Cell knowledge graph construction method, device, computing device and storage medium

Abstract

The present disclosure provides a method for constructing a cell knowledge graph, comprising: determining a plurality of entity types and a plurality of relationship types in a cell-related field, each relationship type representing a relationship between two entities having any entity type among the plurality of entity types; acquiring a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities from a first data set, wherein the first data set includes a plurality of preset entities and a plurality of relationships corresponding to the plurality of entities; constructing a cell knowledge graph with the plurality of first entities as nodes and the plurality of first relationships as edges; acquiring a plurality of second entities and a second relationship corresponding to the plurality of second entities from a second data set, wherein the second data set includes literature data in the cell-related field; and updating the cell knowledge graph according to the plurality of second entities and the second relationships.

Description

Technical Field

[0001] The present disclosure relates to the biomedical field, and in particular, to a method and apparatus for constructing a knowledge graph corresponding to a cell-related field, a computing device, a computer-readable storage medium, and a computer program product. Background Art

[0002] With the continuous development of biomedicine, people have conducted research from various perspectives on different biological system levels (e.g., organs, tissues, etc.). In order to gain a deeper and more comprehensive understanding of human development, physiology, and cases, it is necessary to fully mine and analyze data related to the lower biological system level (e.g., cells).

[0003] In related technologies, high-throughput sequencing technology can be used to obtain large-scale single-cell transcriptome (single cell RNA, scRNA) related data, and then the scRNA data can be analyzed with the help of scRNA analysis tools (e.g., CellRanger) to obtain information such as highly expressed genes and cell marker genes in the scRNA data, thereby assisting subsequent cell type annotation, cell function analysis, etc. However, the above process of obtaining cells, genes and their relationships is relatively complicated, and the researchers' knowledge background and process proficiency will affect the accuracy of the results obtained, and the analysis results of the types and functions of cells obtained are difficult to directly associate with diseases, therapeutic drugs and other information that is helpful for precision medical research. In addition, the above process cannot utilize existing research results based on sequencing data (e.g., research results based on scRNA data to explore the pathogenesis of human diseases at the single-cell level). In addition, when the quality of scRNA data is low (e.g., there are large differences in steps such as single-cell culture separation process, sequencing method, sequencing instrument, sequencing batch, etc.), the analysis results may not be consistent with the actual situation. Summary of the invention

[0004] In view of this, the present disclosure provides a method and apparatus, a computing device, a computer-readable storage medium, and a computer program product for constructing a cell knowledge graph to alleviate, mitigate, or even eliminate the above-mentioned problems.

[0005] According to one aspect of the present disclosure, a method for constructing a cell knowledge graph is provided, comprising: determining a plurality of entity types and a plurality of relationship types in a cell-related field, each relationship type representing a relationship between two entities having any entity type among the plurality of entity types, wherein the plurality of entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe the functions of genes and their products; obtaining a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities from a first data set, the entity type of each first entity being one of the plurality of entity types, and the relationship type of each first relationship being one of the plurality of relationship types, wherein the first data set includes A plurality of preset entities and a plurality of relationships corresponding to the plurality of entities; constructing a cell knowledge graph corresponding to the cell-related field with the plurality of first entities as nodes and the plurality of first relationships as edges; obtaining a plurality of second entities and second relationships corresponding to the plurality of second entities from a second data set, wherein the entity type of each second entity is one of the plurality of entity types, and the relationship type of the second relationship is one of the plurality of relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the plurality of second entities and the second relationships; updating the cell knowledge graph according to the plurality of second entities and the second relationships.

[0006] According to some embodiments of the present disclosure, obtaining multiple second entities and second relationships corresponding to the multiple second entities from a second data set includes: using a trained graph element extraction model to obtain the multiple second entities and the second relationships from the second data set, wherein the graph element extraction model is trained by the following steps: obtaining a training data set, the training data set including sample document data, and sample entity data and sample relationship data including labels for the sample document data; inputting the sample document data into the graph element extraction model to obtain predicted entity data and predicted relationship data corresponding to the sample document data; determining a loss function based on the sample entity data, the predicted entity data, the sample relationship data, and the predicted relationship data; and adjusting the parameters of the graph element extraction model to minimize the loss function.

