Cross-scene adaptive multi-omics knowledge graph construction method and system

Abstract

The invention provides a cross-scene adaptive multi-omics knowledge graph construction method and system, and the method comprises the steps: collecting and preprocessing multi-omics data related to a target research object, and obtaining standardized data comprising entities and association information between the entities; based on the standardized data, performing matching and duplicate removal on the entities by using a multi-omics entity semantic-feature fusion matching algorithm to generate a standardized entity set; performing credibility quantification on association information among entities in the standardized entity set through an omics specific relation weight calculation algorithm, and screening a high credibility relation set; constructing an initial graph structure by taking the standardized entity set as nodes and the high-credibility relation set as edges, and performing feature representation learning on the initial graph structure by adopting a multi-omics enhanced node embedding algorithm to obtain node vector representation; and constructing a knowledge graph based on the node vector representation and the high-credibility relationship set, performing accuracy verification on the knowledge graph, and dynamically updating the knowledge graph according to a verification result and the multi-omics data change.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of bioinformatics and knowledge graph construction, and in particular to a method and system for constructing multi-omics knowledge graphs that are adaptable to different scenarios. Background Technology

[0002] Phenotypic characteristics of complex diseases and biological traits in life science research are often the result of the combined effects of polygenic inheritance and environmental factors. Their underlying mechanisms involve synergistic effects at multiple levels, including gene expression regulation, protein-protein interactions, and metabolic network imbalances. Currently, a large amount of multi-omics data, including genomics, transcriptomics, proteomics, and metabolomics, has been accumulated and stored in authoritative databases such as NCBI, GEO, Uniprot, HMDB, TAIR, and MGnify. However, these data exhibit heterogeneous formats and hidden correlations, making them difficult to use directly for systematic mechanism analysis. Furthermore, existing multi-omics knowledge graph construction technologies suffer from three core shortcomings: First, the integration of multi-omics data is low, with different omics data being simply stacked as independent modules without achieving collaborative feature mining. For example, early multi-omics analysis tools like MetaboAnalyst only matched differentially expressed genes in the transcriptome with differentially expressed metabolites in the metabolome, without integrating proteomics interaction information, failing to reveal the mediating role of key intermediate proteins, and resulting in insufficient mechanistic explanatory power of the maps.

[0003] Second, entity matching algorithms rely on single semantic features and ignore the inherent correlations at the omics feature level, leading to misidentification or missed identification of homologous entities. For example, the AutoSchemaKG framework uses the Word2Vec model to extract semantic features of entities to achieve cross-database entity matching, but it does not integrate the correlations at the omics feature level, resulting in a high matching error rate.

[0004] Third, the calculation of relation weights fails to quantify the contribution and reliability of each omics data, resulting in insufficient credibility of biological associations. For example, the POINT platform integrates multi-omics networks and knowledge graphs, but relation weights are mostly based on document frequency and fixed thresholds, failing to quantify the contribution and database reliability of each omics data.

[0005] The aforementioned shortcomings have led to existing knowledge graph constructions generally suffering from problems such as "incomplete structure, unreliable associations, and poor scene adaptability," making it difficult to support in-depth mechanism research and cross-scene applications. There is an urgent need for a knowledge graph construction technology that combines the ability to deeply integrate multiple omics, algorithmic accuracy, and scene universality. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for constructing multi-omics knowledge graphs that are adaptable to different scenarios, in order to solve the above-mentioned problems in the prior art.

[0007] This invention provides a method for constructing a multi-omics knowledge graph that adapts to different scenarios, including: S1: Collect multi-omics data related to the target research object, and preprocess the multi-omics data to obtain standardized data including entity and inter-entity relationship information; S2: Based on the standardized data, a multi-omics entity semantic-feature fusion matching algorithm is used to match and deduplicate entities from different sources to generate a standardized entity set; S3: The credibility of the association information between entities in the standardized entity set is quantified by the omics-specific relation weight calculation algorithm, and a high credibility relation set is selected based on the quantification results; S4: Using the standardized entity set as nodes and the high-confidence relation set as edges, construct an initial graph structure, and use a multi-omics-enhanced node embedding algorithm to learn feature representations of the initial graph structure to obtain node vector representations; and S5: Construct a knowledge graph based on the node vector representation and the high-confidence relation set, verify the accuracy of the knowledge graph, and dynamically update the knowledge graph according to the verification results and changes in multi-omics data.

[0008] This invention provides a cross-scenario adapted multi-omics knowledge graph construction system, including: The data processing module is used to collect multi-omics data related to the target research object and preprocess the multi-omics data to obtain standardized data including entity and inter-entity relationship information. The entity fusion module is used to match and deduplicate entities from different sources based on the standardized data using a multi-omics entity semantic-feature fusion matching algorithm, and generate a standardized entity set. The association quantification module is used to quantify the credibility of the association information between entities in the standardized entity set through an omics-specific relationship weight calculation algorithm, and to select a set of highly credible relationships based on the quantification results. The graph representation learning module is used to construct an initial graph structure using the standardized entity set as nodes and the high-confidence relation set as edges, and to perform feature representation learning on the initial graph structure using a multi-omics enhanced node embedding algorithm to obtain node vector representations; and The knowledge graph construction module is used to construct a knowledge graph based on the node vector representation and the high-confidence relation set, verify the accuracy of the knowledge graph, and dynamically update the knowledge graph according to the verification results and changes in multi-omics data.

