Metabolic pathway and data-driven breathomics knowledge graph dynamic construction method and system

Abstract

The application discloses a kind of based on metabolic pathway and data-driven exhaled breathomics knowledge graph dynamic construction method and system.The method obtains compound, module and metabolic pathway information from public database, filters target compound matched with human exhaled breath volatile organic compounds VOCs, and constructs initial knowledge network with metabolic pathway priori constraint;Subsequently, obtain the exhaled breath VOCs detection data of subject, construct feature matrix after pretreatment, and mine data-driven candidate edge;The candidate edge is filtered based on metabolic pathway priori, and the candidate edge that meets the condition is supplemented to initial knowledge network, to obtain the completed exhaled breathomics knowledge graph;When new data is added, incremental update is carried out to existing graph, to realize the continuous expansion, dynamic evolution and version management of graph.The application can improve the completeness, biological rationality and interpretability of exhaled breathomics knowledge representation, and is suitable for disease-related exhaled breath marker analysis and mechanism research.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics, medical data analysis and knowledge graph construction technology, specifically to a method and system for dynamically constructing a breathomics knowledge graph based on metabolic pathway priors and data-driven approaches. Background Technology

[0002] Human exhaled breath contains a variety of volatile organic compounds (VOCs), the composition of which is influenced by a combination of factors, including host endogenous metabolism, gut and airway microbial metabolism, dietary intake, and environmental exposure. The composition of VOCs is closely related to the body's physiological and pathological states; therefore, detecting and analyzing them helps to obtain information reflecting the body's metabolic state and health status. Thus, breath analysis, due to its non-invasive, convenient, and repeatable sampling characteristics, shows great potential for application in disease detection and assessment.

[0003] However, the formation and transformation of VOCs involve multiple metabolic pathways and various biochemical reactions, and their association with disease states typically exhibits many-to-many and multi-level relationships. Furthermore, VOCs are influenced by multi-source information, including microbial metabolism, diet, and environmental factors. Therefore, breathwork data not only contains information on single variables but also complex entity relationships and interaction structures. Current breathwork analyses often focus more on differential screening or predictive modeling, and their systematic expression of the association mechanisms between VOCs, metabolic processes, disease phenotypes, and external influencing factors remains insufficient, limiting the systematic analysis and utilization of breathwork data, and consequently, its interpretability remains limited.

[0004] To provide a unified representation and model of the complex relationships described above, a data organization method capable of expressing multiple entities, multiple relationships, and their hierarchical structures is needed. Knowledge graphs, by organizing different types of entities and their relationships into a graph structure, can uniformly represent the associations between VOCs, metabolic pathways, disease phenotypes, and related influencing factors, and support relation querying and reasoning analysis. However, currently, there is a lack of structured knowledge graphs specifically for VOCs. The mapping relationship between existing metabolic pathway databases and exhaled VOCs is insufficient, and the directly corresponding molecular overlap is limited, making it difficult to directly use them to construct the association structure between VOCs and metabolic processes.

[0005] Existing knowledge graph construction methods largely rely on manual organization, rule extraction, or static knowledge import. However, due to insufficient support from publicly available standardized databases for exhaled VOCs, and the complex nature of their sources, multi-source influences, and indirect association pathways, graphs constructed using existing methods struggle to fully reflect the potential association structures within exhalomics. Furthermore, existing methods lack dynamic update mechanisms when faced with new data, hindering the continuous expansion and evolution of the graphs. Therefore, it is necessary to propose a dynamic knowledge graph construction method for exhalomics that combines metabolic pathway priors with a data-driven completion mechanism to address the problems of missing graph relationships, insufficient biological constraints, and difficulty in continuous updating with new data in existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a dynamic construction method for a breathomics knowledge graph based on metabolic pathways and data-driven approaches. In the case of insufficient mapping between exhaled VOCs and existing publicly available metabolic pathway resources, this method first constructs an initial knowledge network with prior constraints on metabolic pathways based on a public database. Then, it combines breathomics sample data to mine candidate associations, and performs edge patching and updates on the initial knowledge network, thereby constructing and dynamically updating the breathomics knowledge graph.

[0007] According to a first aspect of this specification, a method for dynamically constructing a breathomics knowledge graph based on metabolic pathways and data-driven approaches is provided, the method comprising:

[0008] S1. Obtain compound, module and metabolic pathway information from a publicly available metabolic pathway database, determine the metabolic relationships between target volatile organic compounds, and construct an initial knowledge network with prior constraints on metabolic pathways based on the metabolic relationships.

[0009] S2, acquire breath volatile organic compound detection data, perform data preprocessing on the detection data, construct a feature matrix, and mine candidate association relationships between volatile organic compounds based on the feature matrix to obtain a candidate association network. Filter the candidate association relationships in the candidate association network to obtain a data-driven candidate edge set.

[0010] S3, based on the prior knowledge of metabolic pathways, selects candidate edges and completes the knowledge graph. Specifically, it calculates the comprehensive confidence of the candidate edges in the data-driven candidate edge set obtained in S2, and adds the candidate edges that meet the confidence conditions to the initial knowledge network constructed in S1 to form the completed breath mics knowledge graph.

[0011] S4. When new data on the detection of volatile organic compounds in exhaled air or new information in the publicly available metabolic pathway database are obtained, the existing exhaled air mics knowledge graph is incrementally updated.

[0012] Furthermore, step S1 specifically includes:

[0013] Obtain all compound names from the publicly available metabolic pathway database, and filter out compounds that exist simultaneously in both the publicly available metabolic pathway database and a pre-organized list of volatile organic compounds in human exhaled gas, denoting them as the target compound set.