[0007] According to some embodiments of the present disclosure, the literature data in the cell-related field is obtained by the following steps: taking data related to literature in the cell-related field that meets screening criteria as the literature data, wherein the screening criteria include at least one of the following: publication date of the literature, number of citations of the literature, impact factor of the literature publishing platform, user rating of the literature on the corresponding publishing platform.

[0008] According to some embodiments of the present disclosure, the multiple entity types also include at least one of the following: protein, drug, pathway, disease, phenotype, and tissue, wherein the pathway is used to describe the interaction between molecules within a cell, the phenotype is used to describe the symptoms produced by the patient, and the tissue is composed of cells and intercellular matrix with similar morphology and function.

[0009] According to another aspect of the present disclosure, a method for predicting the relationship between entities is provided, comprising: obtaining two entities from the above-mentioned cell knowledge graph as a first prediction entity and a second prediction entity, wherein there is no directly connected edge between a node corresponding to the first prediction entity in the cell knowledge graph and a node corresponding to the second prediction entity in the cell knowledge graph; performing relationship prediction on the first prediction entity and the second prediction entity based on the cell knowledge graph to obtain a relationship prediction result, wherein the relationship prediction result is used to characterize the degree of association between the first prediction entity and the second prediction entity.

[0010] According to some embodiments of the present disclosure, the performing relationship prediction on the first prediction entity and the second prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: determining a first feature vector corresponding to the first prediction entity and a second feature vector corresponding to the second prediction entity, respectively, based on multiple entities and multiple relationships in the cell knowledge graph; calculating the similarity between the first feature vector and the second feature vector, and using the similarity as the relationship prediction result.

[0011] According to some embodiments of the present disclosure, the relationship prediction of the first prediction entity and the second prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: obtaining a reference entity from the cell knowledge graph, wherein a node corresponding to the reference entity in the cell knowledge graph and a node corresponding to the first prediction entity in the cell knowledge graph have a directly connected edge between them, and a node corresponding to the reference entity in the cell knowledge graph and a node corresponding to the second prediction entity in the cell knowledge graph have a directly connected edge between them, and the relationship between the reference entity and the first prediction entity, as well as the relationship between the reference entity and the second prediction entity, is determined.

[0012] According to some embodiments of the present disclosure, the entity type corresponding to the first prediction entity is a cell, and the entity type corresponding to the second prediction entity is a gene ontology term, and the method further includes: in response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a first preset threshold, using the gene ontology term corresponding to the second prediction entity as a functional annotation of the cell corresponding to the first prediction entity.

[0013] According to some embodiments of the present disclosure, the entity type corresponding to the first prediction entity is gene, and the entity type corresponding to the second prediction entity is disease, and the method further includes: in response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a second preset threshold, determining that the disease corresponding to the second prediction entity is associated with the gene corresponding to the first prediction entity.

[0014] According to some embodiments of the present disclosure, the entity type corresponding to the first prediction entity is a cell, and the entity type corresponding to the second prediction entity is a drug, and the method further includes: in response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a third preset threshold, determining that the drug corresponding to the second prediction entity has an impact on the cell corresponding to the first prediction entity.

[0015] According to another aspect of the present disclosure, a cell clustering method is provided, comprising: determining a feature vector corresponding to a node indicating a cell in the cell knowledge graph based on multiple entities and multiple relationships in the cell knowledge graph; clustering the cells in the cell knowledge graph according to the feature vector to obtain a clustering result of the cells corresponding to the feature vector.