[0009] This invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-described cross-scenario adapted multi-omics knowledge graph construction method.

[0010] This invention also provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, implements the steps of the above-described cross-scenario adapted multi-omics knowledge graph construction method.

[0011] The embodiments of the present invention can include the following beneficial effects: Through the innovative design of cross-scenario algorithm decoupling and multi-omics feature fusion, the embodiments of the present invention simultaneously solve the three major problems of scenario limitation, low matching accuracy and low association reliability, construct a multi-omics knowledge graph with complete structure, reliable association and adaptability to various life science research scenarios, and provide core tools for basic research and industrial applications. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a cross-scenario adapted multi-omics knowledge graph construction method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the overall technical framework of an embodiment of the present invention; Figure 3 This is a flowchart of entity recognition and matching according to an embodiment of the present invention; Figure 4 This is a flowchart of the relationship weight calculation according to an embodiment of the present invention; Figure 5 This is a flowchart of the node embedding optimization process according to an embodiment of the present invention; Figure 6 This is a schematic diagram of a cross-scenario adapted multi-omics knowledge graph construction system according to an embodiment of the present invention. Detailed Implementation

[0014] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification 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 this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0015] Method Implementation Examples According to embodiments of the present invention, a method for constructing a multi-omics knowledge graph that adapts to different scenarios is provided. Figure 1 This is a flowchart of a cross-scenario adapted multi-omics knowledge graph construction method according to an embodiment of the present invention, such as... Figure 1 As shown, the cross-scenario adaptation multi-omics knowledge graph construction method according to an embodiment of the present invention specifically includes: Step S101: Collect multi-omics data related to the target research object, and preprocess the multi-omics data to obtain standardized data including entity and inter-entity relationship information; The multi-omics data includes genomics, transcriptomics, proteomics, and metabolomics; the preprocessing includes data cleaning, missing value handling, and format standardization; The data cleaning includes removing data with a missing rate higher than a preset threshold; The standardization of the format includes standardizing transcriptomics and metabolomics data, as well as performing homology comparisons of protein sequence data using the BLAST tool.

[0016] Step S102, based on the standardized data, uses a multi-omics entity semantic-feature fusion matching algorithm to match and deduplicate entities from different sources, generating a standardized entity set, specifically including: Calculate the semantic similarity between entities to be matched from different data sources; Calculate the omics feature similarity of the entities to be matched across multiple omics dimensions; Based on the distribution characteristics of each omics data, the feature weights of multiple omics feature similarities in the comprehensive matching calculation are dynamically determined using Formula 1 and Formula 2. Formula 1; Formula 2; in, This represents the information entropy of the k-th group of learning data. Let represent the probability distribution of the i-th feature in the k-th omics, and n represent the total number of features in that omics. Represents the feature weights of the k-th group; Based on the semantic similarity, omics feature similarity, and feature weights, the comprehensive matching similarity of the entities to be matched is calculated using Formula 3. Formula 3; in, Representing entities and The overall matching similarity, Represents semantic weight coefficients. Indicates the semantic similarity between entities. These correspond to genomics, transcriptomics, proteomics, and metabolomics, respectively. Let be the similarity of omics features of the k-th omics; The comprehensive matching similarity is compared with a preset matching threshold, and entities that are greater than or equal to the preset matching threshold are identified as homologous entities and merged and deduplicated to form a standardized entity set.

[0017] Step S103 involves using an omics-specific relation weighting algorithm to quantify the reliability of the association information between entities in the standardized entity set, and then selecting a high-reliability relation set based on the quantification results. Specifically, this includes: Obtain association strength data that characterizes the strength of association between entities in the standardized entity set; The data reliability contribution weight of each omics data is determined based on the accuracy of each omics data. The reliability coefficients of the databases from which each omics data originates are obtained based on the database impact factor and data update frequency. Based on the aforementioned association strength data, data reliability contribution weight, and reliability coefficient, Formula 4 is used to calculate the comprehensive weight value of the relationships between corresponding entities in the standardized entity set. Formula 4; in, Representing entities and Interrelationship The weight, Represents the 0-1 normalization function. This represents the contribution weight of the k-th group. This represents the association strength of the k-th group. This represents the reliability coefficient of the database from which the k-th omics data originates; The comprehensive weight value is compared with a preset confidence threshold, and relationships that are greater than or equal to the preset confidence threshold are selected to form a high-confidence relationship set.

