Drug interaction knowledge graph construction method and system

Abstract

The invention provides a drug interaction knowledge graph construction method and system, and relates to the technical field of knowledge graphs, and the method comprises the steps: carrying out the attribute semantic annotation of each drug entity, and carrying out the division to obtain entity clusters of different action categories; extracting a path probability entropy of drug interaction between the entity clusters, and analyzing based on the path probability entropy and an indirect dependency relationship between every two drug entities to obtain an indirect logic relationship of pharmacological interaction between every two drug entities; determining a direct logic relationship of pharmacological interaction between every two drug entities according to the dependency of direct interaction between every two drug entities and the attribute semantic annotation information of each drug entity; evidence edge weights are generated through the indirect logic relation and the direct logic relation of pharmacological interaction between every two drug entities, and the drug interaction knowledge graph is constructed based on the evidence edge weights. By adopting the scheme of the invention, the knowledge graph can be constructed based on the direct evidence component and the indirect evidence component of the drug interaction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of knowledge graph, more particularly, the present application relates to a drug interaction knowledge graph construction method and system. BACKGROUND

[0002] Under the background of the intelligent development of medical information, as an important tool for assisting drug research and development, safety evaluation and clinical decision-making, drug interaction knowledge graph is widely used in pharmacological relationship mining and semantic reasoning tasks. By modeling drug entities and their interactions in a structured form, knowledge graph can provide global visual knowledge expression, improve the organization efficiency and correlation mining ability of drug information, and has become an important support means for intelligent drug information processing.

[0003] Existing drug interaction knowledge graph construction methods generally rely on explicit relationship extraction and rule matching, and pay too much attention to the direct association information between drug entities, ignoring the potential multi-hop path, implicit dependency structure and other indirect interaction characteristics existing in large-scale data. This method is prone to information loss and relationship weakening when dealing with complex pharmacological mechanisms, identifying non-explicit interference paths and abstracting semantic relationships, resulting in incomplete graph structure, unbalanced edge weight distribution, insufficient reasoning ability, and difficulty in accurately depicting the potential interaction risks and action patterns between drugs. Therefore, how to construct a knowledge graph based on the direct evidence component and the indirect evidence component of drug interaction has become a difficult problem in the industry. SUMMARY

[0004] The present application provides a drug interaction knowledge graph construction method and system, which can construct a knowledge graph based on the direct evidence component and the indirect evidence component of drug interaction.

[0005] In a first aspect, the present application provides a drug interaction knowledge graph construction method, comprising the following steps: Obtaining a plurality of drug entities, performing attribute semantic labeling on each drug entity, and classifying the corresponding drug entity according to the attribute semantic labeling result to obtain entity clusters of different action categories; Performing latent association path analysis on the drug attributes between each entity cluster, and then extracting the path probability entropy of the drug interaction between each entity cluster, obtaining the indirect logical relationship of the pharmacological interaction between each two drug entities based on all path probability entropies and the indirect dependency relationship between each two drug entities; Determining the dependency degree of the direct interaction between each two drug entities according to the attribute semantic relationship between each two drug entities, and then determining the direct logical relationship of the pharmacological interaction between each two drug entities from all dependency degrees and the attribute semantic labeling information of each drug entity; The evidence edge weight representing the interaction relationship between the drug entities is generated through the indirect logical relationship and the direct logical relationship of the pharmacological interaction between each two drug entities, and then a knowledge graph reflecting the drug interaction relationship is constructed based on all the evidence edge weights and all the drug entities.

[0006] Preferably, the corresponding drug entities are classified according to the attribute semantic annotation result, and entity clusters of different action categories are obtained, which specifically include: The core attribute features of each drug entity are extracted from the attribute semantic annotation result, and the core attribute features include action targets, indications, and pharmacological action categories; An attribute feature vector of each drug entity is constructed based on the core attribute features; The attribute feature vectors with similar core attribute features are classified to obtain entity clusters of different action categories.

[0007] Preferably, the potential association path analysis of the drug attributes between each entity cluster is performed, and then the path probability entropy of the drug interaction between each entity cluster is extracted, which specifically includes: The attribute information of the drugs in each entity cluster is extracted; Based on the attribute information, the association paths existing between each entity cluster are identified; The frequency of each association path appearing in historical drug interaction data is counted, and then the occurrence probability of each association path is determined; The path probability entropy of the drug interaction between each entity cluster is determined according to the occurrence probability of all the association paths.

[0008] Preferably, the indirect logical relationship of the pharmacological interaction between each two drug entities is obtained based on all the path probability entropies and the indirect dependency relationship between each two drug entities, which specifically includes: The association relationship formed by each two drug entities through a third-party drug entity is extracted from a drug interaction database, and then the indirect dependency relationship between each two drug entities is determined; The association paths between the entity clusters corresponding to the path probability entropies in the preset interval are screened out; The indirect dependency relationship between each two drug entities is matched with the screened association paths, and the effective indirect action path is identified; The indirect logical relationship of the pharmacological interaction between each two drug entities is determined according to the effective indirect action path.