[0016] According to another aspect of the present disclosure, there is provided an apparatus for constructing a cell knowledge graph, comprising: a type determination module, configured to determine a plurality of entity types and a plurality of relationship types in a cell-related field, each relationship type representing a relationship between two entities having any entity type among the plurality of entity types, wherein the plurality of entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe the functions of genes and their products; a first data acquisition module, configured to acquire a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities from a first data set, the entity type of each first entity being one of the plurality of entity types, and the relationship type of each first relationship being one of the plurality of relationship types, wherein the first data set includes entities having the plurality of entity types. A plurality of entities, and a plurality of relationships having the plurality of relationship types; a graph construction module, configured to construct a cell knowledge graph corresponding to the cell-related field with the plurality of first entities as nodes and the plurality of first relationships as edges; a second data acquisition module, configured to acquire a plurality of second entities and second relationships corresponding to the plurality of second entities from a second data set, the entity type of each second entity being one of the plurality of entity types, and the relationship type of the second relationship being one of the plurality of relationship types, wherein the second data set comprises literature data in the cell-related field, and the literature data comprises data related to the plurality of second entities and the second relationships; a graph update module, configured to update the cell knowledge graph according to the plurality of second entities and the second relationships.

[0017] According to another aspect of the present disclosure, a computing device is provided, comprising: a memory configured to store computer-executable instructions; and a processor configured to execute any one of the methods provided according to the aforementioned aspects of the present disclosure when the computer-executable instructions are executed by the processor.

[0018] According to another aspect of the present disclosure, a computer-readable storage medium is provided, which stores computer-executable instructions. When the computer-executable instructions are executed, any method provided according to the aforementioned aspects of the present disclosure is executed.

[0019] According to yet another aspect of the present disclosure, a computer program product is provided, comprising computer executable instructions. When the computer executable instructions are executed by a processor, any method provided according to the aforementioned aspects of the present disclosure is executed.

[0020] According to the method for constructing a cell knowledge graph provided by the present disclosure, by determining multiple entity types and multiple relationship types in a cell-related field (wherein the multiple entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe the functions of genes and their products), and by obtaining multiple first entities and multiple first relationships corresponding to the multiple first entities from a first data set, a cell knowledge graph corresponding to the cell-related field is constructed. The multiple entities and the corresponding multiple relationships in the knowledge graph constructed in this way can contain information related to cells and having multiple biological levels (for example, information related to cell types, marker genes, cell functions, etc.), which helps to efficiently process and analyze the entity and relationship information in the cell knowledge graph. In addition, by obtaining multiple second entities and second relationships corresponding to the multiple second entities from a second data set (wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the multiple second entities and the second relationship), the knowledge graph can include the latest and authoritative research results in the cell-related field, thereby improving the accuracy of the information in the knowledge graph.

[0021] These and other aspects of the disclosure will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In the following description of exemplary embodiments in conjunction with the accompanying drawings, more details, features and advantages of the technical solutions of the present disclosure are disclosed, in which:

[0023] Figure 1 Schematically illustrates example scenarios in which the technical solutions provided according to some embodiments of the present disclosure can be applied;

[0024] Figure 2A An example flowchart of a method for constructing a cell knowledge graph according to some embodiments of the present disclosure is schematically shown;

[0025] Figure 2B Schematically shows Figure 2A An exemplary schematic diagram of the method for constructing a cell knowledge graph in ;

[0026] Figure 3A An example flow chart schematically shows a method for training a graph element extraction model;

[0027] Figure 3B Schematic diagram showing the use of Figure 3A The graph element extraction model in the above example is used to obtain an exemplary schematic diagram of entities and relationships;

[0028] Figure 4Schematic diagram showing the use of Figure 3A An exemplary schematic diagram of the graph element extraction model in the example to handle the situation where multiple entity abbreviations are connected together;

[0029] Figure 5 Schematically showing an exemplary principle diagram of a method for constructing a cell knowledge graph according to some other embodiments of the present disclosure;

[0030] Figure 6 Schematically showing the visualization query results of the cell knowledge graph constructed according to some embodiments of the present disclosure;

[0031] Figure 7 An exemplary principle diagram schematically illustrates a method for predicting relationships between entities according to some embodiments of the present disclosure;

[0032] Figure 8 An exemplary principle diagram schematically illustrates a method for predicting relationships between entities according to other embodiments of the present disclosure;

