Disease drug association prediction method and system based on knowledge graph

Abstract

The invention discloses a disease drug association prediction method and system based on a knowledge graph, and solves the problem that effective fusion of a clinical knowledge graph and an omics knowledge graph in the existing knowledge graph is not realized. The method comprises the following steps: acquiring entities in a medical record, and acquiring disease entity information and disease relationship information according to an ICD-9 coding system; according to the disease entity information, obtaining corresponding drug information and drug relationship information in a DugBank database; fusing the disease entity information, the disease relationship information, the drug information and the drug relationship information link into a pre-constructed knowledge graph; inputting the disease and the drug into a prediction model trained by the fused knowledge graph to obtain a disease-drug association score; clinical data and omics data can be effectively fused, and the relationship between diseases and drugs can be predicted more accurately.

Description

Technical Field

[0001] The present invention relates to the field of medical information processing, and more specifically, to a method and system for predicting disease-drug associations based on a knowledge graph. Background Art

[0002] As a key research direction in the field of artificial intelligence, knowledge graph link prediction aims to predict missing entity relationships and potential entities in a knowledge graph. With the wide application of knowledge graph technology in multiple fields, link prediction technology has become a key task in the construction and completion process of knowledge graphs and has attracted extensive attention from researchers. However, existing knowledge graph link prediction methods still face some challenges:

[0003] (1) The construction of knowledge graphs is currently mainly divided into two categories: clinical knowledge graphs and omics knowledge graphs, and the effective integration of the two has not been achieved. This separation limits the potential of knowledge graphs in precision medicine because integrating clinical data with omics data can provide more comprehensive disease characteristics and treatment response information, thereby enhancing the personalization and accuracy of disease management.

[0004] (2) In the process of knowledge graph embedding representation, existing methods often fail to fully consider the information contained in different relationships in the knowledge graph. There are various types of relationships in the knowledge graph, carrying rich semantic information, but existing models have not effectively captured and utilized this relationship information in the representation learning process, limiting the overall improvement of the expression ability of the embedding vector.

[0005] In summary, there are still deficiencies in the clinical and omics data integration of knowledge graphs and the fusion mechanism of different relationship information, which limit the in-depth application and development of link prediction technology in fields such as precision medicine. Summary of the Invention

[0006] The purpose of the present invention is to solve the above problems by providing a method and system for predicting disease-drug associations based on a knowledge graph, which solves the problems of single knowledge graph data and the failure to consider relationships during feature extraction.

[0007] To solve the above problems, the present invention provides the following technical solutions:

[0008] On the one hand, a method for predicting disease-drug associations based on a knowledge graph includes

[0009] Obtaining entities in medical records and obtaining disease entity information and disease relationship information according to the ICD-9 coding system;

[0010] According to the disease entity information, obtaining corresponding drug information and drug relationship information in the DrugBank database;

[0011] Link and integrate disease entity information, disease relationship information, drug information, and drug relationship information into a pre-constructed knowledge graph;

[0012] Input diseases and drugs into a prediction model trained on the fused knowledge graph to obtain disease-drug association scores.

[0013] Furthermore, obtain the entities in the medical record, and obtain disease entity information and disease relationship information according to the ICD-9 coding system, including:

[0014] After obtaining the entity information in the medical record, these entities need to be preprocessed and standardized; among them, the entities specifically include disease names, drug information, surgical methods, clinical manifestations, medical history records, laboratory test results, and imaging examination materials;

[0015] According to the ICD-9 coding system, retrieve complication, gene, protein, phenotype, and small molecule disease entity information and disease relationship information related to the diseases and their subtypes in the medical record from public biomedical databases.

[0016] Furthermore, according to the disease entity information, obtain the corresponding drug information and drug relationship information in the DrugBank database, including:

[0017] In the DrugBank database, according to the identified disease entities, retrieve the related drug information and drug relationship information; compare the drug data obtained from DrugBank with the actual drug usage records in the medical record, and perform deduplication processing to ensure the accuracy and uniqueness of the drug information.

[0018] Furthermore, save the fused knowledge graph as a CSV file in UTF-8 encoding format, and use the LOADCSV statement of Neo4j to import the data in the CSV file into the Neo4j database.

[0019] Furthermore, in the prediction model, use the RGCN based on the attention mechanism as the encoder to process multiple types of relationships in the knowledge graph and learn the embedding vector representations of entities and relationships.

[0020] Furthermore, use Conv-TransE as the decoder in the prediction model. After stacking the disease embedding and relationship embedding, there is no reshaping, but directly perform convolution operations, and then calculate the final embedding vector through a linear transformation and a non-linear activation function. Finally, calculate the distance between the embedding vector and the relationship embedding to obtain the disease-drug association score.