[0009] Preferably, the dependency degree of the direct interaction between each two drug entities is determined according to the attribute semantic relationship between the two drug entities, which specifically includes: The attribute semantic relationship between each two drug entities is determined; determine the dependency degree of the direct interaction between each two drug entities based on the correlation matching degree of the attribute semantic relationship.

[0010] Preferably, the direct logical relationship of the pharmacological interaction between each two drug entities determined by all the dependency degrees and the attribute semantic annotation information of each drug entity specifically includes: set a dependency threshold, and then screen out the drug entity pair with a dependency degree higher than the dependency threshold; extract the attribute semantic annotation information of the two drug entities in the screened drug entity pair, and then determine whether there is a direct action attribute basis feature between the two drug entities; determine the direct logical relationship of the pharmacological interaction between each two drug entities according to the dependency degree corresponding to the two drug entities and the attribute basis feature.

[0011] Preferably, the knowledge graph reflecting the drug interaction relationship constructed based on all evidence edge weights and all drug entities specifically includes: each drug entity is taken as a node of the knowledge graph; according to the size of the evidence edge weight, a connection edge is established between the corresponding two drug entity nodes, and the evidence edge weight is taken as the attribute of the connection edge; integrate all nodes and connection edges to obtain the knowledge graph reflecting the drug interaction relationship.

[0012] In the second aspect, the present application provides a drug interaction knowledge graph construction system, comprising: an acquisition module for acquiring a plurality of drug entities, performing attribute semantic annotation on each drug entity, and classifying the corresponding drug entity according to the attribute semantic annotation result to obtain entity clusters of different action categories; a processing module for performing latent correlation path analysis on the drug attributes between each entity cluster, and then extracting the path probability entropy of the drug interaction between each entity cluster, and obtaining the indirect logical relationship of the pharmacological interaction between each two drug entities based on all path probability entropies and the indirect dependency relationship between each two drug entities; The processing module is also used to determine the dependency degree of the direct interaction between each two drug entities according to the attribute semantic relationship between each two drug entities, and then determine the direct logical relationship of the pharmacological interaction between each two drug entities from all the dependency degrees and the attribute semantic annotation information of each drug entity. An execution module is used to generate evidence edge weights representing the interaction relationship between drug entities through the indirect logical relationship and the direct logical relationship of the pharmacological interaction between each two drug entities, and then construct a knowledge graph reflecting the drug interaction relationship based on all evidence edge weights and all drug entities.

[0013] In a third aspect, the present application provides a computer device, comprising a memory and a processor, the memory stores a code, and the processor is configured to acquire the code and execute the drug interaction knowledge graph construction method.

[0014] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the drug interaction knowledge graph construction method.

[0015] The technical scheme provided by the embodiments of the present application has the following beneficial effects: In the embodiments of the present application, a plurality of drug entities are obtained, attribute semantic labels are added to each drug entity, and the corresponding drug entities are classified according to the attribute semantic label results to obtain entity clusters of different action categories. The drug attributes between each entity cluster are analyzed for potential association paths, and then the path probability entropy of the drug interaction between each entity cluster is extracted. The indirect logical relationship of the pharmacological interaction between each two drug entities is obtained based on all path probability entropies and the indirect dependency relationship between each two drug entities. The dependency degree of the direct interaction between each two drug entities is determined according to the attribute semantic relationship between each two drug entities, and then the direct logical relationship of the pharmacological interaction between each two drug entities is determined by all dependency degrees and the attribute semantic label information of each drug entity. The evidence edge weight representing the interaction relationship between drug entities is generated through the indirect logical relationship and the direct logical relationship of the pharmacological interaction between each two drug entities, and then a knowledge graph reflecting the drug interaction relationship is constructed based on all evidence edge weights and all drug entities.