[0033] Fig. 9 Schematically illustrates an exemplary principle diagram of a cell clustering method according to some embodiments of the present disclosure;

[0034] Fig.10 Schematically shows an example block diagram of an apparatus for constructing a cell knowledge graph according to some embodiments of the present disclosure;

[0035] Fig.11 An example system is illustrated that includes an example computing device that is representative of one or more systems and / or devices that may implement the various techniques described herein. DETAILED DESCRIPTION

[0036] Several embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings so that those skilled in the art can implement the technical solutions of the present disclosure. The technical solutions of the present disclosure can be embodied in many different forms and purposes, and should not be limited to the embodiments described herein. These embodiments are provided to make the technical solutions of the present disclosure clear and complete, but the described embodiments do not limit the protection scope of the present disclosure.

[0037] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present disclosure belongs. It will be further understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the relevant art and / or the context of this specification, and will not be interpreted in an idealized or overly formal sense unless explicitly defined as such herein.

[0038] Before describing the embodiments of the present disclosure in detail, some related concepts are first explained.

[0039] 1. Knowledge Graph (KG): It is a modern theory that combines the theories and methods of applied mathematics, graphics, information visualization technology, information science and other disciplines with methods such as metrology citation analysis and co-occurrence analysis, and uses visual graphs to vividly display the core structure, development history, frontier fields and overall knowledge architecture of the discipline to achieve the purpose of multidisciplinary integration. The knowledge graph stores entities in the real world and the relationship between entities in the form of structured triples, expressed as KG = (ε, r, s), where ε represents the entity set, r represents the relationship set, and s represents the set of triples in the knowledge graph. Triples usually describe facts in a specific field, and are composed of a head entity, a tail entity, and a description of the relationship between the two entities.

[0040] 2. Knowledge graph representation: It represents the entities and relations in the knowledge graph as low-dimensional continuous entity vectors, namely embedding, which can be applied to downstream tasks such as knowledge graph completion and relationship prediction, and therefore has received widespread attention.

[0041] 3. Gene Ontology (GO) provides terms (GO terms) for describing gene functions. Gene Ontology terms describe the functions of genes and their products from three aspects: molecular function, cellular component, and biological process. Specifically, molecular function defines the process or activity at the molecular level, cellular component defines the specific area where the process occurs, and biological process defines the broader biological process of the process. For example, a gene encodes a protein (the protein is the product of the gene), which performs a function (i.e., molecular function) at the molecular level in a specific area of ​​the cell (i.e., cellular component), and this function is part of a larger biological process (i.e., biological process).

[0042] 4. Pathway: It is a series of interactions between cellular molecules that lead to the production of a specific product or changes in certain cells. Pathways can trigger the combination of new molecules, such as fats or proteins. Pathways can also turn genes on and off, or stimulate cell movement. Common pathways include metabolism (i.e. metabolic pathways), regulation of gene expression (i.e. genetic pathways), and signal transmission (i.e. signal transduction pathways).

[0043] 5. Phenotype: It is the observable structural and functional characteristics of an organism, such as morphological and behavioral characteristics. Phenotype is the product of the interaction between genotype and environment, that is, the manifestation of a specific genotype under certain environmental conditions. The Human Phenotype Ontology (HPO) provides a standardized vocabulary for phenotypic abnormalities encountered in human diseases. Each term in the HPO describes a phenotypic abnormality, such as atrial septal defect.

[0044] Figure 1 An example scenario 100 is schematically shown in which the technical solutions provided according to some embodiments of the present disclosure can be applied. Figure 1 As shown, the scene 100 may include a user 110, a terminal device 120 (e.g., a computer), a terminal device 130 (e.g., a tablet), a network 140, and a remote facility 150. As an example, the remote facility 150 includes a server 151 and optionally a database device 152 for storing information to be presented, and these servers or devices may communicate via the network 140.