[0021] Furthermore, the prediction model is based on ConvE but has a transformation with the translational characteristics of TransE.

[0022] Second aspect, a disease-drug association system based on a knowledge graph, for the disease-drug association prediction method described above; it includes:

[0023] Knowledge graph module: used to convert entities and relationships into three-way vectors of diseases, relationships, and drugs, store them in a graph database, and provide a visualization display function;

[0024] Embedding vector learning module: used to obtain low-dimensional vector representations of entities and relationships by using a pre-constructed relationship graph convolutional network (RGCN) encoder with an ECA channel attention mechanism;

[0025] Association prediction module: used to calculate the embedding vector distance between disease entities and drug entities for entity embedding vectors and relationship embedding vectors through a Conv-TransE decoder, and predict the potential association relationship between diseases and drugs.

[0026] Third aspect, an electronic device, including a memory and a processor, where computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the processor implements the disease-drug association prediction method described above.

[0027] Fourth aspect, a computer-readable storage medium, where a computer program is stored in the readable storage medium, and when the computer program runs on a computer, the computer implements the disease-drug association prediction method described above.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] (1) The present invention associates the data in medical records with a combined data including a public biomedical database and a DrugBank database through the ICD-9 coding system, overcoming the limitations of only having a clinical knowledge graph or an omics knowledge graph in the existing knowledge graph. The present invention integrally provides more comprehensive disease characteristics and treatment response information, enhancing the application potential of the knowledge graph in precision medicine.

[0030] (2) The present invention introduces an ECA channel attention mechanism, improving the sensitivity of the model to different types of relationships in the knowledge graph. This helps the model more accurately capture and utilize relationship information, thereby enhancing the expression ability of the embedding vector for improving the association prediction ability.

[0031] (3) The present invention uses Conv-TransE as the decoder. After stacking the disease embedding and the relationship embedding, the decoder directly performs a convolution operation without reshaping, and then calculates the final embedding vector through a linear transformation and a non-linear activation function. Finally, the distance between the embedding vector and the relationship embedding is calculated to obtain the disease-drug association score. This method efficiently preserves the local interaction features between diseases and drugs, enhances the model's ability to express complex association patterns, and thus improves the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings, where:

[0033] Figure 1 is the flowchart of the disease-drug association prediction method based on the knowledge graph of the present invention;

[0034] Figure 2 is the schematic diagram of the knowledge graph in the present invention;

[0035] Figure 3 is the schematic flowchart of the embedding representation of the knowledge graph in the present invention;

[0036] Figure 4 is the schematic flowchart of the disease-drug link prediction in the present invention;

[0037] Figure 5 is the functional module diagram of the disease-drug association prediction system based on the knowledge graph of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the following will further describe the present invention in detail. The described embodiments should not be regarded as limitations of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention. Figures 1 to 5 As shown in

[0039] As Figure 1 shown, a disease-drug association prediction method and system based on a knowledge graph provided by the present invention include the following steps:

[0040] S1: Define entity concepts and their relationships using a top-down method, construct the schema layer of the knowledge graph, and form an ontology concept framework;

[0041] S2: Obtain clinical omics data from electronic medical records and biomedical databases;

[0042] S3: According to the ontology concept framework, link relevant entities extracted from different sources to achieve knowledge fusion, and use Neo4j for knowledge storage;

[0043] S4: Based on the triples in the knowledge graph, obtain entity embedding representations using knowledge graph embedding technology;

[0044] S5: Input the disease and drug representation vectors into the disease-drug link prediction model to obtain the prediction results.

[0045] In the above step S1, according to the cyclic acquisition method and the Stanford seven-step method, the tool uses a top-down approach to define entity concepts and their relationships. The determined ontology includes seven categories: diseases, drugs, genes, proteins, small molecules, symptoms, and examination items.

[0046] The above step S2, obtaining clinical omics data from electronic medical records and biomedical databases includes the following sub-steps:

[0047] S21: Extract, preprocess, and standardize the key medical entities of diseases, drugs, symptoms, and examination items in the electronic medical records;

[0048] S22: According to the International Classification of Diseases, 9th Revision (ICD-9) coding system, identify the disease entities and map them to the corresponding numbers and names. Furthermore, according to the ICD-9 coding, retrieve the complication, gene, protein, phenotype, and small molecule disease entity information and disease relationship information related to the disease and its subtypes from public biomedical databases;

[0049] S23: In the DrugBank database, retrieve the drug information and relationship information related to the identified disease entities. In addition, compare the drug data obtained from DrugBank with the actual drug usage records in the electronic medical records, and perform deduplication processing to ensure the accuracy and uniqueness of drug information.