[0018] Step S104: Using the standardized entity set as nodes and the high-confidence relation set as edges, construct an initial graph structure, and use a multi-omics enhanced node embedding algorithm to learn the feature representation of the initial graph structure to obtain node vector representations, specifically including: The neighboring nodes of each node in the initial graph structure are grouped according to their respective omics types; Calculate the average weight value of the relationship between neighbor nodes and the central node for each omics type; The aggregation weight of each group of omics type neighbor nodes in the information aggregation process is dynamically determined based on the average weight value; wherein, the aggregation weight is positively correlated with the corresponding average weight value; Based on the aggregation weight, the features of neighbor nodes of each omics type are weighted and aggregated. The aggregated features of neighbor nodes are then fused with the features of the central node, and a nonlinear transformation is performed through a neural network layer to obtain the vector representation of the node in the current layer. By stacking multiple layers of a neural network and iteratively updating them, the final node vector representation used for downstream tasks is obtained.

[0019] Step S105 involves constructing a knowledge graph based on the node vector representation and the high-confidence relation set, verifying the accuracy of the knowledge graph, and dynamically updating the knowledge graph based on the verification results and changes in multi-omics data. Specifically, this includes: The entities in the standardized entity set, the node vector representations, and the relations and weights in the high-confidence relation set are imported into a graph database to construct a multi-omics knowledge graph. The multi-omics knowledge graph is matched with at least one known standard biological pathway, and at least one preset threshold or algorithm parameter in the graph construction process is optimized and adjusted based on the matching rate results. When new multi-omics data is available or the data in the graph database is updated, steps S101-S105 are repeated to dynamically iterate and update the multi-omics knowledge graph.

[0020] The following details the cross-scenario adaptation method for constructing multi-omics knowledge graphs according to embodiments of the present invention, such as... Figure 2 As shown, the above technical solutions of the embodiments of the present invention will be described in detail.

[0021] This invention proposes a general knowledge graph construction method integrating multi-omics data, including genomics, transcriptomics, proteomics, and metabolomics. This method can be widely applied to various life science research scenarios, such as research on the pathogenesis of complex diseases, biomarker screening, personalized treatment formulation, agricultural biotechnology, and microbial metabolic regulation analysis. Specifically, this invention provides a cross-scenario adapted multi-omics integrated knowledge graph construction method, including the following steps: 1. Data Acquisition and Preprocessing: Collect multi-omics data related to the target research object from life science databases corresponding to genomics, transcriptomics, proteomics, and metabolomics. Clean, standardize, and unify the format of the data, and extract entities and preliminary association information. Entities include genes, mRNA, proteins, metabolites, and phenotypes. Target research objects include complex diseases, biological traits, microbial metabolic systems, or plant stress resistance systems. 2. Multi-omics entity integration: Based on the multi-omics entity semantic-feature fusion matching algorithm, the comprehensive matching similarity between entities in different databases is calculated, and the matched entities are deduplicated and standardized; the matching algorithm simultaneously integrates semantic similarity and feature similarity of each omics, and determines the weight of each omics feature through the entropy method. 3. Relationship weight calculation: Based on the omics-specific relationship weight calculation algorithm, the credibility of biological associations between entities is quantified and high-credibility relationships are screened; the weight calculation algorithm integrates the association strength of each omics, data accuracy and database reliability coefficient, and determines the contribution weight of each omics through accuracy. 4. Node Embedding Optimization: Based on a multi-omics enhanced node embedding algorithm, omics-specific information is incorporated to learn feature representations of graph nodes and optimize the graph structure; the node embedding algorithm is an improved GraphSAGE algorithm, which divides neighboring nodes according to omics type and dynamically assigns aggregation weights; 5. Knowledge Graph Construction and Update: Construct a multi-omics knowledge graph, define node attributes and relationship attributes, regularly synchronize database data to achieve dynamic updates, and verify the accuracy of the graph by matching it with known biological pathways.

[0022] In the multi-omics knowledge graph construction method of this invention, the core algorithm is decoupled from the target research object and data type. It can be quickly adapted to new research scenarios by simply replacing the multi-omics data source, without modifying the core logic of the algorithm.

[0023] The core technical solutions of this invention include data acquisition and standardization processing, entity recognition and matching, relationship weight calculation, node embedding optimization, and knowledge graph construction and updating, as detailed below: I. Overall Technical Framework 1. Multi-omics data collection: Multi-omics data on the target research subjects are collected from authoritative databases such as NCBI, GEO, Uniprot, HMDB, TAIR, and MGnify. This covers genomics (SNP sites, gene sequences, etc.); transcriptomics (mRNA expression profiles, differential expression data, etc.); proteomics (protein sequences, interactions, etc.); and metabolomics (metabolite structure, concentration data, etc.).

[0024] 2. Data Standardization and Screening: The collected multi-omics data were cleaned and standardized. Missing value handling: Data with a missing value rate >20% were removed; Format standardization: Transcriptomics and metabolomics data were standardized using Z-score, and protein sequences were compared for homology using BLAST; Data screening: High-reliability data were retained.