[0014] Extract module and metabolic pathway information related to the target compound set from the publicly available metabolic pathway database, and extract the adjacency relationships between the screened metabolic pathways;

[0015] After standardizing the names and mapping the identifiers of the target compounds, metabolic relationships between the compounds are constructed, and an initial knowledge network is built based on these metabolic relationships.

[0016] Furthermore, for any two target compounds, the metabolic association is as follows: (1) If they are located in the same metabolic pathway and directly connected, they are transformation relationships; (2) If they are located in the same module, they are module-related relationships; (3) If they are located in the same metabolic pathway and not directly connected, they are pathway-related relationships; (4) If they are located in two adjacent metabolic pathways, they are cross-pathway relationships; (5) If there is a clear upstream / downstream metabolic relationship, they are upstream / downstream metabolic relationships; (6) If they are both metabolized from the same compound, they are co-source pathway relationships; The edge weights corresponding to the above metabolic associations decrease sequentially from (1) to (6).

[0017] Furthermore, in step S2, the data preprocessing includes background subtraction and outlier processing, and the preprocessed data is standardized to obtain the feature matrix of volatile organic compounds.

[0018] Further, in step S2, the screening of candidate associations specifically involves:

[0019] The characteristic matrix of volatile organic compounds (VOCs) is denoted as... Where n represents the number of breath genome samples, p represents the number of VOC features, and x ki This represents the detection value of the i-th VOC in the k-th sample;

[0020] Calculate the Pearson correlation coefficient r between the i-th VOC and the j-th VOC. ij and two-tailed test p-value ij ;

[0021] Define a candidate Pearson correlation coefficient (PCC) threshold sequence. For any candidate PCC threshold t... l Construct the corresponding adjacency matrix and calculate the number of network edges based on the adjacency matrix; traverse the candidate PCC threshold sequence and calculate the threshold when the candidate PCC threshold is t l Increase to t l+1Rate of change of the number of sides;

[0022] A preset threshold η is used to determine the abrupt change in the number of edges. If there exists a candidate PCC threshold whose rate of change in the number of edges is greater than η, then that candidate PCC threshold is used as the final dynamic threshold t. * If the rate of change of the number of edges corresponding to all candidate PCC thresholds is less than or equal to η, then the candidate PCC threshold with the largest rate of change of the number of edges is taken as the final dynamic threshold t. * ;

[0023] When r is satisfied at the same time ij Greater than t * And p ij When the value is less than 0.05, it is considered that there is a data-driven candidate association relationship between the i-th VOC and the j-th VOC, thus obtaining the data-driven candidate edge set.

[0024] Furthermore, in step S3, the comprehensive confidence C of the candidate edge is calculated by combining the data-driven evidence of the candidate edge with its metabolic prior support in the initial knowledge network. ij Specifically:

[0025] For each data-driven candidate edge, a classification function g is defined. ij If both VOC nodes at the two ends of a candidate edge belong to the initial knowledge network, and there is already a prior metabolic association between them, then g ij =1; If both VOC nodes at the two ends of the candidate edge belong to the initial knowledge network, and there is no prior metabolic association between them, then g ij =2; if only one end of the candidate edge has a VOC node belonging to the initial knowledge network, then g ij =3; if neither of the VOC nodes at either end of the candidate edge belongs to the initial knowledge network, then g ij =4;

[0026] Data-driven support D for calculating candidate edges ij The calculation formula is as follows:

[0027]

[0028] Where, ω d The weights of the relevant strength terms;

[0029] According to the hierarchical function g ij Define the prior hierarchical support P of the candidate edges ij :g ij When =1, P ij =1; g ij When =2, P ij =0.75; g ij When =3, P ij =0.5; g ijWhen =4, P ij =0;

[0030] Data-driven support based on candidate edges D ij Prior hierarchical support P ij Calculate the comprehensive confidence C of candidate edges. ij :

[0031]

[0032] Where, ω c The weight of data-driven evidence in the overall confidence level;

[0033] The overall confidence level C of the candidate edges ij Candidate edges with a confidence level greater than or equal to the edge-filling threshold θ are included in the set of edges to be filled, and the combined confidence level C is set accordingly. ij As edge weight.

[0034] Further, in step S3, for candidate edges that meet the confidence criteria, they are added to the initial knowledge network. Specifically: if the two VOC nodes corresponding to the candidate edge already have prior edges in the initial knowledge network, the edge is not added again, but the Pearson correlation coefficient, two-sided test p-value, and comprehensive confidence score corresponding to the candidate edge are written into the graph database as attributes of the existing prior edge; if the two VOC nodes corresponding to the candidate edge do not have prior edges in the initial knowledge network, the candidate edge is added to the initial knowledge network, and the Pearson correlation coefficient, two-sided test p-value, and comprehensive confidence score corresponding to the candidate edge are also written into the graph database as attributes of the newly added candidate edge; if the VOC node corresponding to the candidate edge does not exist in the initial knowledge network, it is added to the data-driven VOC node set.

[0035] Furthermore, in step S4, when adding new public metabolic pathway database information, firstly, step S1 is repeated to obtain the updated prior knowledge network; then, the updated prior knowledge network is merged with the prior part in the current knowledge graph; for newly added nodes and newly added prior edges, they are directly written into the graph database; for existing nodes and edges, their source, database version, last update time, and evidence attributes are updated; then, the candidate edge screening and graph completion process in step S3 is repeated to obtain the updated breathomics knowledge graph.