[0016] It can be seen that the present application generates the evidence edge weight representing the interaction relationship between the drug entities through the indirect logical relationship and the direct logical relationship of the pharmacological interaction between every two drug entities, and then constructs the knowledge graph reflecting the drug interaction relationship based on all the evidence edge weights and all the drug entities. First, through attribute semantic annotation and entity classification, the drug entities are divided into semantic structures according to their action characteristics, which provides a semantic basis and classification structure for establishing a reasonable entity cluster relationship network, facilitating the improvement of the organization level of the graph and the semantic distinction of the entity relationship. Second, the indirect logical relationship of the pharmacological interaction between every two drug entities is obtained based on the path probability entropy of all paths and the indirect dependence relationship between every two drug entities. By mining the implicit association characteristics such as pharmacological attribute similarity, co-occurrence path probability distribution and multi-hop entity dependence path, the semantic blank areas that cannot be covered by direct evidence can be completed, the modeling capability of the graph for non-explicit mechanisms can be enhanced, and the problems of sparse and uneven edge weight distribution can be improved. Third, the direct logical relationship of the pharmacological interaction between every two drug entities is determined by the dependence degree of the direct interaction between every two drug entities and the attribute semantic annotation information of each drug entity, which can ensure the accuracy and authority of the graph in the construction of the core interaction chain, and effectively support the clarification of the main structure of the graph and the improvement of the trustworthiness. Finally, the indirect logical relationship and the direct logical relationship of the pharmacological interaction between drug entities are mapped to the evidence edge weight, and then the knowledge graph reflecting the drug interaction relationship is constructed. The scheme can realize the interaction credibility measurement of multi-source evidence fusion by mapping the direct and indirect evidence into unified graph edge weight, improve the performance of the graph in terms of structural integrity and semantic continuity, and ensure that the graph has stronger stability and expandability in high-dimensional semantic space. In summary, the present application can construct a knowledge graph based on the direct evidence component and the indirect evidence component of drug interaction, thereby improving the recognition accuracy and coverage of drug interaction relationship, effectively making up for the neglect of implicit action paths and multi-hop associations by traditional methods, and enhancing the structural integrity and semantic expression capability of the knowledge graph. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is an exemplary flowchart of a method for constructing a drug interaction knowledge graph according to some embodiments of the present application; Figure 2 is a schematic diagram of an application scenario of a system for constructing a drug interaction knowledge graph according to some embodiments of the present application; Figure 3 is a flowchart of determining a direct logical relationship according to some embodiments of the present application; Figure 4 is a structural schematic diagram of a system for constructing a drug interaction knowledge graph according to some embodiments of the present application; Figure 5is a structural schematic diagram of a computer device for implementing a drug interaction knowledge graph construction method according to some embodiments of the present application. DETAILED DESCRIPTION

[0018] In order to better understand the technical solutions of the present application, the technical solutions of the present application will be described in detail below in combination with the drawings in the specification and specific embodiments.

[0019] Reference Figure 1 The figure is an exemplary flowchart of a drug interaction knowledge graph construction method according to some embodiments of the present application, which mainly includes the following steps: In step 101, a plurality of drug entities are obtained, each drug entity is semantically annotated with attributes, and the corresponding drug entity is classified according to the attribute semantic annotation result to obtain entity clusters of different action categories.

[0020] In some embodiments, as shown in Figure 2 The figure is a schematic diagram of an application scenario of a drug interaction knowledge graph construction system according to some embodiments of the present application, which includes three main components: a collection device, a server and a data storage device. The collection device is used to collect drug data, transmit the drug data to the server through a communication network, run the execution code of the drug interaction knowledge graph construction system in the server, and finally store the knowledge graph obtained by execution in the data storage device through the server.

[0021] It should be noted that the drug entity in the present application refers to a drug name or a drug component with independent semantics and attributes in drug data, which is used as a basic node unit representing pharmacological relationships in a knowledge graph.

[0022] It should also be noted that the attribute semantic annotation of each drug entity in the present application refers to assigning an attribute label to the drug entity that can represent semantic information such as pharmacological action, component characteristics or indications.

[0023] In some embodiments, the classification of the corresponding drug entity according to the attribute semantic annotation result to obtain entity clusters of different action categories can be implemented by the following steps: Extract the core attribute features of each drug entity from the attribute semantic annotation result, the core attribute features including action targets, indications, and pharmacological action categories; Construct an attribute feature vector of each drug entity based on the core attribute features; Classify the attribute feature vectors with similar core attribute features to obtain entity clusters of different action categories.

[0024] It should be noted that the core attribute feature in the present application refers to key attribute information capable of reflecting the drug action mechanism; the entity cluster in the present application refers to a category set formed by clustering drug entities with similar attribute features, representing a drug population with similar pharmacological functions.

[0025] In a specific implementation, first, the core attribute feature of each drug entity can be extracted from the attribute semantic annotation result in the following manner: the core attribute feature of each drug entity can be obtained from the attribute semantic annotation result, and the core attribute feature includes the action target point, the indication, and the pharmacological action category; second, the attribute feature vector of each drug entity can be constructed based on the core attribute feature in the following manner: the action target point, the indication, and the pharmacological action category are converted into vector representation through one-hot encoding or multi-hot encoding, and the converted vector is taken as the attribute feature vector; it should be noted that the influence of different dimensions can be eliminated by normalizing the elements in the vector; then, the attribute feature vectors with similar core attribute features are classified to obtain entity clusters of different action categories in the following manner: the K-means algorithm is used for unsupervised clustering of all attribute feature vectors, wherein the K-means needs to first calculate the within-cluster sum of squares under different K values to determine the optimal clustering number through the elbow method, so that multiple clusters can be obtained through unsupervised clustering, and the obtained clusters are taken as entity clusters with similar action mechanisms.

[0026] In step 102, the potential association path analysis of the drug attributes between each entity cluster is performed, and then the path probability entropy of the drug interaction between each entity cluster is extracted, and the indirect logical relationship of the pharmacological interaction between each two drug entities is analyzed based on all path probability entropies and the indirect dependence relationship between each two drug entities.