[0025] 3. Entity Recognition and Matching (Algorithm 1: Multi-omics Entity Semantic-Feature Fusion Matching Algorithm): Based on the multi-omics entity semantic-feature fusion matching algorithm, this algorithm achieves deduplication and standardization of entities across databases. Entity Recognition: BERT+CRF model is used to extract entities such as genes, mRNA, proteins, and metabolites; Multi-dimensional Matching: Semantic similarity and omics feature similarity are fused; Matching Decision: Entities with a comprehensive similarity ≥ 0.8 are considered homologous and standardized.

[0026] 4. Relationship Weight Calculation (Algorithm 2: Omics-Specific Relationship Weight Calculation Algorithm): Based on the omics-specific relationship weight calculation algorithm, the credibility of biological associations between entities is quantified. Multi-source association data integration: Collect the association strength of each omics; Dynamic weight allocation: Combine the accuracy of omics data with the database reliability coefficient to calculate the omics contribution weight. High-confidence relation screening: Relationships with a weight ≥ 0.6 are considered high-confidence and retained for graph construction.

[0027] 5. Node Embedding Optimization (Algorithm 3: Multi-omics Enhanced Node Embedding Algorithm): Based on the multi-omics enhanced node embedding algorithm, optimize the feature representation of graph nodes. Neighbor Grouping: Divide node neighbors according to omics type; Dynamic Aggregation Weight: Assign aggregation weights based on the average relation weight of omics neighbors. Embedding training: LeakyReLU activation function is used to learn node embedding vectors specific to fusion omics.

[0028] 6. Knowledge Graph Construction and Dynamic Updates. Graph Construction: Based on the Neo4j graph database, define node attributes and relationship attributes to construct a multi-omics knowledge graph; Dynamic Updates: Periodically synchronize database updates and repeat steps 1-5 above to achieve graph iteration; Reliability Verification: Verify the accuracy of the graph by checking its matching rate with known biological pathways.

[0029] II. Core Algorithm Design 1. Multi-omics entity semantic-feature fusion matching algorithm This algorithm addresses the problem of heterogeneous entities across different databases, while also considering semantic consistency and correlation of omics features, making it compatible with all multi-omics data types.

[0030] The core formula is as follows: ; ; ; In the formula, Representing entities and The overall matching similarity score ranges from 0 to 1, and a score ≥ 0.8 is considered a match. This represents the semantic weight coefficient, determined through 5-fold cross-validation, with a value ranging from 0.3 to 0.5. For entity semantic similarity, cosine similarity of word vectors is calculated based on the biomedical dictionary Word2Vec. These correspond to genomics, transcriptomics, proteomics, and metabolomics, respectively. The feature weights of the k-th group are determined by the entropy method. Calculations reflect the dispersion and discriminative power of data, adapting to the omics data characteristics of different research subjects; For the k-th omics, the feature similarity is: SNP sequence similarity for genomics, Pearson correlation coefficient for expression profiles for transcriptomics, sequence homology for proteomics, and structural fingerprint similarity for metabolomics. The information entropy of the k-th group of learning data, Let be the probability distribution of the i-th feature in the k-th omics, and n be the total number of features in that omics.

[0031] 2. Algorithm for calculating weights of omics-specific relationships This algorithm quantifies the credibility of biological associations between entities, integrates the strength of associations across different omics, data reliability, and database quality, avoids bias from a single data source, and is applicable to the evaluation of various multi-omics associations.

[0032] The core formula is as follows: ; ; ; In the formula, For entities and Interrelationship The weight of the relation is between 0 and 1, and a value of ≥0.6 is considered a highly reliable relation. The function is a 0-1 normalization function, standardized using min-max. The contribution weight of the k-th group is determined by the accuracy. This confirmation demonstrates the reliability of the omics data; For the association strength of the k-th omics, genomics uses the GWAS association score. Transcriptomics uses fold change (FC) normalized value; proteomics uses interaction confidence; metabolomics uses concentration correlation coefficient. The reliability coefficient of the database from which the k-th omics data originates is assigned based on the database impact factor and data update frequency, and is between 0.7 and 1.0. Let be the precision of the k-th group of data. The number of true positive associations. This represents the number of false positive associations.

[0033] 3. Multi-omics Enhanced Node Embedding Algorithm This algorithm optimizes the feature representation of graph nodes, incorporates omics-specific information, improves the accuracy of downstream association prediction, mechanism analysis and other tasks, and adapts to the downstream application needs of different research scenarios.

[0034] The core formula is as follows: ; ; ; In the formula, Let v be the embedding vector of node v in the l-th layer of the network. Let v be the embedding vector of node v in the (l-1)th layer of the network; For the activation function, this invention uses LeakyReLU to avoid gradient vanishing; Let be the trainable parameter matrix of the l-th layer, with dimension . ; This is a vector concatenation operation; Let v be the set of neighboring nodes. Let k be the set of nodes corresponding to the k-th group of learning; The aggregate weight of the k-th group's neighbors, and the average relation weight with the neighbors of that group. Positive correlation; Let be the embedding vector of neighbor node u in the (l-1)th layer of the network; The relationship weight between node v and its neighbor u is calculated using Algorithm 2.