[0036] Furthermore, in step S4, when adding new exhaled volatile organic compound detection data from the same source, the new sample data is first merged with the original sample data to obtain an updated feature matrix; then, based on the updated feature matrix, step S2 is re-executed to obtain an updated set of data-driven candidate edges, and then based on step S3, an updated set of edges to be supplemented is obtained; for existing data-driven edges in the current graph, if they exist in the updated set of edges to be supplemented, their Pearson correlation coefficient, two-sided test p-value, edge weight, and most recent update time are updated; if they no longer exist in the updated set of edges to be supplemented, the data-driven edge is invalidated.

[0037] Furthermore, in step S4, if the new data is sample data of different disease subtypes, or data that has obvious category differences from the original data, the new data is not directly mixed with the original sample data for calculation, but is processed as an independent data group; first, the candidate edge mining process in step S2 and the comprehensive confidence screening process in step S3 are executed separately; then, the category to which it belongs is recorded in the graph database by edge attributes or subgraph identifiers, the data-driven relationship corresponding to different disease subtypes or different categories of sample data is stored as a category subgraph, and it is incorporated into the overall graph.

[0038] According to a second aspect of this specification, a dynamic construction system for a breathomics knowledge graph based on metabolic pathways and data-driven methods is provided. This system is used to implement the dynamic construction method for a breathomics knowledge graph as described in the first aspect. The system includes:

[0039] An initial knowledge network construction module is used to acquire compound, module and metabolic pathway information from a public metabolic pathway database, determine the metabolic relationships between target volatile organic compounds, and construct an initial knowledge network with prior constraints on metabolic pathways based on the metabolic relationships.

[0040] The data-driven candidate edge mining module is used to acquire exhaled volatile organic compound detection data, perform data preprocessing on the detection data, construct a feature matrix, mine candidate association relationships between volatile organic compounds based on the feature matrix to obtain a candidate association network, and filter the candidate association relationships in the candidate association network to obtain a data-driven candidate edge set.

[0041] The candidate edge screening and graph completion module is used to calculate the comprehensive confidence of the candidate edges in the data-driven candidate edge set, and to add the candidate edges that meet the confidence conditions to the initial knowledge network to form the completed breath mics knowledge graph.

[0042] The graph update module is used to perform incremental updates on the existing breathomics knowledge graph when new data is added.

[0043] According to a third aspect of this specification, an electronic device is provided, including a memory and a processor, the memory being coupled to the processor; wherein the memory is used to store program data, and the processor is used to execute the program data to implement the dynamic construction method of the breathomics knowledge graph as described in the first aspect.

[0044] According to a fourth aspect of this specification, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method for dynamically constructing a breathomics knowledge graph as described in the first aspect.

[0045] According to a fifth aspect of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the dynamic construction method for a breathomics knowledge graph as described in the first aspect.

[0046] The beneficial effects of this invention are as follows: Given the insufficient mapping between exhaled VOCs and existing publicly available metabolic pathway resources, this invention first constructs an initial knowledge network with prior constraints on metabolic pathways based on a public database. Then, it mines candidate associations using exhalomics sample data to supplement and update the initial knowledge network, thereby constructing and dynamically updating an exhalomics knowledge graph. This invention solves the problems of missing graph relationships, insufficient biological constraints, and difficulty in continuously updating with new data in existing technologies. It can improve the completeness, biological rationality, and interpretability of exhalomics knowledge representation, and is suitable for the analysis and mechanism research of disease-related exhaled biomarkers. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 The flowchart illustrates an exemplary embodiment of a dynamic construction method for a knowledge graph based on metabolic pathways and data-driven exhalomics.

[0049] Figure 2 A flowchart illustrating the construction of an initial knowledge network with prior constraints on metabolic pathways, as shown in an exemplary embodiment;

[0050] Figure 3 A flowchart illustrating, as an exemplary embodiment, of obtaining a data-driven candidate edge set based on sample data mining;

[0051] Figure 4A schematic diagram illustrating a constructed exhalomics knowledge graph for diabetes, as an exemplary embodiment;

[0052] Figure 5 This is a schematic diagram of the structure of an electronic device as an exemplary embodiment. Detailed Implementation

[0053] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0054] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0055] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0056] like Figure 1 As shown, this embodiment provides a dynamic construction method for a breathomics knowledge graph based on metabolic pathways and data-driven approaches for diabetes. The method first constructs an initial knowledge network with prior constraints on metabolic pathways based on a publicly available metabolic pathway database. Then, it mines candidate associations using sample data obtained through breathomics detection, and filters and fills in the edges of these candidate associations based on metabolic pathway priors. Finally, it constructs and dynamically updates a breathomics knowledge graph for diabetes.

[0057] In this embodiment, the publicly available metabolic pathway database can be the Kyoto Encyclopedia of Genes and Genomes (KEGG), but is not limited to it; breathwork sample data can be obtained through a high-sensitivity surface acoustic wave gas sensor combined with gas chromatography-mass spectrometry (GC-MS), but is not limited to it. The dynamic construction method of breathwork knowledge graph provided in this embodiment specifically includes the following steps:

[0058] S1: Obtain data from publicly available metabolic pathway databases and construct an initial knowledge network with prior constraints on metabolic pathways.

[0059] In this step, the KEGG PATHWAY database from the KEGG database is selected as the source of prior information on metabolic pathways. The KEGG PATHWAY database stores metabolic pathway maps, signaling pathway maps, and their corresponding compound, module, and reaction information, which can be used to characterize the metabolic relationships between compounds.

[0060] First, compound information from the KEGG database is obtained through the interface provided by KEGG. Specifically, http: / / rest.kegg.jp / list / compound is called to retrieve all compound names from the KEGG database. These compounds are then compared with a pre-compiled list of VOCs reported in human exhaled breath from existing studies. Compounds present in both the KEGG database and the human exhaled breath VOCs list are selected and designated as the target compound set. In this embodiment, 16 VOCs were obtained after screening.