[0050] In the above step S3, the step of linking relevant entities extracted from different sources according to the ontology concept framework to achieve knowledge fusion and using Neo4j for knowledge storage is as follows:

[0051] S31: Save the extracted entities and their mutual relationships in UTF-8 encoding format as a CSV file;

[0052] S32: Use the LOAD CSV statement of Neo4j to import the data in the CSV file into the Neo4j database and perform visual display. The display effect is as Figure 2 shown.

[0053] In step S4, as Figure 3 shown, the RGCN based on the attention mechanism is used as the encoder. Each layer of the network uses the ECA channel attention mechanism to act on multiple types of relationships in the knowledge graph, assigns weights to different relationships, and obtains the knowledge graph embedding representation. The specific calculation formula is as follows:

[0054] The graph convolution operation in each layer of the RGCN is as follows:

[0055]

[0056] Among them, and are the feature representations of nodes i and j in the l-th layer, and are both mapping matrices, R is the set of relationship types, is the set of neighbor nodes j of node i, σ is the activation function, and α r is the weight of relationship r.

[0057] For each layer of the network, the ECA channel attention mechanism weights different relationships r, and α r is calculated as follows:

[0058]

[0059] Among them, exp() represents the exponential function, and LeakyReLU is the activation function.

[0060] In step S5, as Figure 4 shown, the disease and drug embedding representations are used as the inputs of the decoder of the disease-drug link prediction model, and a stacking operation is performed to obtain a composite embedding. Then, through a convolutional layer and a fully connected layer, a series of linear and non-linear transformation processes are carried out, and then the distance metric is used to calculate the link score between the embedding vector and the tail entity.

[0061] Furthermore, the disease-drug link prediction model uses Conv-TransE as the decoder. This model is based on ConvE but has the transformation of the translational property of TransE: e d +e r ≈e g . Among them, e d represents the disease embedding, e r represents the relationship embedding, and e g represents the drug embedding. The decoder stacks e d and e rThere is no reshaping and the convolution operation is directly performed. In the convolution, the size of the filter is 2×k, where k ∈ {1, 2, 3,...}, and the convolution calculation formula is as follows:

[0062]

[0063] where K is the kernel size, n refers to the entry index in the output vector, ω c is the trainable kernel parameter, and are the padded versions of the head entity e d and the relation e r respectively, and τ is the offset of the convolution kernel sliding on the input sequence. After the convolution operation, the matrix is obtained, and then the final embedding representation is calculated through a linear transformation and a non-linear activation function.

[0064] The scoring function is used to measure the rationality of the triple, and the formula is as follows:

[0065] ψ(e d , e g ) = f(vec(M(e d , e r )))W)e g

[0066] where, is the matrix of the linear transformation, vec(M) converts M(e d , e r ) into a long vector, and f represents the non-linear activation function. The predicted score is obtained by calculating the distance between the embedding vector and e g , and the calculation formula is as follows:

[0067] p(e d , e r , e g ) = σ(ψ(e d , e g ))

[0068] where p is the predicted score and σ is the activation function.

[0069] As Figure 5 shown, it is the functional module diagram of a disease-drug association prediction system provided by the present invention.

[0070] Furthermore, the knowledge graph module is used to convert relevant entities (such as diseases, drugs, symptoms, etc.) and relationships (such as treatment, side effects, associations, etc.) into the form of triples (d, r, g), store them in the graph database, and provide a visualization display function;

[0071] The embedded vector learning module is used to obtain the low-dimensional vector representations of entities and relationships by using the relationship graph convolutional network (RGCN) encoder with a pre-constructed ECA channel attention mechanism;

[0072] The association prediction module is used to calculate the embedding vector distance between disease entities and drug entities through a Conv-TransE decoder for the entity embedding vectors and relationship embedding vectors, so as to predict whether there is an association between them and give the confidence score of the association, and predict the potential association relationship between diseases and drugs.

[0073] The present invention first uses a top-down method to define entity concepts, attributes and relationships, constructs the schema layer of the knowledge graph, and forms an ontology concept framework; secondly, obtains relevant clinical omics data from electronic medical records and biomedical public databases; according to the ontology concept framework, links the relevant entities extracted from different sources and performs knowledge fusion processing to form triples, and uses the LOAD CSV statement to implement knowledge storage and visualization display through Neo4j; then, the triples in the knowledge graph are obtained through a relationship graph convolutional neural network based on the attention mechanism to obtain the knowledge graph embedding representation; finally, the disease and drug representation vectors are input into the decoder to predict the disease-drug association score. Therefore, the present invention provides an auxiliary tool for medical professionals and patients, improves the efficiency and accuracy of drug discovery, and reduces the overall time and cost of drug research and development.