[0027] In some embodiments, the potential association path analysis of the drug attributes between each entity cluster is performed, and then the path probability entropy of the drug interaction between each entity cluster is extracted in the following steps: Extracting attribute information of drugs in each entity cluster; Identifying the association paths existing between each entity cluster based on the attribute information; Statistically analyzing the frequency of each association path appearing in historical drug interaction data, and then determining the occurrence probability of each association path; Determining the path probability entropy of the drug interaction between each entity cluster according to the occurrence probability of all association paths.

[0028] It should be noted that the path probability entropy in the present application refers to an index for measuring the uncertainty of all potential association paths between each entity cluster in the historical drug interaction data, reflecting the complexity of the attribute association relationship between drugs.

[0029] In a specific implementation, first, the attribute information of the drugs in each entity cluster can be implemented in the following manner, that is, for each entity cluster, the field features of each drug entity attribute feature in the entity cluster are obtained from the attribute semantic annotation result, and the attribute features specifically include target points, indications, and pharmacological action categories. Then, the field features of all attribute features are combined into a set, and the set is taken as the attribute set of the drugs in the entity cluster, and thus the attribute set of the drugs in each entity cluster is obtained. Second, the association paths existing between each entity cluster can be implemented in the following manner based on the attribute information, that is, the association path patterns across the entity clusters are constructed based on the attribute information, such as “drug A-target point X-drug B” or “drug C-indication Y-drug D”. The two ends of the path come from different clusters, and the intermediate nodes of the path are shared attribute entities. The cross-cluster relationship is established by means of multi-hop connection. Further, the breadth-first search traversal algorithm is applied in the pre-constructed drug interaction network to search for all reachable cross-cluster paths from each drug entity under the conditions of limiting path length (which can be set to 2 hops or 3 hops) and attribute type, and the structure representation of each unique path is recorded by using a hash structure, and thus the association paths existing between each entity cluster are obtained. Third, the frequency of each association path appearing in the historical drug interaction data is counted, and then the occurrence probability of each association path can be implemented in the following manner, that is, all records can be extracted from the historical drug interaction database, each association path is traversed, the number of times of occurrence of the path in the records is counted, and the occurrence probability is obtained by dividing the number of times by the total number of records. For example, if a path appears 200 times and the total number of records is 1000 times, the occurrence probability is 0.2. Finally, the path probability entropy of the drug interaction between each entity cluster can be implemented in the following manner based on the occurrence probability of all association paths, that is, the information entropy formula is used, the occurrence probability of all association paths is substituted, the logarithm of the occurrence probability of each path is multiplied by the probability, the sum of all results is calculated and the inverse number is taken, and the obtained value is taken as the path probability entropy of the drug interaction between the corresponding entity clusters. The obtained path probability entropy can be used as an important quantitative index reflecting the closeness of the potential pharmacological contact and the information complexity between clusters.

[0030] In some embodiments, the indirect logical relationship of the pharmacological interaction between each two drug entities can be obtained based on all path probability entropies and indirect dependence relationship analysis between each two drug entities in the following steps: extracting, from the drug interaction database, an association relationship formed by a third-party drug entity between each two drug entities, and determining an indirect dependency relationship between each two drug entities; screening an association path between entity clusters corresponding to a preset interval of path probability entropy; matching the indirect dependency relationship between each two drug entities with the screened association path, and identifying an effective indirect action path; determining an indirect logical relationship of pharmacological interaction between each two drug entities according to the effective indirect action path.

[0031] It should be noted that the drug interaction database in the present application refers to a structured data resource collecting interaction relationships and mechanism information between different drugs; the indirect dependency relationship in the present application refers to an index reflecting the indirect association degree between two drug entities in the absence of direct association through shared intermediary entities; the association path in the present application refers to an attribute chain connecting two drug entities, used to represent a potential relationship path established through shared attributes or intermediary nodes in a knowledge graph; and the indirect logical relationship in the present application refers to a potential pharmacological association indirectly embodied between two drug entities through shared intermediary attributes.