[0061] Next, module and metabolic pathway information related to the target compound set is extracted. Specifically, http: / / rest.kegg.jp / list / module and http: / / rest.kegg.jp / list / pathway / map are called to obtain the complete list of modules and the list of metabolic pathways, respectively. Then, based on the target compound set, http: / / rest.kegg.jp / get / {module_id} is called to traverse each module and filter out modules containing the target compounds. Subsequently, http: / / rest.kegg.jp / get / {pathway_id} is called to traverse each metabolic pathway and filter out metabolic pathways containing the above-selected modules or target compounds. In this embodiment, a total of 17 modules and 19 metabolic pathways were obtained after screening.

[0062] Furthermore, to determine the adjacency relationships between metabolic pathways, http: / / rest.kegg.jp / get / {pathway_id} is called again to obtain detailed information about each metabolic pathway, and the adjacency relationships between the selected metabolic pathways are extracted.

[0063] After obtaining the aforementioned compounds, modules, metabolic pathways, and their adjacency relationships, the target compounds are named and mapped. In this embodiment, the Chemical Substance Identifier (CAS number) is preferably used to uniformly match VOCs from different data sources, thereby reducing ambiguity caused by synonyms, aliases, and abbreviations.

[0064] Based on this, metabolic relationships among VOCs are constructed. Specifically, these metabolic relationships include:

[0065] (1) If any two target compounds are located in the same metabolic pathway and are directly connected, then it is considered that there is a transformation relationship between them;

[0066] (2) If any two target compounds are located in the same module, then it is considered that there is a module association relationship between them;

[0067] (3) If any two target compounds are located in the same metabolic pathway and are not directly connected, then they are considered to have a pathway association relationship.

[0068] (4) If any two target compounds are located in two adjacent metabolic pathways, then a cross-pathway association is considered to exist between them;

[0069] (5) If any two target compounds have a clear upstream and downstream metabolic relationship, then it is considered that there is an upstream / downstream metabolic relationship between them;

[0070] (6) If any two target compounds are derived from the metabolism of the same compound, then a common source pathway relationship is considered to exist between them.

[0071] The edge weights corresponding to the above metabolic associations decrease sequentially from (1) to (6), and in this embodiment they are set to 6, 5, 4, 3, 2, 1.

[0072] Finally, an initial knowledge network with prior constraints on metabolic pathways is constructed based on the aforementioned metabolic relationships, such as... Figure 2 As shown.

[0073] S2: Collect exhaled VOCs data and mine data-driven candidate edges.

[0074] In this step, exhaled gas samples are collected from the subjects, and corresponding VOCs detection data are obtained. The detection data includes information on the types of compounds and their relative abundance.

[0075] Specifically, a highly sensitive surface acoustic wave (SAW) gas sensor is used to detect the exhaled gas of the subject, and a capillary separation module separates different VOCs components according to different transit times. The separated components condense and deposit on the SAW gas sensor, and quantitative detection of VOCs components is achieved by monitoring changes in sensor parameters such as frequency and phase. To improve the accuracy of compound identification, GC-MS can be further used to confirm the types of VOCs.

[0076] In this embodiment, sample collection was conducted in a clean, well-ventilated room free of any unusual odors. Subjects fasted from 8:00 PM the day before testing and abstained from water after midnight. On the day of collection, subjects did not brush their teeth with toothpaste, rinsed their mouths five times with purified water, and then exhaled into a 3L Tedlar sampling bag while remaining quiet. Sampling was performed using a deep nasal inhalation and slow oral exhalation, collecting only the mid-exhalation air through a three-way connector. Simultaneously, air from the testing environment was collected as a background control sample. After collection, the samples were sealed and tested within 24 hours.

[0077] During the data preprocessing stage, the detection values ​​of various VOCs in the subject's exhaled gas are subtracted from the detection values ​​of the corresponding VOCs in the background control sample to obtain the final VOCs data; when the detection value of a certain VOC in the background control sample is greater than the detection value of that VOC in the exhaled gas, the detection value of that VOC is recorded as 0.

[0078] Given that the number of target compounds that can be mapped in the KEGG database is lower than the total number of VOCs in the breath mics detection data, it is difficult to build a complete knowledge network by relying solely on public databases. Therefore, it is necessary to further explore the potential relationships between VOCs based on breath mics sample data in order to achieve data-driven edge completion.

[0079] Specifically, the collected VOCs data were first screened. Since some VOCs only appeared in a small number of samples, making robust statistical analysis difficult, VOCs detected in at least 50% of the samples were selected for subsequent analysis. In this embodiment, the total number of samples was 219, and after screening, 202 VOCs were obtained.

[0080] Then, for each VOC, its detection values ​​in all samples are concatenated in the order of the samples to form a feature sequence, and a feature matrix is ​​constructed; if a VOC is not detected in a sample, the detection value of that VOC in that sample is recorded as 0.

[0081] Furthermore, a metabolic association mining algorithm is used to screen for potential associations between VOCs. In this embodiment, by scanning the topological indicators of the VOC network, such as the number of edges, network density, average degree, and maximum connected component, under different Pearson correlation coefficient (PCC) thresholds, locations where the network structure changes significantly are identified, and the PCC value corresponding to that location is used as a dynamic threshold.

[0082] Specifically, let the collected VOC feature matrix be... for:

[0083]

[0084] Where n represents the number of subject samples, p represents the number of VOC features, and x ki This represents the detection value of the i-th VOC in the k-th sample.