[0074] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the steps of a disease-drug association prediction method based on a knowledge graph are implemented.

[0075] The electronic device can be a desktop computer, a notebook, a palm computer, a cloud server and other electronic devices. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the figure is only an example of the electronic device, and does not constitute a limitation on the electronic device. It may include more or fewer components than shown in the figure, or different components.

[0076] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0077] The memory can be an internal storage unit of an electronic device, for example, the hard disk or memory of the electronic device. The memory can also be an external storage device of the electronic device, for example, a plug-in hard disk equipped on the electronic device, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. The memory can also include both the internal storage unit of the electronic device and the external storage device. The memory is used to store computer programs and other programs and data required by the electronic device.

[0078] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each block in the flowchart or block diagram can represent a module, a program segment, or a part of code, and the module, program segment, or part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0079] In addition, each functional module in various embodiments of the present invention can be integrated together to form an independent part, or each module can exist alone, or two or more modules can be integrated to form an independent part.

[0080] When the above-described functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprise", "include" or any other variation thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the said element.

[0081] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention. It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0082] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or replacements, which should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A disease-drug association prediction method based on knowledge graph, characterized in that: include Obtain entities in medical records and obtain disease entity information and disease relationship information based on the ICD-9 coding system; According to the disease entity information, the corresponding drug information and drug relationship information are obtained in the DrugBank database; Integrate disease entity information, disease relationship information, drug information, and drug relationship information links into the pre-built knowledge graph; Diseases and drugs are input into the prediction model trained by the fused knowledge graph to obtain the disease-drug association score.

2. A disease-drug association prediction method based on knowledge graph according to claim 1, characterized in that: Obtain entities in medical records and obtain disease entity information and disease relationship information based on the ICD-9 coding system, including: After obtaining the entities in the medical records, they need to be preprocessed and standardized; entities include diseases, drugs, surgeries, clinical manifestations, medical history, laboratory tests, and imaging examinations; According to the ICD-9 coding system, complications, genes, proteins, phenotypes, and small molecule disease entity information and disease relationship information related to the diseases and their subtypes in the medical records were retrieved from public biomedical databases.

3. The disease-drug association prediction method based on knowledge graph according to claim 1, characterized in that: According to the disease entity information, the corresponding drug information and drug relationship information are obtained in the DrugBank database, including: In the DrugBank database, according to the identified disease entities, the relevant drug information and drug relationship information are retrieved; the drug data obtained from DrugBank are compared with the actual drug usage recorded in the medical records, and after deduplication processing, the corresponding drug information and drug relationship information are obtained.

4. The disease-drug association prediction method based on knowledge graph according to claim 1, characterized in that: The fused knowledge graph is saved as a CSV file in UTF-8 encoding format, and the data in the CSV file is imported into the Neo4j database using the Neo4j LOAD CSV statement.

5. The disease-drug association prediction method based on knowledge graph according to claim 1, characterized in that: In the prediction model, RGCN based on the attention mechanism is used as an encoder to process multi-type relations in the knowledge graph and learn the embedded vector representation of entities and relations.

6. The disease-drug association prediction method based on knowledge graph according to claim 1, characterized in that: Conv-TransE is used as the decoder in the prediction model. The decoder stacks the disease embedding and the relationship embedding without reshaping, but directly performs a convolution operation. Then, a linear transformation and a nonlinear activation function are used to calculate the final embedding vector. Finally, the distance between the embedding vector and the relationship embedding is calculated to obtain the disease-drug association score.

7. The disease-drug association prediction method based on knowledge graph according to claim 6, characterized in that: The prediction model is based on ConvE, but with the translation feature of TransE.

8. A disease-drug association system based on knowledge graph, characterized in that: Used to implement the disease-drug association prediction method according to any one of claims 1 to 7; It includes: Knowledge graph module: used to convert entities and relationships into ternary vectors of diseases, relationships, and drugs, and store them in a graph database to provide visualization functions; Embedding vector learning module: used to use the pre-built ECA channel attention mechanism relational graph convolutional network RGCN encoder to obtain low-dimensional vector representations of entities and relations; Association prediction module: used to calculate the embedding vector distance between disease entities and drug entities through the Conv-TransE decoder for entity embedding vectors and relationship embedding vectors, and predict the potential association relationship between diseases and drugs.

9. An electronic device, comprising a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the electronic device comprises: The processor is enabled to execute the disease-drug association prediction method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed on a computer, the computer is enabled to execute the disease-drug association prediction method according to any one of claims 1 to 7.

Images