[0032] In a specific implementation, firstly, the association relationship between each two drug entities through a third-party drug entity is extracted from a drug interaction database, and then the indirect dependency relationship between each two drug entities can be determined by the following method, that is, all drug interaction relationships in the form of triplets are retrieved from the drug interaction database (such as DrugBank), for example, “drug A-drug C-drug B”, and the indirect association path between drug A and drug B through the intermediary drug C is extracted through graph structure analysis, and then the extracted path relationship is mapped to the indirect dependency relationship between the two drug entities, that is, the indirect dependency relationship set between all drug pairs is obtained; secondly, the association path between the entity clusters with the path probability entropy in the preset interval can be screened by the following method, that is, the association paths between all entity clusters and the corresponding path probability entropy values are summarized to form a path-entropy mapping table, and then the upper and lower threshold values of the preset interval (such as setting the threshold range based on the median of the distribution of all entropy values) are set to screen out paths with too little information (indicating high repetition rate and low differentiation) or too much information (indicating extremely rare and lack of representativeness), and then the path entropy between all entity clusters is compared, and only the paths meeting the interval constraint are retained, that is, a set of high-confidence association paths between entity clusters is obtained as the candidate path basis for subsequent matching of indirect logical relationships between drugs; then, the indirect dependency relationship between each two drug entities and the screened association path are matched, and the effective indirect action path can be identified by the following method, that is, the drug pairs in the indirect dependency relationship and the effective paths screened by the path probability entropy are structurally matched, which can be achieved by comparing the path topological structure or the path vector cosine similarity, to realize the coincidence identification of the actual indirect path between drugs and the theoretical high-confidence path, and the matching successful path is regarded as the effective indirect action path with statistical support and structural basis; finally, the indirect logical relationship of the pharmacological interaction between each two drug entities can be determined according to the effective indirect action path by the following method, that is, the indirect logical relationship between the corresponding drug entity pairs is established according to these effective paths, which is represented as “drug A has potential pharmacological action with drug B through path P”, and the matching strength of the path is taken as the relationship confidence, and then the relationship confidence and the corresponding indirect action path vector are used to describe the characteristic quantity of the indirect logical relationship of the pharmacological interaction between two drug entities.

[0033] In step 103, the dependency degree of the direct interaction between each two drug entities is determined according to the attribute semantic relationship between the drug entities, and then the direct logical relationship of the pharmacological interaction between each two drug entities is determined by all dependency degrees and attribute semantic annotation information of each drug entity.

[0034] In some embodiments, determining the dependency degree of the direct interaction between each pair of drug entities according to the attribute semantic relationship between each pair of drug entities can be achieved by the following steps: determining the attribute semantic relationship between each pair of drug entities; determining the dependency degree of the direct interaction between each pair of drug entities based on the association matching degree of the attribute semantic relationship.

[0035] It should be noted that the attribute semantic relationship in the present application is a measure of the semantic similarity between the key attribute features of two drug entities; the dependency degree of the direct interaction in the present application is an index for measuring the association strength of two drug entities in the attribute semantic space.

[0036] In specific implementation, first, determining the attribute semantic relationship between each pair of drug entities can be achieved by the following manner, that is, for each pair of drug entities, extracting their respective attribute semantic vectors, the attributes including the structured fields of action target, indication, metabolic pathway, pharmacokinetic parameters, and generating a vector representation by using a unified encoding method, the present application uses one-hot encoding method, then calculating the association matching degree between the two drug attribute semantic vectors, the association matching degree calculation can be quantified by using cosine similarity, which is used to measure the closeness of two drugs in the semantic space, and the obtained association matching degree is used to describe the attribute semantic relationship between the two drug entities; then, determining the dependency degree of the direct interaction between each pair of drug entities based on the association matching degree of the attribute semantic relationship can be achieved by the following manner, that is, normalizing the association matching degree of the attribute semantic relationship, and then taking the result obtained by the normalization as the dependency degree of the direct interaction between each pair of drug entities.

[0037] In some embodiments, referring to Figure 3 The figure is a flowchart for determining the direct logical relationship in some embodiments of the present application, and the direct logical relationship of the pharmacological interaction between each pair of drug entities in the present embodiment is determined by all dependency degrees and attribute semantic annotation information of each drug entity, which can be achieved by the following steps: In step 1031, a dependency degree threshold is set, and then drug entity pairs with a dependency degree higher than the dependency degree threshold are screened out; In step 1032, the attribute semantic annotation information of the two drug entities in the screened drug entity pair is extracted, and then it is determined whether the two drug entities have a direct action attribute basis feature; In step 1033, the direct logical relationship of the pharmacological interaction between each pair of drug entities is determined according to the dependency degree corresponding to the two drug entities and the attribute basis feature.

[0038] It should be noted that the attribute basis in the present application is a feature for measuring whether there is a direct pharmacological action between two drug entities; the direct logical relationship in the present application is a structural association index reflecting that two drug entities directly form a pharmacological action due to attribute matching and mutual dependence.