[0035] III. Implementation Steps of the Technical Solution 1. Data Acquisition: Collect multi-omics data related to the target research object from life science databases corresponding to genomics (NCBI, TCGA, etc.), transcriptomics (GEO, etc.), proteomics (Uniprot, STRING, etc.), and metabolomics (HMDB, etc.), covering information such as SNP, mRNA expression profile, protein interaction, and metabolite concentration; 2. Data preprocessing: Data with a missing rate >20% were removed. Transcriptomics and metabolomics data were standardized using Z-score. Protein sequences were preprocessed by homology alignment using the BLAST tool to achieve data format uniformity. 3. Entity Recognition and Matching (Algorithm 1): Entities are extracted based on the biomedical named entity recognition model (BERT+CRF), and the results are used in Algorithm 1 to calculate the overall entity similarity. Matched entities (S≥0.8) are then deduplicated and standardized. Figure 3 As shown; 4. Relationship Weight Calculation (Algorithm 2): For the preprocessed multi-omics association data, substitute the data into Algorithm 2 to calculate the relationship weights, and filter out high-confidence relationships (≥0.6), such as... Figure 4 As shown; 5. Node Embedding and Knowledge Graph Construction (Algorithm 3): Input the filtered entities and relationships into Algorithm 3 to obtain node embedding vectors. Construct a knowledge graph based on Neo4j, defining node attributes (entity name, omics type, feature value) and relationship attributes (weight, association type), such as... Figure 5 As shown; the map matches known biological pathways at a rate of ≥85%; 6. Map Update and Verification: Periodically synchronize database data and repeat steps 1-5 above to achieve dynamic updates.

[0036] IV. Key Points of the Invention 1. Multi-omics entity semantic-feature fusion matching algorithm: It innovatively integrates entity semantic similarity with the specific feature similarity of each omics, and dynamically calculates the omics feature weights through the entropy method to solve the problems of heterogeneous entities across databases and low matching accuracy, and adapts to the multi-omics data features of different research objects; 2. Omics-specific relationship weight calculation algorithm: Integrates omics association strength, data precision (TP / TP+FP) and database reliability coefficient, dynamically allocates omics contribution weights based on precision, quantifies the credibility of biological associations, and avoids bias from a single data source; 3. Multi-omics Enhanced Node Embedding Algorithm: The core difference between this algorithm and the traditional GraphSAGE is that the traditional GraphSAGE uses a uniform aggregation weight for all neighbor nodes, ignoring omics specificity; this invention, however, is based on omics type (… ) Divide neighbor nodes, through (Positively correlated with the average relation weight of omics neighbors) Dynamically allocate aggregation weights to make the embedding vectors more accurately reflect the biological significance of different omics, and improve the accuracy of downstream association prediction compared with traditional GraphSAGE; 4. General-purpose multi-omics integration framework: It covers the entire process of "data preprocessing - entity integration - association mining - map construction - dynamic update", supports standardized processing of multi-source heterogeneous data, adapts to various research scenarios such as complex diseases, plant stress resistance, and microbial metabolism, and ensures the reliability of the map through known pathway matching verification. 5. Highly scalable technical design: The core algorithm is decoupled from data types and research objects, and can be quickly adapted to new scenarios by simply replacing multiple omics data sources without modifying the core algorithm logic.

[0037] V. Specific Examples The embodiments of this invention are highly versatile. The implementation process is described in detail below using "type 2 diabetes" (a complex disease scenario) and "lung cancer" (a complex disease scenario) as examples. Other research scenarios (such as plant stress resistance and microbial metabolism) can be implemented by reference, only requiring the replacement of the multi-omics data of the target research object.

[0038] Case 1: Construction of a Multi-omics Knowledge Graph for Type 2 Diabetes 1. Data Acquisition and Preprocessing 1.1 Genomics data: Type 2 diabetes-related SNP data were obtained from the GWAS Catalog and NCBI platform; 1.2 Transcriptomics data: The GSE81608 dataset was downloaded from the GEO database; 1.3 Proteomics Data: Download the protein sequences and interaction data corresponding to the above genes from Uniprot and STRING. 1.4 Metabolomics data: Downloaded type 2 diabetes-related metabolite data from HMDB.

[0039] The collected data were preprocessed, and data with a missing rate >20% were removed. Finally, 1127 SNP sites, 876 associated genes, 15148 mRNA expression profiles, 861 protein sequences, 3215 interaction relationships, and 643 metabolites were retained, as shown in Table 1.

