reaches and the target disease status.

[0105] At the same time, the number of disease-related nodes with valid path connections to the candidate key node is counted, which is taken as the number of disease-related nodes that the candidate key node can reach.

[0106] This yields the cross-layer regulation contribution value of candidate key nodes;

[0107] Step S46: Generation of cross-layer regulation results, which is used to form the data required for subsequent key node identification. Specifically, it records the disease-related nodes that can be reached by the candidate key nodes, the corresponding cross-layer regulation paths, the confidence of the path edges, the path level span, and the cross-layer regulation contribution value to obtain cross-layer regulation contribution data.

[0108] In specific implementation, the candidate key node identifier is used as the data index, and the cross-level regulation calculation results formed in steps S42 to S45 are merged according to the candidate nodes. For each candidate key node, firstly, its omics level and the number of different disease-related nodes that can be reached through the effective cross-level regulation path are recorded, and then the cross-level regulation contribution value of the candidate key node calculated in step S45 is recorded.

[0109] For each cross-layer control path corresponding to a candidate key node, the node identifier sequence and connection edge type are recorded according to the order of nodes in the path, and the credibility of each control edge determined in step S43 is saved according to the corresponding connection order; the path level span is recorded by the number of cross-layer control edges contained in the path, so that subsequent steps can distinguish between paths mainly formed by same-layer transmission and paths formed by continuously crossing multiple omics levels.

[0110] As a further optimization of this implementation, the effective contribution value of the path and the contribution value of the deduplicated path can be saved simultaneously as intermediate calculation fields for subsequent tracing of the source of cross-layer regulation contribution of candidate key nodes. However, the intermediate fields do not change the predetermined data composition of the cross-layer regulation contribution data.

[0111] The cross-layer regulation contribution data specifically includes candidate key node identifiers, the omics level to which the candidate key node belongs, the number of disease-related nodes reachable by the candidate key node, and the cross-layer regulation contribution value of the candidate key node; the cross-layer regulation contribution data also includes the cross-layer regulation path corresponding to the candidate key node, the edge credibility of the cross-layer regulation path, and the hierarchical span of the cross-layer regulation path;

[0112] In one specific implementation, the disease-related node set is determined based on the correlation strength between nodes in the COPD lung-kidney qi deficiency syndrome sample and pulmonary function indicators, inflammatory response indicators, oxidative stress indicators, and syndrome quantification indicators; thereby, cross-layer regulatory contribution data are obtained.

[0113] By performing the above operations, this solution addresses the technical problems in existing cross-level regulatory relationship evaluations, such as the repeated accumulation of multiple highly overlapping paths, the overestimation of contributions from excessively long indirect paths, and the ineffective differentiation of the importance of target nodes for different diseases. It creatively employs single-path contribution calculation based on regulatory edge credibility, path length, and the number of cross-levels, combined with a bounded contribution aggregation method that weights path overlap redundancy removal and disease-related correlation strength. This reasonably suppresses duplicate regulatory evidence and weak indirect regulatory relationships, improving the ability of cross-level regulatory contributions to represent the true and effective regulatory chain. For example, when a 3-sided regulatory path is essentially contained within another 5-sided path, the contribution of subsequent paths can be reduced based on the proportion of common regulatory edges, preventing the same regulatory evidence from being repeatedly amplified due to different path combinations.

[0114] Example 6, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S5, the key node identification is used to determine the final key node identification result based on the cross-layer regulation contribution and network stability of the candidate key nodes. Specifically, it involves calculating the key node score of the candidate key nodes based on the candidate key node set and the cross-layer regulation contribution data, performing perturbation verification on the multi-omics regulation network, correcting the key node score based on the ranking stability of the candidate key nodes after perturbation verification, and determining the candidate key nodes whose corrected key node scores meet the preset identification conditions as key nodes to obtain the key node identification result. The embodiment includes the following steps:

[0115] Step S51: Initial score of candidate nodes, used to determine the basic ranking of each candidate key node without network disturbance. Specifically, it is to combine the basic score of candidate key nodes and the cross-layer regulation contribution value according to a unified scale to obtain the initial key node score and initial ranking of candidate nodes.