[0039] In specific implementation, first, a dependency threshold is set, and then drug entity pairs with a dependency higher than the dependency threshold are screened out, which can be achieved in the following manner: a reasonable dependency threshold is set based on the dependency of the direct interaction between historical drug entities, and drug entity pairs with a dependency higher than the set dependency threshold are screened out through the set dependency threshold, ensuring that the screened candidate pairs have strong attribute correlation; then, the attribute semantic annotation information of the two drug entities in the screened drug entity pairs is extracted, and then the attribute basis feature of whether the two drug entities have a direct action can be achieved in the following manner: for each screened drug entity pair, the attribute semantic annotation features of the two drug entities in the drug entity pair are obtained from the attribute semantic annotation results, and then the attribute semantic annotation features of the two drug entities are taken as the input parameters of the pre-trained attribute matching model, and then the attribute matching model is used to analyze whether the two drug entities have a direct action, and the output result is taken as the attribute basis feature of whether the two drug entities have a direct action; it should be further noted that the attribute matching model in the present application is a discrimination model for discriminating whether two entities have a correlation relationship at the attribute semantic level, and its core lies in semantic representation and matching degree calculation of input attribute features of two entities, and the attribute matching model is based on a deep learning structure such as an attention mechanism network, which encodes attribute semantic features (including pharmacological mechanism, target information and metabolic characteristics) of two drug entities respectively, and then performs similarity calculation in a shared vector space to capture the co-occurrence pattern and semantic consistency between attributes, and finally outputs a matching score for representing the attribute basis of whether two drugs have a potential direct interaction; finally, the direct logical relationship of pharmacological interaction between each two drug entities can be achieved in the following manner: the dependency of the two drug entities and the attribute basis feature can be constructed into a joint vector as a description feature representing the direct logical relationship of pharmacological interaction between the two drug entities.

[0040] In step 104, the evidence edge weight representing the interaction relationship between drug entities is generated through the indirect logical relationship and the direct logical relationship of pharmacological interaction between each two drug entities, and then a knowledge graph reflecting the drug interaction relationship is constructed based on all evidence edge weights and all drug entities.

[0041] In some embodiments, the evidence edge weight representing the interaction relationship between each two drug entities generated by the indirect logical relationship and the direct logical relationship of the pharmacological interaction between each two drug entities can be achieved by the following steps: determining an indirect evidence component representing the interaction relationship between each two drug entities according to the indirect logical relationship of the pharmacological interaction between each two drug entities; determining a direct evidence component representing the interaction relationship between each two drug entities according to the direct logical relationship of the pharmacological interaction between each two drug entities; fusing the indirect weight component and the direct weight component to obtain the evidence edge weight representing the interaction relationship between each two drug entities.

[0042] It should be noted that the indirect evidence component in the present application is a reasoning support quantity for measuring the strength of the potential mutual pharmacological interaction relationship between two drug entities; the direct evidence component in the present application is a characteristic index for measuring the strength of the direct pharmacological interaction relationship between two drug entities; and the evidence edge weight in the present application is an index for measuring the strength of the pharmacological interaction evidence between drug entities.

[0043] In specific implementation, firstly, determining an indirect evidence component representing the interaction relationship between each two drug entities according to the indirect logical relationship of the pharmacological interaction between each two drug entities can be achieved by the following manner, i.e., for each two drug entities, the relationship confidence in the indirect logical relationship of the pharmacological interaction between the two drug entities is normalized, and the value obtained by the normalization is taken as the indirect evidence component of the interaction relationship between the two drug entities; then, determining a direct evidence component representing the interaction relationship between each two drug entities according to the direct logical relationship of the pharmacological interaction between each two drug entities can be achieved by the following manner, i.e., for each two drug entities, the dependence degree and attribute basis characteristics corresponding to the two drug entities are obtained from the direct logical relationship of the pharmacological interaction between the two drug entities, the dependence degree and the attribute basis characteristics are fused, the value obtained by the fusion is normalized, and the value obtained by the normalization is further taken as the direct evidence component of the interaction relationship between the two drug entities; finally, fusing the indirect weight component and the direct weight component to obtain the evidence edge weight representing the interaction relationship between each two drug entities can be achieved by the following manner, i.e., for each two drug entities, the product of the indirect weight component and the direct weight component is normalized, and the result obtained by the normalization is taken as the evidence edge weight of the interaction relationship between the two drug entities, and thus the evidence edge weight of the interaction relationship between each two drug entities can be obtained by the above manner.

[0044] It should be noted that the scheme in the present application effectively solves the problem of incomplete information and insufficient accuracy caused by single evidence source in traditional drug interaction recognition by fusing the evidence components of indirect logical relationship and direct logical relationship. The indirect evidence is used to capture potential associations in multi-hop paths, and the direct evidence is used to reflect the explicit connection between drug attributes, thereby realizing multi-dimensional comprehensive evaluation of drug interaction relationship. The fusion of evidence edge weight enhances the confidence and expression ability of the interaction relationship, and significantly improves the accuracy and robustness of relationship discrimination in the knowledge graph.

[0045] In some embodiments, constructing a knowledge graph reflecting drug interaction relationship based on all evidence edge weights and all drug entities can be achieved by the following steps: Each drug entity is taken as a node of the knowledge graph; According to the size of the evidence edge weight, a connection edge is established between the corresponding two drug entity nodes, and the evidence edge weight is taken as the attribute of the connection edge; Integrating all nodes and connection edges, a knowledge graph reflecting drug interaction relationship is obtained.

[0046] It should be noted that the knowledge graph in the present application refers to a graph structure reflecting drug entities and their interaction relationships.