[0085] For any two VOC features i and j, calculate their Pearson correlation coefficient to obtain the PCC matrix. :

[0086]

[0087] Where, r ij Let r represent the Pearson correlation coefficient between the i-th VOC and the j-th VOC, and r ii = 1, r ij = r ji .

[0088] A two-sided significance test was performed on the Pearson correlation coefficient for each pair of VOCs to obtain the significance matrix. :

[0089]

[0090] Where, p ij Let p represent the two-tailed test p-value corresponding to the Pearson correlation coefficient between the i-th VOC and the j-th VOC.

[0091] Set candidate PCC threshold sequence for:

[0092]

[0093] In this embodiment, a starting threshold of 0.3 and a step size of 0.05 are used, and the threshold is set to... .

[0094] For any candidate PCC threshold t l For each l=1,…,L, construct the corresponding adjacency matrix. :

[0095]

[0096] Further calculations reveal that for the candidate PCC threshold t l The number of its network edges for:

[0097]

[0098] Traversing the aforementioned candidate PCC threshold sequence The calculation yields the threshold value when t is... l Increase to t l+1 When, the rate of change of the number of sides for:

[0099]

[0100] The preset threshold for determining abrupt changes in the number of edges is η. If ρ exists... l If η > t, then it is considered that at t l To t l+1 When a sudden change in the number of edges occurs in the interval, the candidate PCC threshold t corresponding to l at this time is used. l As the final dynamic threshold If ρ satisfies for all l l If ≤ η, then the candidate PCC threshold t with the largest rate of change in the number of edges is selected. l As the final dynamic threshold In this embodiment, the mutation threshold η is set to 0.2.

[0101] When both conditions are met and At that time, it is believed that there is a data-driven candidate association relationship between the i-th VOC and the j-th VOC.

[0102] For VOC pairs that meet the above conditions, candidate triple relations are constructed. This yields a data-driven candidate edge set based on sample data mining, such as... Figure 3 As shown.

[0103] Finally, the initial knowledge network constructed in step S1 and the data-driven candidate edge set obtained in this step are imported into the graph database. In this embodiment, the graph database is preferably Neo4j. After importation, nodes represent VOC entities, and edges represent prior metabolic associations or data-driven candidate associations between entities.

[0104] S3: Based on metabolic pathway priors, candidate edges are screened and the graph is completed.

[0105] In this step, the data-driven candidate edge set obtained in step S2 is subjected to confidence screening and weight calculation, and the candidate edges that meet the conditions are added to the initial knowledge network constructed in step S1, thereby forming a complete breathomics knowledge graph, such as... Figure 4 As shown.

[0106] Specifically, let the initial knowledge network with prior constraints on metabolic pathways constructed in step S1 be:

[0107]

[0108] Wherein, V0 represents the set of target VOC nodes jointly selected by the KEGG database and the list of human exhaled VOCs, and E0 represents the set of prior metabolic association edges constructed based on the same module association, same pathway association, cross pathway association, upstream / downstream metabolic relationship, transformation relationship and common source path relationship.

[0109] Let the set of data-driven candidate edges obtained through sample data mining in step S2 be:

[0110]

[0111] Among them, e ij data Let represent the data-driven candidate edge between the i-th VOC and the j-th VOC. For any candidate edge e ij data The corresponding Pearson correlation coefficient is denoted as r. ij The p-value for a two-tailed test is denoted as p. ij The dynamically determined PCC threshold in step S2 is denoted as t. * .

[0112] The purpose of this step is to calculate the overall confidence C of the candidate edges by combining the data-driven evidence for the candidate edges with their metabolic prior support in the initial knowledge network. ij Only candidate edges that meet the confidence criteria are added to the initial knowledge network.

[0113] First, based on whether the VOC nodes at both ends of the candidate edge belong to the initial knowledge network and whether there is a prior metabolic association between the two ends of the node, each data-driven candidate edge obtained in step S2 is classified.

[0114] For any candidate edge e ij data Define node v i and node v j Whether it belongs to the initial knowledge network:

[0115]

[0116] Further define the candidate edge grading function g ij :

[0117]

[0118] The meanings of the above classifications are as follows:

[0119] When g ij When =1, it means that both VOC nodes at both ends of the candidate edge belong to the initial knowledge network and there is a prior metabolic relationship between them. This candidate edge is a strong prior support edge.

[0120] When g ij When =2, it means that both VOC nodes at both ends of the candidate edge belong to the initial knowledge network and there is no prior metabolic association between them. This candidate edge is an indirect prior support edge.

[0121] When g ij When =3, it means that if one end of the candidate edge is a VOC node belonging to the initial knowledge network, then the other end of the candidate edge, which is not a VOC node belonging to the initial knowledge network, can be anchored to the known metabolic network through the candidate edge. This candidate edge is a knowledge anchoring extension edge.

[0122] When g ij When the value is 4, it means that neither of the VOC nodes at the two ends of the candidate edge belongs to the initial knowledge network, and the candidate edge is a completely data-driven edge.

[0123] Then, the data-driven support D of the candidate edges is calculated. ij Since the candidate edges in step S2 already satisfy: and Therefore, the data-driven support of candidate edges can be determined by the degree to which the Pearson correlation coefficient exceeds the dynamic threshold and the significance level:

[0124]

[0125] Where, ω d The weights of the relevant strength terms satisfy: .

[0126] Furthermore, according to the candidate edge classification function g ij Define the prior hierarchical support P of the candidate edges. ij :

[0127]

[0128] Subsequently, the support degree D is driven by the candidate edges. ij Prior hierarchical support P ij Calculate the comprehensive confidence C of candidate edges. ij :

[0129]

[0130] Where, ω c The weight of data-driven evidence in the overall confidence score is given by satisfying: .