[0040] Table 12 Multi-omics data information for type 2 diabetes

[0041] 2. Algorithm Execution Process 2.1 Entity Matching: Taking "insulin receptor gene (INSR)" as an example, calculate its semantic similarity in the NCBI and GEO databases. =0.94, genomic feature similarity =0.96, transcriptomic feature similarity =0.90, proteomics =0.93, metabolomics =0.85; calculated using the entropy method =0.33、 =0.29、 =0.24、 =0.14, =0.43, substituting into Algorithm 1 yields =0.43×0.94+0.57×(0.33×0.96+0.29×0.90+0.24×0.93+0.14×0.85)=0.92≥0.8, therefore it is determined to be a matching entity and is merged.

[0042] 2.2 Relationship Weight Calculation: The relationship weight between the INSR gene and the insulin (INS) protein was calculated. =5.6 (GWAS correlation score) =4.9 (FC ​​normalized value for differential mRNA expression) =0.91 (mutual confidence level) =0.79 (metabolite correlation); =0.97、 =0.92、 =0.94、 =0.89; =0.36、 =0.31、 =0.23、 =0.10, substituting into Algorithm 2, we get: = (0.36×5.6×0.97+0.31×4.9×0.92+0.23×0.91×0.94+0.10×0.79×0.89)=0.89≥0.6, which is determined to be a highly reliable relationship.

[0043] 2.3 Node Embedding: Taking the INSR gene as node v, its neighboring nodes include 15 genomics nodes, 20 transcriptomics nodes, 12 proteomics nodes, and 8 metabolomics nodes. Calculation... =0.28、 =0.35、 =0.22、 =0.15, substituting it into Algorithm 3 yields a 128-dimensional node embedding vector, which is used for subsequent association prediction.

[0044] 3. Map Construction Results The constructed knowledge graph of type 2 diabetes contains 3,982 entities (876 genes, 1,602 mRNAs, 861 proteins, and 643 metabolites), with 8,972 highly reliable relationships; the matching rate with the insulin signaling pathway is 93%, which is 23% higher than the existing type 2 diabetes knowledge graph.

[0045] Case 2: Construction of a Multi-omics Knowledge Graph for Lung Cancer 1. Data Acquisition and Preprocessing 1.1 Genomics data: Download lung cancer-related SNP data from TCGA and NCBI; 1.2 Transcriptomics data: The GSE11969 dataset was downloaded from the GEO database; 1.3 Proteomics Data: Download protein sequences and interaction data from Uniprot and STRING; 1.4 Metabolomics data: Download lung cancer-related metabolite data from HMDB and MetaboAnalyst.

[0046] The collected data were preprocessed, and data with a missing rate >20% were removed. Transcriptomics and metabolomics data were standardized using Z-score, and protein sequence homology was compared using BLAST. Finally, 1289 SNPs, 893 associated genes, 14892 mRNA expression profiles, 852 protein sequences, 3189 protein interactions, and 582 metabolites were retained, as shown in Table 2.

[0047] Table 2. Multi-omics data information for lung cancer

[0048] 2. Algorithm Execution Process 2.1 Entity Matching: Taking "epidermal growth factor receptor gene (EGFR)" as an example, the semantic similarity of EGFR in the TCGA and GEO databases was calculated. =0.95, genomic feature similarity =0.97, transcriptomic feature similarity =0.91, proteomics =0.94, metabolomics =0.86; calculated using the entropy method =0.34、 =0.28、 =0.25、 =0.13, =0.42, substituting into Algorithm 1 yields =0.45×0.95+0.55×(0.34×0.97+0.28×0.91+0.25×0.94+0.13×0.86)=0.93≥0.8, therefore it is determined to be a matching entity and is merged; 2.2 Relationship Weight Calculation: The relationship weight between the EGFR gene and epidermal growth factor (EGF) protein was calculated. =6.1 (GWAS correlation score) =5.4 (FC normalized value for differential mRNA expression) =0.93 (mutual confidence level) =0.82 (metabolite correlation); =0.98、 =0.93、 =0.94、 =0.88; =0.36、 =0.31、 =0.24、 =0.09, substituting into Algorithm 2 yields = (0.36×6.1×0.98+0.31×5.4×0.93+0.24×0.93×0.94+0.09×0.82×0.88)=0.91≥0.6, which is determined to be a highly reliable relationship; 2.3 Node Embedding: Taking the EGFR gene as node v, its neighboring nodes include 18 genomics nodes, 22 transcriptomics nodes, 14 proteomics nodes, and 10 metabolomics nodes. Calculation... =0.29、 =0.34、 =0.23、 =0.14, substituting into Algorithm 3 yields a 128-dimensional node embedding vector, which is used for subsequent lung cancer-related association prediction.

[0049] 3. Map Construction Results The constructed lung cancer knowledge graph contains 3,614 entities (893 genes, 1,287 mRNAs, 852 proteins, and 582 metabolites), 9,018 highly reliable relationships, and the matching rate between the graph and the lung cancer EGFR signaling pathway reaches 92%, which is 24% higher than the existing lung cancer-related graphs.