[0047] In specific implementation, firstly, each drug entity is taken as a node of the knowledge graph can be achieved by the following way, i.e., each drug entity is mapped to a node in the graph structure; then, according to the size of the evidence edge weight, a connection edge is established between the corresponding two drug entity nodes, and the evidence edge weight is taken as the attribute of the connection edge can be achieved by the following way, i.e., based on the obtained evidence edge weight, a graph edge structure is constructed using a graph database (such as Neo4j), for each pair of drug entity nodes with interaction relationship, whether to connect edges and the edge weight value are determined according to the evidence edge weight value, and the weight value is usually taken as the attribute of the graph edge to reflect the interaction strength or credibility; finally, integrating all nodes and connection edges, a knowledge graph reflecting drug interaction relationship is obtained can be achieved by the following way, i.e., all nodes and edges are integrated into a graph data structure to form a complete heterogeneous drug interaction graph, thereby obtaining a knowledge graph reflecting drug interaction relationship.

[0048] On the other hand, in some embodiments, the present application provides a drug interaction knowledge graph construction system, referring to Figure 4 The figure is a structural schematic diagram of the drug interaction knowledge graph construction system according to some embodiments of the present application. The drug interaction knowledge graph construction system 400 includes an acquisition module 401, a processing module 402 and an execution module 403, which are described as follows. The obtaining module 401 is mainly used for obtaining a plurality of drug entities, performing attribute semantic labeling on each drug entity, and classifying the corresponding drug entity according to the attribute semantic labeling result to obtain entity clusters of different action categories. The processing module 402 is used for performing potential correlation path analysis on the drug attributes between each entity cluster, and further extracting path probability entropy of the drug interaction between each entity cluster, to obtain an indirect logical relationship of pharmacological interaction between each two drug entities based on all path probability entropies and an indirect dependency relationship between each two drug entities. The processing module 402 is also used for determining a dependency degree of direct interaction between each two drug entities according to the attribute semantic relationship between the two drug entities, and further determining a direct logical relationship of pharmacological interaction between each two drug entities based on all dependency degrees and attribute semantic labeling information of each drug entity. The execution module 403 is mainly used for generating evidence edge weight representing the interaction relationship between drug entities through the indirect logical relationship and the direct logical relationship of pharmacological interaction between each two drug entities, and further constructing a knowledge graph reflecting the drug interaction relationship based on all evidence edge weights and all drug entities.

[0049] In addition, the present application also provides a computer device, which comprises a memory and a processor, the memory stores code, and the processor is configured to acquire the code and execute the above-mentioned drug interaction knowledge graph construction method.

[0050] In some embodiments, with reference to Figure 5 The figure is a structural schematic diagram of a computer device for implementing the drug interaction knowledge graph construction method according to some embodiments of the present application. The drug interaction knowledge graph construction method in the above-mentioned embodiments can be implemented by the computer device shown in the figure. Figure 5 The computer device 500 comprises at least one processor 501, a communication bus 502, a memory 503 and at least one communication interface 504.

[0051] The processor 501 can be a general central processing unit (CPU) or an application specific integrated circuit (ASIC).

[0052] The communication bus 502 can be used for transmitting information between the above-mentioned components.

[0053] The memory 503 can be a read only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read only memory (EEPROM), a compact disc read only memory (CD ROM) or other optical disk storage, a magnetic disk or other magnetic storage device, or any other medium capable of storing desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to. The memory 503 can exist independently, and is connected to the processor 501 through the communication bus 502. The memory 503 can also be integrated with the processor 501.

[0054] The memory 503 is configured to store program code for implementing the solutions of the present application, and the processor 501 is configured to execute the program code stored in the memory 503. The program code can include one or more software modules. The method for constructing a drug interaction knowledge graph in the above embodiments can be implemented by one or more software modules in the program code of the processor 501 and the memory 503.

[0055] The communication interface 504 is configured to communicate with other devices or communication networks, such as an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), etc., using any transceiver-like device.

[0056] In a specific implementation, as an embodiment, the computer device can include a plurality of processors, each of which can be a single CPU processor or a multi-CPU processor. The processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0057] The computer device described above can be a general-purpose computer device or a special-purpose computer device. In a specific implementation, the computer device can be a desktop computer, a laptop computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of the computer device.

[0058] In addition, the present application also provides a computer readable storage medium, the computer readable storage medium stores a computer program, the computer program is executed by a processor to realize the drug interaction knowledge graph construction method.

[0059] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to the embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to be interpreted as including all the preferred embodiments and all the changes and modifications falling within the scope of the present application.

[0060] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application also intends to include these modifications and variations.