[0116] In practice, the basic scores corresponding to the candidate key nodes in step S3 and the cross-layer regulatory contribution values ​​obtained in step S4 are read. Since the basic scores of candidate key nodes are formed by the combination of basic network attributes, omics difference strength, and phenotypic association strength, and the cross-layer regulatory contribution values ​​obtained in step S4 are limited to the range of 0 to 1, the basic scores of all candidate key nodes are first normalized according to the minimum and maximum values ​​in the current candidate set to obtain normalized basic scores within the range of 0 to 1. When all candidate key nodes have the same base score, the scores of each candidate node will be... The value is uniformly set to 0.5 to avoid a denominator of 0;

[0117] Initial critical node score of candidate critical node i Calculate according to the following formula:

[0118] ;

[0119] in, Represents the normalized baseline score of candidate key node i; This represents the cross-layer regulation contribution value obtained in step S4; Indicates the base score weight; This indicates the contribution weight of cross-level regulation and satisfies... Preferably, Take 0.4, The value is set to 0.6, which uses the original network importance of candidate nodes as the basis for identification, while also increasing the influence of actual cross-layer control capabilities on the final identification process.

[0120] , and All are dimensionless indicators, therefore there is no issue of directly adding data with different dimensions; according to The candidate key nodes are sorted from high to low to obtain the initial ranking. When two candidate nodes When they are the same, priority shall be given to... Determine the order from high to low, when They are still arranged according to the fixed coding order of the candidate node identifiers, so as to ensure that the ranking rules are consistent in subsequent repeated perturbation processes;

[0121] This yields the initial critical node scores and initial rankings for the candidate nodes.

[0122] Step S52: Hierarchical conservation network perturbation, used to generate network change samples while maintaining the basic structure of the original multi-omics network. Specifically, it involves perturbing the confidence of the control edges with a limited amplitude, and performing degree-preserving permutation on some node connections according to edge type and connection omics hierarchy to obtain multiple perturbation control networks.

[0123] In specific implementation, the original multi-omics regulation network obtained in step S2 is used as the baseline network. Under the condition of not changing the node set, the omics level to which the node belongs and the total number of various connection edges, a perturbation regulation network is generated. Each perturbation consists of two parts: a finite amplitude change in edge credibility and a permutation that maintains the degree of connection relationship, so as to simulate the fluctuation of the degree of credibility of regulation relationship and the uncertainty of local network connection, respectively.

[0124] For the original edge confidence level is The reliability of the perturbated edge e is... Determined according to the following formula:

[0125] ;

[0126] Where b represents the b-th network disturbance; This indicates the proportion of random perturbation corresponding to the current connected edge; δ indicates that the value is uniformly taken between -δ and δ; clip indicates that the calculation result is restricted to the range of 0 to 1; preferably, δ is 0.10, so that the single change of the edge confidence relative to the original value does not exceed 10%, which is used to simulate detection error, statistical estimation error and external confidence score fluctuation, without changing the basic confidence level of the original connection relationship;

[0127] After completing the edge credibility perturbation, the node connection data is divided into intra-layer related edges and regulatory edges of different cross-layer types according to the connection structure. Among them, cross-layer regulatory edges are further processed according to three levels: gene to protein, protein to metabolite, and metabolite to phenotypic state. Different types of cross-layer connections are prohibited from being interchanged.

[0128] Preferably, 10% of the number of connecting edges of each type are randomly selected from the edge set to maintain the permutation; for cross-layer directed edges... and Try replacing with and The replacement is accepted only if the replaced starting node still belongs to the original starting omics level, the target node still belongs to the original target omics level, and there are no self-connections or duplicate connections. This operation keeps the number of in-degrees and out-degrees of the nodes participating in the replacement unchanged, so the number of node connections will not be artificially increased or decreased due to the perturbation process.

[0129] For undirected associated edges in the same layer, the corresponding bilateral exchange method is used for replacement, while keeping the degree of the participating nodes unchanged; when a random selection does not meet the replacement conditions, the current combination is discarded and the connecting edges are re-extracted until the preset proportion of effective replacements is completed or the preset maximum number of attempts is reached; preferably, the maximum number of attempts is set to 20 times the number of target replacement edges to avoid infinite loops caused by insufficient replacement combinations in high-density local networks;

[0130] Preferably, the above process is repeated to generate 200 perturbation control networks; each perturbation control network maintains the original candidate node set, the omics level to which the nodes belong, and the basic scale of various connection structures, and only changes the edge credibility and specific connection objects within a limited range, so that the subsequent stability evaluation reflects reasonable network uncertainty, rather than the score change caused by the destruction of the overall network structure;

[0131] This results in multiple disturbance control networks;

[0132] Step S53: Repeated scoring of perturbation network is used to observe the identification results of candidate key nodes under the condition of local network change. Specifically, the cross-layer control contribution and key node score of candidate key nodes are recalculated in each perturbation control network to obtain the perturbation ranking sequence of candidate nodes.