[0050] System Implementation Examples According to embodiments of the present invention, a multi-omics knowledge graph construction system adaptable to different scenarios is provided. Figure 6 This is a schematic diagram of a cross-scenario adapted multi-omics knowledge graph construction system according to an embodiment of the present invention, such as... Figure 6As shown, the cross-scenario adapted multi-omics knowledge graph construction system according to an embodiment of the present invention specifically includes: The data processing module 60 is used to collect multi-omics data related to the target research object and preprocess the multi-omics data to obtain standardized data including entities and inter-entity relationship information. The multi-omics data includes genomics, transcriptomics, proteomics, and metabolomics; the preprocessing includes data cleaning, missing value handling, and format standardization; The data cleaning includes removing data with a missing rate higher than a preset threshold; The standardization of the format includes standardizing transcriptomics and metabolomics data, as well as performing homology comparisons of protein sequence data using the BLAST tool.

[0051] The entity fusion module 62 is used to match and deduplicate entities from different sources based on the standardized data using a multi-omics entity semantic-feature fusion matching algorithm to generate a standardized entity set. The association quantification module 64 is used to perform credibility quantification on the association information between entities in the standardized entity set through an omics-specific relationship weight calculation algorithm, and to select a high-credibility relationship set based on the quantification results. Graph representation learning module 66 is used to construct an initial graph structure using the standardized entity set as nodes and the high-confidence relation set as edges, and to perform feature representation learning on the initial graph structure using a multi-omics enhanced node embedding algorithm to obtain node vector representations; and The knowledge graph construction module 68 is used to construct a knowledge graph based on the node vector representation and the high-confidence relation set, verify the accuracy of the knowledge graph, and dynamically update the knowledge graph according to the verification results and changes in multi-omics data.

[0052] The embodiments of the present invention are system embodiments corresponding to the above method embodiments. The specific operation of each module can be understood by referring to the description of the method embodiments, and will not be repeated here.

[0053] In summary, the knowledge graph constructed using the embodiments of this invention can be applied to research on the pathogenesis of complex diseases, biomarker screening, personalized treatment plan formulation, plant stress resistance gene mining or microbial metabolic regulation analysis, plant stress resistance gene functional annotation, and microbial metabolic pathway optimization, covering both disease and non-disease application scenarios. Specifically, it includes the following beneficial effects: 1. Deep integration of multiple omics + strong versatility: It achieves the organic integration of data from the four major omics through three core algorithms, solving the problems of data dispersion and heterogeneity; the framework design does not depend on specific research objects and can be flexibly adapted to multiple scenarios such as complex diseases, agricultural biotechnology, and microbial metabolism, making it more widely applicable than existing single-scenario map construction technologies; 2. The algorithm is advanced and reliable: entity matching integrates semantic and omics features, relation weights quantify data reliability, node embedding enhances omics specificity, and the three algorithms work together to improve the accuracy of the graph and the success rate of downstream tasks (association prediction, mechanism analysis); 3. Wide range of applications and great potential for transformation: The constructed knowledge graph can intuitively present the multi-omics association network of research objects, providing core tools for the analysis of complex disease mechanisms, screening of biomarkers, discovery of personalized treatment targets, and mining of plant stress resistance genes, and has greater potential for cross-domain transformation than existing technologies; 4. Dynamic updates and easy maintenance: Supports synchronous database updates and dynamic map iteration, solving the problem of "static" existing maps and adapting to research progress over a long period of time; the modular design of the technical framework facilitates the subsequent expansion to more omics types (such as epigenomics and microbiome).

[0054] Device Example 1 This invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it performs the steps described in the method embodiment.

[0055] Device Example 2 This invention provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, performs the steps described in the method embodiment.

[0056] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a multi-omics knowledge graph that adapts to different scenarios, characterized in that, include: S1: Collect multi-omics data related to the target research object, and preprocess the multi-omics data to obtain standardized data including entity and inter-entity relationship information; S2: Based on the standardized data, a multi-omics entity semantic-feature fusion matching algorithm is used to match and deduplicate entities from different sources to generate a standardized entity set; S3: The credibility of the association information between entities in the standardized entity set is quantified by the omics-specific relation weight calculation algorithm, and a high credibility relation set is selected based on the quantification results; S4: Using the standardized entity set as nodes and the high-confidence relation set as edges, construct an initial graph structure, and use a multi-omics enhanced node embedding algorithm to learn the feature representation of the initial graph structure to obtain node vector representation; as well as S5: Construct a knowledge graph based on the node vector representation and the high-confidence relation set, verify the accuracy of the knowledge graph, and dynamically update the knowledge graph according to the verification results and changes in multi-omics data.

2. The method according to claim 1, characterized in that, The multi-omics data includes genomics, transcriptomics, proteomics, and metabolomics; the preprocessing includes data cleaning, missing value handling, and format standardization; The data cleaning includes removing data with a missing rate higher than a preset threshold; The standardization of the format includes standardizing transcriptomics and metabolomics data, as well as performing homology comparisons of protein sequence data using the BLAST tool.