[0131] Finally, data-driven candidate edges that meet the following edge-filling conditions are included in the set E of edges to be filled. s And the comprehensive confidence C corresponding to the candidate edge. ij As its edge weight:

[0132]

[0133] Where θ is the confidence threshold for edge padding.

[0134] In this embodiment, the parameter values ​​mentioned above are: .

[0135] Furthermore, for candidate edges that satisfy the edge completion condition, they are mapped one by one to the corresponding VOC nodes in the initial knowledge network. If candidate edge e ij data If the two corresponding VOC nodes already have prior edges in the initial knowledge network, then no new edges will be added. Instead, the Pearson correlation coefficient r corresponding to the candidate edge will be used. ij Two-tailed test p-value ij Overall confidence level C ij This information is written into the graph database as an attribute of an existing prior edge to strengthen the data evidence for that prior edge. If candidate edge e... ij data If the two corresponding VOC nodes do not have prior edges in the initial knowledge network, but meet the above-mentioned edge addition conditions, then a new data-driven candidate edge is added to the initial knowledge network, and the Pearson correlation coefficient r corresponding to the candidate edge is also added. ij Two-tailed test p-value ij Overall confidence level C ij As an attribute of the newly added data-driven candidate edge, it is written into the graph database. If the candidate edge e ij data If the corresponding VOC node is not in the initial knowledge network, it is added to the data-driven VOC node set, denoted as V. c .

[0136] Furthermore, in this embodiment, different types of labels and attributes can be set for edges from different sources. For example, edges that originate solely from the prior knowledge of the KEGG metabolic pathway are labeled as "prior_edge", edges that originate solely from sample data-driven analysis are labeled as "data_edge", and edges that are both supported by prior knowledge and validated by data are labeled as "validated_edge".

[0137] Finally, the complete breath genomics knowledge graph was obtained. :

[0138]

[0139] in, , W represents the set of edge weights in the knowledge graph.

[0140] Through the above screening and edge-filling operations, potential associations that are not explicitly included in the public database but have strong statistical support in the sample data and meet the prior constraints of metabolic pathways can be added to the original initial knowledge network. The completed exhalomics knowledge graph not only retains the prior relationships of metabolic pathways in the KEGG database, but also introduces high-confidence potential associations obtained from the mining of exhaled VOCs sample data of subjects. This can make up for the problem of insufficient coverage of exhaled VOCs in the public metabolic pathway database, improve the completeness of the exhalomics knowledge graph and its ability to express diabetes-related exhalomics features.

[0141] S4: Perform dynamic updates to the graph.

[0142] In this step, when new exhaled VOCs detection data or new public database information is acquired, the existing exhaled mics knowledge graph is incrementally updated to achieve continuous expansion, dynamic evolution, and version management of the knowledge graph.

[0143] After acquiring new data, different incremental update strategies are executed based on the type of new data.

[0144] Specifically, if the new data consists of compound, module, and metabolic pathway information from a public database, the process of obtaining public database information, matching target compounds, screening modules, screening metabolic pathways, and extracting metabolic associations in step S1 is first repeated to obtain an updated prior knowledge network. Then, the updated prior knowledge network is merged with the prior portion of the current graph. New nodes and new prior edges are directly written to the graph database; for existing nodes and edges, their source, database version, most recent update time, and evidence attributes are updated. Then, the candidate edge screening and graph completion process in step S3 is repeated to obtain an updated breathomics knowledge graph.

[0145] Specifically, if the new data is exhaled VOCs detection data from the same source, the new sample data is merged with the original sample data to obtain an updated VOCs feature matrix. Subsequently, based on the updated feature matrix, the VOC screening, Pearson correlation coefficient calculation, dynamic threshold determination, significance test, and candidate edge mining processes in step S2 are re-executed to obtain an updated set of data-driven candidate edges. Then, based on the candidate edge classification, comprehensive confidence calculation, and edge supplementation rules in step S3, an updated set of edges to be supplemented is obtained. For existing data-driven edges in the current graph, if they exist in the updated set of edges to be supplemented, their Pearson correlation coefficient, two-sided test p-value, edge weight, and most recent update time are updated; if they no longer exist in the updated set of edges to be supplemented, the data-driven edge is invalidated and its status is marked as "superseded". Thus, a breathomics knowledge graph updated based on new sample data from the same source can be obtained.

[0146] Specifically, if the new data consists of samples from different disease subtypes, or data with significant categorical differences from the existing data, the new data is not directly mixed with the existing sample data for calculation. Instead, it is processed as an independent data group. For this type of data, the candidate edge mining process in step S2 and the comprehensive confidence screening and edge weight calculation process in step S3 are performed separately. Then, the original data-driven edges are not invalidated; instead, their categories are recorded in the graph database through edge attributes or subgraph identifiers. Thus, the data-driven relationships corresponding to different disease subtypes or different categories of sample data are stored as category subgraphs and incorporated into the overall graph.

[0147] Furthermore, to support continuous updates, historical tracking, and evidence backtracking of the knowledge graph, this embodiment sets version and time attributes for graph nodes, edges, and update activities. Specifically, "created_time" indicates the time when a node or edge was first written into the graph, "updated_time" indicates the most recent update time, "valid_from" and "valid_to" indicate the valid start and end times of the node or edge in the graph, "status" indicates the current status, "source" indicates the data source, and "graph_version" indicates the graph version. Through this method, each node and edge in the graph can trace its source, generation time, version, and current valid status, thereby supporting versioned management and time-series tracking of the graph.