[0133] In practical implementation, the candidate key node set determined in step S3 is used as a fixed evaluation object, and candidate node screening is not re-executed after each network perturbation to avoid interference from changes in the candidate set on the ranking comparison between different perturbation results. For the b-th perturbation control network, the basic network attributes of the candidate nodes affected by the network structure are recalculated according to the same basic scoring rules in step S3, while keeping the omics difference strength and phenotypic association strength of the corresponding nodes unchanged. Thus, the basic scores of the candidate nodes under the current perturbation network are obtained again, and normalized in the same way as in step S51. ;

[0134] Subsequently, the original multi-omics control network in step S4 is replaced with the current perturbation control network. While keeping the set of disease-related nodes, the upper limit of path length, and the calculation parameters in steps S43 to S45 unchanged, the constrained cross-layer path search, single-path credible contribution calculation, path redundancy removal, and disease-target contribution aggregation are re-executed to obtain the cross-layer control contribution value of candidate key node i in the b-th perturbation control network. ;

[0135] Key node score of candidate key node i in the b-th perturbation Calculate according to the following formula:

[0136] ;

[0137] in, and The same parameters as in step S51 are used to ensure consistency in the scoring rules between the original network and each perturbation network; according to Sort the candidate nodes from highest to lowest, and use the same tie-breaking rules as in step S51 to obtain the perturbation ranking of the candidate nodes. ;

[0138] When a total of 200 perturbation control networks are generated, each candidate key node forms a ranking sequence consisting of 200 perturbation rankings, which is used to describe the change in the relative importance of the candidate node under the condition of local network structure change.

[0139] This yields the candidate node perturbation ranking sequence;

[0140] Step S54: Ranking robustness calculation, used to quantify the sensitivity of candidate key nodes to network disturbances. Specifically, it calculates the robustness score based on the average deviation of the candidate node's disturbance ranking relative to the initial ranking and the retention ratio within the allowed ranking range, thus obtaining the candidate node robustness data.

[0141] In practical implementation, let the number of candidate key nodes be N, and the number of disturbance control networks be B; for candidate key node i, the disturbance rankings obtained in step S53 will be... Compare with the initial ranking obtained in step S51 The ranking shift caused by a single perturbation is represented by the absolute value of the ranking difference.

[0142] To simultaneously evaluate the overall drift of candidate node rankings and their ability to maintain their initial rankings, a pre-defined allowable ranking offset range Δr is set. Preferably, when the number of candidate nodes is N, Δr is determined to be the larger of 5% of N rounded up and 2. That is, when the number of candidate nodes is small, local changes between the two rankings are allowed, and as the number of candidate nodes increases, the allowable range is expanded synchronously according to the candidate size.

[0143] Robustness score of candidate key node i Calculate according to the following formula:

[0144] ;

[0145] The first term represents the degree of preservation of the normalized average ranking; the second term represents the proportion of the ranking that remains within the allowable deviation range after perturbation; I(·) represents the indicator function, which takes 1 when the condition in parentheses is met, and 0 otherwise; the two parts are combined with equal weights so that the robustness score is neither completely masked by the average value due to a few large ranking jumps, nor judged solely based on whether a fixed ranking boundary is crossed.

[0146] When the number of candidate key nodes N is 1, it is... Set it directly to 1; for cases where N is greater than 1, The value ranges from 0 to 1. The closer it is to 1, the more closely the candidate key node can maintain its identification position in the original network under multiple finite network perturbations;

[0147] This yields robustness data for candidate nodes, including candidate node identifiers, initial rankings, perturbation ranking sequences, and robustness scores.

[0148] Step S55: Key node score correction, used to reduce the identification priority of high-scoring nodes that rely on accidental network structure. Specifically, the initial key node score is reduced according to the robustness score of the candidate node, and the key node set and ranking are determined according to the corrected key node score.