3. The method according to claim 1, characterized in that, S2 specifically includes: Calculate the semantic similarity between entities to be matched from different data sources; Calculate the omics feature similarity of the entities to be matched across multiple omics dimensions; Based on the distribution characteristics of each omics data, the feature weights of multiple omics feature similarities in the comprehensive matching calculation are dynamically determined using Formula 1 and Formula 2. Official 1; Official 2; in, This represents the information entropy of the k-th group of learning data. Let represent the probability distribution of the i-th feature in the k-th omics, and n represent the total number of features in that omics. Represents the feature weights of the k-th group; Based on the semantic similarity, omics feature similarity, and feature weights, the comprehensive matching similarity of the entities to be matched is calculated using Formula 3. Official 3; in, Representing entities and The overall matching similarity, Represents semantic weight coefficients. Indicates the semantic similarity between entities. These correspond to genomics, transcriptomics, proteomics, and metabolomics, respectively. Let be the similarity of omics features of the k-th omics; The comprehensive matching similarity is compared with a preset matching threshold, and entities that are greater than or equal to the preset matching threshold are identified as homologous entities and merged and deduplicated to form a standardized entity set.

4. The method according to claim 3, characterized in that, S3 specifically includes: Obtain association strength data that characterizes the strength of association between entities in the standardized entity set; The data reliability contribution weight of each omics data is determined based on the accuracy of each omics data. The reliability coefficients of the databases from which each omics data originates are obtained based on the database impact factor and data update frequency. Based on the aforementioned association strength data, data reliability contribution weight, and reliability coefficient, Formula 4 is used to calculate the comprehensive weight value of the relationships between corresponding entities in the standardized entity set. Official 4; in, Representing entities and Interrelationship The weight, Represents the 0-1 normalization function. This represents the contribution weight of the k-th group. This represents the association strength of the k-th group. This represents the reliability coefficient of the database from which the k-th omics data originates; The comprehensive weight value is compared with a preset confidence threshold, and relationships that are greater than or equal to the preset confidence threshold are selected to form a set of highly reliable relationships.

5. The method according to claim 1, characterized in that, S4 specifically includes: The neighboring nodes of each node in the initial graph structure are grouped according to their respective omics types; Calculate the average weight value of the relationship between neighbor nodes and the central node for each omics type; The aggregation weight of each group of omics type neighbor nodes in the information aggregation process is dynamically determined based on the average weight value; wherein, the aggregation weight is positively correlated with the corresponding average weight value; Based on the aggregation weight, the features of neighbor nodes of each omics type are weighted and aggregated. The aggregated features of neighbor nodes are then fused with the features of the central node, and a nonlinear transformation is performed through a neural network layer to obtain the vector representation of the node in the current layer. By stacking multiple layers of a neural network and iteratively updating them, the final node vector representation used for downstream tasks is obtained.

6. The method according to claim 1, characterized in that, S5 specifically includes: The entities in the standardized entity set, the node vector representations, and the relations and weights in the high-confidence relation set are imported into a graph database to construct a multi-omics knowledge graph. The multi-omics knowledge graph is matched with at least one known standard biological pathway, and at least one preset threshold or algorithm parameter in the graph construction process is optimized and adjusted based on the matching rate results. When new multi-omics data is available or the data in the graph database is updated, steps S1-S5 are repeated to dynamically iterate and update the multi-omics knowledge graph.

7. A multi-omics knowledge graph construction system that adapts to different scenarios, characterized in that, include: The data processing module is used to collect multi-omics data related to the target research object and preprocess the multi-omics data to obtain standardized data including entity and inter-entity relationship information. The entity fusion module is used to match and deduplicate entities from different sources based on the standardized data using a multi-omics entity semantic-feature fusion matching algorithm, and generate a standardized entity set. The association quantification module is used to quantify the credibility of the association information between entities in the standardized entity set through an omics-specific relationship weight calculation algorithm, and to select a set of highly credible relationships based on the quantification results. The graph representation learning module is used to construct an initial graph structure with the standardized entity set as nodes and the high-confidence relation set as edges, and to learn the feature representation of the initial graph structure using a multi-omics enhanced node embedding algorithm to obtain node vector representations. as well as The knowledge graph construction module is used to construct a knowledge graph based on the node vector representation and the high-confidence relation set, verify the accuracy of the knowledge graph, and dynamically update the knowledge graph according to the verification results and changes in multi-omics data.

8. The system according to claim 7, characterized in that, The multi-omics data includes genomics, transcriptomics, proteomics, and metabolomics; the preprocessing includes data cleaning, missing value handling, and format standardization; The data cleaning includes removing data with a missing rate higher than a preset threshold; The standardization of the format includes standardizing transcriptomics and metabolomics data, as well as performing homology comparisons of protein sequence data using the BLAST tool.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the cross-scenario adapted multi-omics knowledge graph construction method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for information transmission, which, when executed by a processor, implements the steps of the cross-scenario adaptation multi-omics knowledge graph construction method as described in any one of claims 1-6.

Images