Claims

1. A method for constructing a drug interaction knowledge graph, characterized in that, The method comprises the following steps: obtaining a plurality of drug entities, performing attribute semantic labeling on each drug entity, and classifying the corresponding drug entity according to the attribute semantic labeling result to obtain entity clusters of different action categories; performing potential correlation path analysis on the drug attributes between each entity cluster, and then extracting the path probability entropy of the drug interaction between each entity cluster; determining the indirect logical relationship of the pharmacological interaction between each two drug entities based on all path probability entropies and the indirect dependency relationship between each two drug entities; determining the dependency degree of the direct interaction between each two drug entities according to the attribute semantic relationship between each two drug entities, and then determining the direct logical relationship of the pharmacological interaction between each two drug entities from all dependency degrees and the attribute semantic labeling information of each drug entity; 2. The method of claim 1, wherein, generating the evidence edge weight representing the interaction relationship between drug entities through the indirect logical relationship and the direct logical relationship of the pharmacological interaction between each two drug entities, and then constructing a knowledge graph reflecting the drug interaction relationship based on all evidence edge weights and all drug entities. The classification of the corresponding drug entity according to the attribute semantic labeling result to obtain entity clusters of different action categories specifically comprises: extracting the core attribute features of each drug entity from the attribute semantic labeling result, wherein the core attribute features include action targets, indications, and pharmacological action categories; constructing the attribute feature vector of each drug entity based on the core attribute features; 3. The method of claim 1, wherein, classifying the attribute feature vectors with similar core attribute features to obtain entity clusters of different action categories. The potential correlation path analysis on the drug attributes between each entity cluster, and then extracting the path probability entropy of the drug interaction between each entity cluster specifically comprises: extracting the attribute information of drugs in each entity cluster; identifying the correlation paths existing between each entity cluster based on the attribute information; determining the occurrence probability of each correlation path by counting the frequency of each correlation path in historical drug interaction data; 4. The method of claim 1, wherein, determining the path probability entropy of the drug interaction between each entity cluster according to the occurrence probability of all correlation paths. The determination of the indirect logical relationship of the pharmacological interaction between each two drug entities based on all path probability entropies and the indirect dependency relationship between each two drug entities specifically comprises: extracting the correlation relationship between each two drug entities formed through a third-party drug entity from a drug interaction database, and then determining the indirect dependency relationship between each two drug entities; screening out the correlation paths between the entity clusters corresponding to the path probability entropy in the preset interval; matching the indirect dependency relationship between each two drug entities with the screened correlation paths to identify effective indirect action paths; 5. The method of claim 1, wherein, determining the indirect logical relationship of the pharmacological interaction between each two drug entities according to the effective indirect action paths. The determination of the dependency degree of the direct interaction between each two drug entities according to the attribute semantic relationship between each two drug entities specifically comprises: determining the attribute semantic relationship between each two drug entities; determining the dependency degree of the direct interaction between each two drug entities based on the correlation matching degree of the attribute semantic relationship.

6. The method of claim 1, wherein, The direct logical relationship of pharmacological interaction between each two drug entities is determined by all the dependency degrees and the attribute semantic annotation information of each drug entity, and specifically includes: A dependency degree threshold is set, and then drug entity pairs with a dependency degree higher than the dependency degree threshold are screened out; The attribute semantic annotation information of the two drug entities in the screened drug entity pairs is extracted, and then it is determined whether the two drug entities have a direct action attribute basis feature; The direct logical relationship of pharmacological interaction between each two drug entities is determined according to the dependency degree corresponding to the two drug entities and the attribute basis feature.

7. The method of claim 1, wherein, The knowledge graph reflecting the drug interaction relationship is constructed based on all evidence edge weights and all drug entities, and specifically includes: Each drug entity is taken as a node of the knowledge graph; According to the size of the evidence edge weight, a connection edge is established between the corresponding two drug entity nodes, and the evidence edge weight is taken as the attribute of the connection edge; All nodes and connection edges are integrated to obtain the knowledge graph reflecting the drug interaction relationship. 8.A drug interaction knowledge graph construction system, characterized by, It includes: An acquisition module is configured to acquire a plurality of drug entities, perform attribute semantic annotation on each drug entity, and classify the corresponding drug entity according to the attribute semantic annotation result to obtain entity clusters of different action categories; A processing module is configured to perform potential correlation path analysis on the drug attributes between each entity cluster, and then extract the path probability entropy of the drug interaction between each entity cluster, and obtain the indirect logical relationship of pharmacological interaction between each two drug entities based on all path probability entropies and the indirect dependency relationship between each two drug entities; The processing module is further configured to determine the dependency degree of the direct interaction between each two drug entities according to the attribute semantic relationship between the two drug entities, and then determine the direct logical relationship of pharmacological interaction between each two drug entities by all the dependency degrees and the attribute semantic annotation information of each drug entity; An execution module is configured to generate evidence edge weights representing the interaction relationship between drug entities through the indirect logical relationship and the direct logical relationship of pharmacological interaction between each two drug entities, and then construct a knowledge graph reflecting the drug interaction relationship based on all evidence edge weights and all drug entities. 9.A computer device, comprising a memory and a processor, wherein the memory stores code, and the code comprises the following steps: The processor is configured to acquire the code and execute the drug interaction knowledge graph construction method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the drug interaction knowledge graph construction method according to any one of claims 1 to 7.

Images