[0149] In practical implementation, a one-to-one correspondence is established between the initial critical node score obtained in step S51 and the robustness score obtained in step S54, and the robustness score is used to correct the retention ratio of the initial score; the corrected critical node score of candidate critical node i Calculate according to the following formula:

[0150] ;

[0151] in, This represents the initial key node score of the candidate key node in the original multi-omics regulatory network; This indicates its robustness score during the perturbation verification process; This indicates the corrected critical node score;

[0152] The meaning of this correction method is that high scores in the original network are only fully preserved if they can be stably maintained under reasonable perturbation conditions; when the high scores of candidate nodes mainly depend on a small number of specific connections, their perturbation ranking fluctuations increase, corresponding to... Lower, make Synchronization is reduced, thereby minimizing the impact of accidental network structures on the final key node identification results;

[0153] Preferably, the corrected critical node score is among the top 10% of all candidate critical nodes, and the robustness score is... A score not less than 0.75 is set as the preset recognition condition; when the number of candidate nodes is less than 10, the node with the highest corrected score is selected. Nodes with a score of 0.75 or higher are considered as candidates that meet the scoring and ranking criteria; for nodes that meet the preset recognition criteria, they are ranked according to... Sort the nodes from highest to lowest to form a set of key nodes and a ranking of key nodes;

[0154] This determines the set and order of critical nodes after network stability correction;

[0155] Step S56: Key node verification information is generated to determine the subsequent verification order of key nodes. Specifically, the verification priority is determined based on the cross-layer regulation contribution and robustness score of key nodes, and the corresponding main cross-layer regulation paths are extracted to obtain the key node identification results.

[0156] In specific implementation, for each key node determined in step S55, its corresponding cross-layer control path is extracted from the cross-layer control contribution data formed in step S4, and arranged from high to low according to the redundancy removal path contribution value obtained in step S44; starting from the path with the highest contribution value, the redundancy removal contribution is accumulated sequentially, and when the cumulative redundancy removal contribution corresponding to the selected path reaches 80% of the total redundancy removal contribution of all effective paths of the key node, the selection is stopped, and the selected path is determined as the main cross-layer control path of the key node; preferably, the number of main cross-layer control paths retained for each key node does not exceed 5, and when the cumulative contribution of the current 5 paths has not reached 80%, the 5 with the highest contribution value are still used as the main cross-layer control paths to control the final verification range;

[0157] To determine the order of subsequent experimental verification for different key nodes, the cross-layer regulatory contribution values ​​obtained in step S4 will be used. Robustness score obtained in step S54 Together, they serve as the basis for verification, with the verification priority of key node i. Calculate according to the following formula:

[0158] ;

[0159] in, This represents the contribution value of cross-level regulation at key nodes; Indicates the robustness score of key nodes; This indicates the priority score for critical node verification; due to and All are dimensionless indices in the range of 0 to 1, therefore Similarly, it is a dimensionless indicator; the product method can simultaneously constrain cross-layer control capability and network stability, so that nodes with high cross-layer contribution but unstable after disturbance, as well as nodes with stable network ranking but weak cross-layer control contribution, will not receive too high verification priority.

[0160] according to The key node verification priority is determined from high to low, and the key node identifier, its corresponding omics level and ranking information are associated and saved. At the same time, the cross-layer regulation contribution value, robustness score and main cross-layer regulation path of the key node are recorded, thus forming the key node identification result.

[0161] The key node identification results specifically include a set of key nodes, a ranking of key nodes, the omics level to which the key nodes belong, and the cross-layer regulatory contribution value of the key nodes; the key node identification results also include a key node robustness score, the main cross-layer regulatory path corresponding to the key node, and the key node verification priority; wherein, the key node verification priority is determined jointly based on the key node cross-layer regulatory contribution value and the robustness score; thus, the key node identification results are obtained.

[0162] By performing the above operations, this solution addresses the technical problem in existing key node rankings where high node scores may depend on a small number of specific connections, and rankings fluctuate significantly after slight changes in network structure, leading to insufficient stability of the screening results. This solution creatively employs hierarchical conservation network perturbation, perturbation network repetition scoring, and ranking robustness correction. While maintaining the basic structure of the omics hierarchy and connection scale, it simulates changes in edge credibility and local connectivity, and uses perturbation ranking stability to correct key node scores. This reduces the impact of accidental network structure on the final identification results and improves the robustness of key node screening and subsequent verification order. For example, when a candidate node's high score mainly depends on a small number of specific control edges, its ranking is prone to significant shifts after edge credibility perturbation and degree-preserving connection replacement, resulting in a decrease in its robustness score and a corresponding decrease in the corrected key node score.

[0163] Example 7, see Figure 1 and Figure 2 Based on the above embodiments, this embodiment provides a key node identification system for multi-omics regulatory networks, including a multi-omics data standardization processing module, a multi-omics regulatory network construction module, a candidate key node initial screening module, an omics-level cross-layer regulation module, and a key node identification module.