[0148] Through the aforementioned dynamic update mechanism, the breathwork knowledge graph constructed in this invention can dynamically expand as publicly available database information, breath VOCs detection samples, and disease subtype data continue to increase. Compared to a static knowledge graph constructed all at once, this method can retain historical versions, track evidence sources, distinguish subgraphs corresponding to different data categories, and update the confidence and weight of edges when new evidence appears, thereby achieving dynamic knowledge graph construction for breathwork analysis of diabetes.

[0149] On the other hand, embodiments of the present invention provide a dynamic construction system for a breathomics knowledge graph based on metabolic pathways and data-driven approaches, used to implement the aforementioned dynamic construction method for a breathomics knowledge graph based on metabolic pathways and data-driven approaches. The system includes:

[0150] An initial knowledge network construction module is used to acquire compound, module and metabolic pathway information from a public metabolic pathway database, determine the metabolic relationships between target volatile organic compounds, and construct an initial knowledge network with prior constraints on metabolic pathways based on the metabolic relationships.

[0151] The data-driven candidate edge mining module is used to acquire exhaled volatile organic compound detection data, perform data preprocessing on the detection data, construct a feature matrix, mine candidate association relationships between volatile organic compounds based on the feature matrix to obtain a candidate association network, and filter the candidate association relationships in the candidate association network to obtain a data-driven candidate edge set.

[0152] The candidate edge screening and graph completion module is used to calculate the comprehensive confidence of the candidate edges in the data-driven candidate edge set, and to add the candidate edges that meet the confidence conditions to the initial knowledge network to form the completed breath mics knowledge graph.

[0153] The graph update module is used to perform incremental updates on the existing breathomics knowledge graph when new data is added.

[0154] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0155] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0156] Accordingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described method for dynamically constructing a breathomics knowledge graph based on metabolic pathways and data-driven principles. Figure 5 The diagram shown is a hardware structure diagram of any device with data processing capabilities for the dynamic construction method of exhalomics knowledge graph based on metabolic pathways and data-driven methods provided in this embodiment of the invention, except for... Figure 5 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0157] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the above-described method for dynamically constructing a breathomics knowledge graph based on metabolic pathways and data-driven principles. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0158] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only.

[0159] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

[0160] The above description is merely a preferred embodiment of the present invention. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the technical solutions of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall still fall within the protection scope of the technical solutions of the present invention.

Claims

1. A dynamic construction method for a breathomics knowledge graph based on metabolic pathways and data-driven approaches, characterized in that, include: S1. Obtain compound, module and metabolic pathway information from a publicly available metabolic pathway database, determine the metabolic relationships between target volatile organic compounds, and construct an initial knowledge network with prior constraints on metabolic pathways based on the metabolic relationships. S2, acquire breath volatile organic compound detection data, perform data preprocessing on the detection data, construct a feature matrix, and mine candidate association relationships between volatile organic compounds based on the feature matrix to obtain a candidate association network. Filter the candidate association relationships in the candidate association network to obtain a data-driven candidate edge set. S3, based on the prior knowledge of metabolic pathways, selects candidate edges and completes the knowledge graph. Specifically, it calculates the comprehensive confidence of the candidate edges in the data-driven candidate edge set obtained in S2, and adds the candidate edges that meet the confidence conditions to the initial knowledge network constructed in S1 to form the completed breath mics knowledge graph. S4. When new data on the detection of volatile organic compounds in exhaled air or new information in the publicly available metabolic pathway database are obtained, the existing exhaled air mics knowledge graph is incrementally updated.

2. The method according to claim 1, characterized in that, Step S1 is as follows: Obtain all compound names from the publicly available metabolic pathway database, and filter out compounds that exist simultaneously in both the publicly available metabolic pathway database and a pre-organized list of volatile organic compounds in human exhaled gas, denoting them as the target compound set. Extract module and metabolic pathway information related to the target compound set from the publicly available metabolic pathway database, and extract the adjacency relationships between the screened metabolic pathways; After standardizing the names and mapping the identifiers of the target compounds, metabolic relationships between the compounds are constructed, and an initial knowledge network is built based on these metabolic relationships.

3. The method according to claim 1, characterized in that, For any two target compounds, the metabolic association is as follows: (1) If they are located in the same metabolic pathway and directly connected, they are transformation relationships; (2) If they are located in the same module, they are module-related relationships; (3) If they are located in the same metabolic pathway and not directly connected, they are pathway-related relationships; (4) If they are located in two adjacent metabolic pathways, they are cross-pathway relationships; (5) If there is a clear upstream / downstream metabolic relationship, they are upstream / downstream metabolic relationships; (6) If they are both metabolized from the same compound, they are co-source pathway relationships. The edge weights corresponding to the above metabolic associations decrease sequentially from (1) to (6).

4. The method according to claim 1, characterized in that, In step S2, the screening of candidate associations specifically involves: The characteristic matrix of volatile organic compounds (VOCs) is denoted as... Where n represents the number of breath genome samples, p represents the number of VOC features, and x ki This represents the detection value of the i-th VOC in the k-th sample; Calculate the Pearson correlation coefficient r between the i-th VOC and the j-th VOC. ij and two-tailed test p-value ij ; Define a candidate Pearson correlation coefficient (PCC) threshold sequence. For any candidate PCC threshold t... l Construct the corresponding adjacency matrix and calculate the number of network edges based on the adjacency matrix; Traverse the candidate PCC threshold sequence and calculate the threshold when the candidate PCC threshold is t l Increase to t l+1 Rate of change of the number of sides; A preset threshold η is used to determine the abrupt change in the number of edges. If there exists a candidate PCC threshold whose rate of change in the number of edges is greater than η, then that candidate PCC threshold is used as the final dynamic threshold t. * If the rate of change of the number of edges corresponding to all candidate PCC thresholds is less than or equal to η, then the candidate PCC threshold with the largest rate of change of the number of edges is taken as the final dynamic threshold t. * ; When r is satisfied at the same time ij Greater than t * And p ij When the value is less than 0.05, it is considered that there is a data-driven candidate association relationship between the i-th VOC and the j-th VOC, thus obtaining the data-driven candidate edge set.