[0164] The multi-omics data standardization processing module is used to uniformly organize multi-omics data from different sources, with different detection scales and different data structures. Through multi-omics data standardization processing, multi-omics standardized data is obtained, and the multi-omics standardized data is transmitted to the multi-omics regulatory network construction module.

[0165] The multi-omics regulation network construction module is used to establish node association relationships between different omics levels. Through the construction of the multi-omics regulation network, a multi-omics regulation network is obtained, and the multi-omics regulation network is transmitted to the candidate key node screening module and the omics level cross-level regulation module.

[0166] The candidate key node screening module is used to screen candidate nodes with further computational value from the multi-omics regulation network. Through the candidate key node screening, a set of candidate key nodes is obtained, and the set of candidate key nodes is transmitted to the omics-level cross-layer regulation module.

[0167] The omics-level cross-layer regulation module is used to calculate the effective contribution of candidate key nodes to the transmission of regulatory influence to disease-related nodes. Through omics-level cross-layer regulation, cross-layer regulation contribution data is obtained, and the cross-layer regulation contribution data is transmitted to the key node identification module.

[0168] The critical node identification module is used to determine the final critical node based on the cross-layer regulation contribution and network stability of the candidate critical nodes, and to obtain the critical node identification result through critical node identification.

[0169] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0170] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0171] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for identifying key nodes in multi-omics regulatory networks, characterized in that: The method includes the following steps: Step S1: Standardize the multi-omics data to obtain multi-omics standardized data; Step S2: Construction of multi-omics regulatory network, identifying gene nodes, protein nodes, metabolite nodes and phenotype nodes, and establishing intra-layer association edges and cross-layer regulatory edges between nodes to obtain the multi-omics regulatory network; Step S3: Initial screening of candidate key nodes. Based on the basic network attributes, omics difference strength and phenotypic association strength of each node in the multi-omics regulatory network, a basic score is given to each node, and nodes whose basic scores meet the preset initial screening conditions are determined as candidate key nodes, thus obtaining a set of candidate key nodes. Step S4: Cross-level regulation at the omics level. Based on the multi-omics regulation network, determine the set of disease-related nodes. Starting from the candidate key nodes in the candidate key node set, search for cross-level regulation paths between the candidate key nodes and the set of disease-related nodes. Based on the confidence of the regulation edges, path length, and number of omics levels crossed in the cross-level regulation path, combined with the overlap between the cross-level regulation paths and the disease-related association strength of the disease-related nodes, calculate the cross-level regulation contribution of the candidate key nodes to obtain the cross-level regulation contribution data. Step S5: Key node identification. Based on the candidate key node set and the cross-layer regulation contribution data, calculate the key node score of the candidate key nodes, and perform perturbation verification on the multi-omics regulation network. Correct the key node score based on the ranking stability of the candidate key nodes after perturbation verification. Determine the candidate key nodes whose corrected key node scores meet the preset identification conditions as key nodes, and obtain the key node identification result.

2. The method for identifying key nodes in a multi-omics regulatory network according to claim 1, characterized in that: In step S1, the multi-omics standardized data specifically includes transcriptomic standardized data, proteomic standardized data, metabolomic standardized data, and phenotypic state standardized data.

3. The method for identifying key nodes in a multi-omics regulatory network according to claim 2, characterized in that: In step S2, the multi-omics regulatory network specifically includes omics node data and node connection data; the omics node data includes gene nodes, protein nodes, metabolite nodes and phenotype nodes; the node connection data includes intra-layer association edges, gene-to-protein cross-layer mapping edges, protein-to-metabolite functional association edges and metabolite-to-phenotype state phenotypic association edges.

4. The method for identifying key nodes in a multi-omics regulatory network according to claim 3, characterized in that: In step S3, the candidate key node set specifically includes candidate node identifiers, candidate node omics level, candidate node basic network attributes, and candidate node state association data; the candidate node state association data includes candidate node omics difference strength and candidate node phenotypic association strength; wherein, the basic network attributes include node degree, betweenness centrality, proximity centrality, and eigenvector centrality.