5. The method according to claim 4, characterized in that, In step S3, the comprehensive confidence C of the candidate edge is calculated by combining the data-driven evidence of the candidate edge with its metabolic prior support in the initial knowledge network. ij Specifically: For each data-driven candidate edge, a classification function g is defined. ij If both VOC nodes at the two ends of a candidate edge belong to the initial knowledge network, and there is already a prior metabolic association between them, then g ij =1; If both VOC nodes at the two ends of the candidate edge belong to the initial knowledge network, and there is no prior metabolic association between them, then g ij =2; if only one end of the candidate edge has a VOC node belonging to the initial knowledge network, then g ij =3; if neither of the VOC nodes at either end of the candidate edge belongs to the initial knowledge network, then g ij =4; Data-driven support D for calculating candidate edges ij The calculation formula is as follows: Where, ω d The weights of the relevant strength terms; According to the hierarchical function g ij Define the prior hierarchical support P of the candidate edges ij :g ij When =1, P ij =1; g ij When =2, P ij =0.75; g ij When =3, P ij =0.5; g ij When =4, P ij =0; Data-driven support based on candidate edges D ij Prior hierarchical support P ij Calculate the comprehensive confidence C of candidate edges. ij : Where, ω c The weight of data-driven evidence in the overall confidence level; The overall confidence level C of the candidate edges ij Candidate edges with a confidence level greater than or equal to the edge-filling threshold θ are included in the set of edges to be filled, and the combined confidence level C is set accordingly. ij As edge weight.

6. The method according to claim 5, characterized in that, In step S3, for candidate edges that meet the confidence criteria, they are added to the initial knowledge network. Specifically: if the two VOC nodes corresponding to the candidate edge already have a priori edges in the initial knowledge network, the edge is not added again, but the Pearson correlation coefficient, two-sided test p-value, and overall confidence score corresponding to the candidate edge are written into the graph database as attributes of the existing prior edge; if the two VOC nodes corresponding to the candidate edge do not have a priori edges in the initial knowledge network, the candidate edge is added to the initial knowledge network, and the Pearson correlation coefficient, two-sided test p-value, and overall confidence score corresponding to the candidate edge are also written into the graph database as attributes of the newly added candidate edge; if the VOC node corresponding to the candidate edge does not exist in the initial knowledge network, it is added to the data-driven VOC node set.

7. The method according to claim 1, characterized in that, In step S4, when adding new public metabolic pathway database information, first, step S1 is repeated to obtain the updated prior knowledge network; then, the updated prior knowledge network is merged with the prior part in the current knowledge graph; for newly added nodes and newly added prior edges, they are directly written into the graph database; for existing nodes and edges, their source, database version, last update time, and evidence attributes are updated; then, the candidate edge screening and graph completion process in step S3 is repeated to obtain the updated breath mics knowledge graph.

8. The method according to claim 1, characterized in that, In step S4, when adding new exhaled volatile organic compound detection data from the same source, the new sample data is first merged with the original sample data to obtain an updated feature matrix; then, based on the updated feature matrix, step S2 is re-executed to obtain an updated set of data-driven candidate edges, and then based on step S3, an updated set of edges to be supplemented is obtained; for existing data-driven edges in the current graph, if they exist in the updated set of edges to be supplemented, their Pearson correlation coefficient, two-sided test p-value, edge weight, and most recent update time are updated. If it no longer exists in the updated set of edges to be added, then the data-driven edge is invalidated.

9. The method according to claim 1, characterized in that, In step S4, if the newly added data is sample data of different disease subtypes, or data that has a significant class difference from the original data, it is treated as an independent data group for processing. First, the candidate edge mining process in step S2 and the comprehensive confidence screening process in step S3 are executed separately. Then, the category to which the edge belongs is recorded in the graph database by edge attributes or subgraph identifiers. The data-driven relationships corresponding to sample data of different disease subtypes or different categories are stored as category subgraphs and incorporated into the overall graph.

10. A dynamic construction system for a breathomics knowledge graph based on metabolic pathways and data-driven methods, characterized in that, The system for implementing the dynamic construction method of a breath omics knowledge graph as described in any one of claims 1-6 includes: An initial knowledge network construction module is used to acquire compound, module and metabolic pathway information from a public metabolic pathway database, determine the metabolic relationships between target volatile organic compounds, and construct an initial knowledge network with prior constraints on metabolic pathways based on the metabolic relationships. The data-driven candidate edge mining module is used to acquire exhaled volatile organic compound detection data, perform data preprocessing on the detection data, construct a feature matrix, mine candidate association relationships between volatile organic compounds based on the feature matrix to obtain a candidate association network, and filter the candidate association relationships in the candidate association network to obtain a data-driven candidate edge set. The candidate edge screening and graph completion module is used to calculate the comprehensive confidence of the candidate edges in the data-driven candidate edge set, and to add the candidate edges that meet the confidence conditions to the initial knowledge network to form the completed breath mics knowledge graph. The graph update module is used to perform incremental updates on the existing breathomics knowledge graph when new data is added.

Images