5. The method for identifying key nodes in a multi-omics regulatory network according to claim 4, characterized in that: In step S4, the omics-level cross-layer regulation includes the following steps: Step S41: Determine disease-related nodes. Filter nodes that meet the disease-related conditions based on the association between multi-omics nodes and phenotypic status, and record the corresponding disease-related strength to obtain a set of disease-related nodes. Step S42: Constrained cross-layer path search. Starting from the candidate key node, search for cross-layer regulation paths according to the omics hierarchy direction, path length constraints, and node non-repetition conditions to obtain a set of candidate cross-layer regulation paths. Step S43: Single path credibility contribution calculation. The basic credibility of the path is determined based on the credibility of each control edge in the path, and attenuation processing is performed according to the path length and the number of omics levels crossed to obtain the effective contribution value of the path. Step S44: Redundancy removal of overlapping path contributions. Calculate the degree of edge overlap between cross-layer regulatory paths pointing to the same disease-related node, and reduce the effective contribution of subsequent paths according to the degree of path overlap to obtain redundancy-removed path contribution data. Step S45: Aggregate the contribution of disease and syndrome targets. Perform bounded aggregation of the redundancy removal path contributions corresponding to the same disease and syndrome related nodes, and combine the disease and syndrome association strength to integrate the cross-layer regulation contribution values ​​of candidate key nodes. Step S46: Generate cross-layer regulation results. Record the disease-related nodes reachable by the candidate key nodes, the corresponding cross-layer regulation paths, the confidence level of the path edges, the path level span, and the cross-layer regulation contribution value to obtain cross-layer regulation contribution data.

6. The method for identifying key nodes in a multi-omics regulatory network according to claim 5, characterized in that: In step S4, the cross-layer regulatory contribution data specifically includes the candidate key node identifier, the omics level to which the candidate key node belongs, the number of disease-related nodes that the candidate key node can reach, and the cross-layer regulatory contribution value of the candidate key node.

7. The method for identifying key nodes in a multi-omics regulatory network according to claim 6, characterized in that: In step S5, the identification of key nodes includes the following steps: Step S51: Initial scoring of candidate nodes. Based on the basic scores of candidate key nodes and the cross-layer regulation contribution values, a unified scale combination is performed to obtain the initial key node scores and initial rankings of candidate nodes. Step S52: Hierarchical conservation network perturbation, using a limited amplitude perturbation on the confidence of the control edges, and performing degree-preserving permutation on some node connections according to edge type and connection omics hierarchy to obtain multiple perturbation control networks; Step S53: Repeated scoring of perturbation networks. In each perturbation control network, the cross-layer control contribution and key node score of the candidate key nodes are recalculated to obtain the perturbation ranking sequence of candidate nodes. Step S54: Calculate the robustness of the ranking. Calculate the robustness score based on the average deviation of the candidate node's perturbation ranking relative to the initial ranking and the retention ratio within the allowed ranking range to obtain the candidate node robustness data. Step S55: Critical node score correction. The initial critical node score is reduced based on the robustness score of the candidate nodes, and the critical node set and ranking are determined based on the corrected critical node score. Step S56: Generate key node verification information. Determine the verification priority based on the cross-layer regulation contribution and robustness score of key nodes, and extract the corresponding main cross-layer regulation paths to obtain the key node identification results.

8. A key node identification system for multi-omics regulatory networks, used to implement the key node identification method for multi-omics regulatory networks as described in any one of claims 1-7, characterized in that: It includes a multi-omics data standardization processing module, a multi-omics regulatory network construction module, a candidate key node initial screening module, an omics-level cross-layer regulation module, and a key node identification module.

9. A key node identification system for multi-omics regulatory networks according to claim 8, characterized in that: The multi-omics data standardization processing module is used to uniformly organize multi-omics data from different sources, with different detection scales and different data structures. Through multi-omics data standardization processing, multi-omics standardized data is obtained, and the multi-omics standardized data is transmitted to the multi-omics regulatory network construction module. The multi-omics regulation network construction module is used to establish node association relationships between different omics levels. Through the construction of the multi-omics regulation network, a multi-omics regulation network is obtained, and the multi-omics regulation network is transmitted to the candidate key node screening module and the omics level cross-level regulation module. The candidate key node screening module is used to screen candidate nodes with further computational value from the multi-omics regulation network. Through the candidate key node screening, a set of candidate key nodes is obtained, and the set of candidate key nodes is transmitted to the omics-level cross-layer regulation module. The omics-level cross-layer regulation module is used to calculate the effective contribution of candidate key nodes to the transmission of regulatory influence to disease-related nodes. Through omics-level cross-layer regulation, cross-layer regulation contribution data is obtained, and the cross-layer regulation contribution data is transmitted to the key node identification module. The critical node identification module is used to determine the final critical node based on the cross-layer regulation contribution and network stability of the candidate critical nodes, and to obtain the critical node identification result through critical node identification.

Images