[0045] An application (not shown) may be installed on the terminal device 120 or 130, and the application may provide a visual user interface for the user 110, and the user 110 may interact with the application on the user interface to obtain information of interest. Exemplarily, the user 110 triggers a construction instruction for a cell knowledge graph through the user interface provided by the application installed on the terminal device 120, and the construction instruction includes information related to multiple entity types and multiple relationship types in the cell-related field, and each relationship type represents a relationship between two entities having any entity type among the multiple entity types, wherein the multiple entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe the functions of genes and their products; the terminal device 120 sends the construction instruction triggered by the user 110 to the server 151 via the network 140; after receiving the construction instruction, the server 151 obtains multiple first entities and multiple first relationships corresponding to the multiple first entities from the first data set, wherein the entity type of each first entity is one of the multiple entity types, and the relationship type of each first relationship is one of the multiple relationship types. A method for constructing a cell knowledge graph corresponding to the cell-related field, wherein the first data set includes a plurality of preset entities and a plurality of relationships corresponding to the plurality of entities; after obtaining the plurality of first entities and the plurality of first relationships, the server 151 uses the plurality of first entities as nodes and the plurality of first relationships as edges to construct a cell knowledge graph corresponding to the cell-related field; then, the server 151 obtains a plurality of second entities and a second relationship corresponding to the plurality of second entities from a second data set, wherein the entity type of each second entity is one of the plurality of entity types, and the relationship type of the second relationship is one of the plurality of relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the plurality of second entities and the second relationship; the server 151 then updates the cell knowledge graph according to the plurality of second entities and the second relationship. The server 151 stores the cell knowledge graph that is finally constructed (for example, in a database device 152). After the cell knowledge graph is constructed, the server 151 can send the relevant data of the cell knowledge graph to the client 120 via the network 140 for visual presentation.

[0046] It should be noted that the first data set can be any data set having relevant data corresponding to the multiple entity types and the multiple relationship types. Exemplarily, the first data set can be a set of the following biomedical databases: CellOntology (which includes various types of cell entities and the relationships between them), CellMarker (which includes the relationships between various types of cell entities and gene entities), HGNC (which includes the relationships between various types of gene entities and gene ontology term entities, GeneOntology (which includes various gene ontology term entities and the relationships between them); the second data set can be a collection of research literature (for example, papers, patents, etc.) in the cell-related field.

[0047] It should be noted that the term "user" used in this article can refer to any party that needs to review, analyze or access information, including but not limited to human users, network platforms, and even machines.

[0048] In the present disclosure, the server 151 may be a single server or a server cluster, on which an application for a method for constructing a cell knowledge graph may be run and related data may be stored. Alternatively, the server 151 may also run other applications and store other data. For example, the server 151 may include multiple virtual hosts to run different applications and provide different services.

[0049] In the present disclosure, the terminal devices 120 and 130 may be various types of devices, such as mobile phones, tablet computers, laptop computers, wearable devices such as smart watches, vehicle-mounted devices, etc. A client may be deployed on the terminal devices 120 and 130, which may be used to present the cell knowledge graph and optionally provide other services, and may take any of the following forms: a locally installed application, a small program accessed via other applications, a web program accessed via a browser, etc. The user 110 may view the information presented by the client and perform corresponding interactive operations through the input / output interface of the terminal devices 120 and 130. Optionally, the terminal devices 120 and 130 may be integrated with the server 151.

[0050] In the present disclosure, the database device 152 can be regarded as an electronic file cabinet, that is, a place to store electronic files, and users can add, query, update, delete, etc. data in the files. The so-called "database" is a collection of data that is stored together in a certain way, can be shared with multiple objects, has as little redundancy as possible, and is independent of the application program.

[0051] In addition, in the present disclosure, the network 140 may be a wired network connected via a cable, optical fiber, etc., or a wireless network such as 2G, 3G, 4G, 5G, Wi-Fi, Bluetooth, ZigBee, Li-Fi, etc.

[0052] Figure 2A The following schematically shows an example flow chart of a method 210 for constructing a cell knowledge graph (hereinafter referred to as graph construction method 210 for brevity) according to some embodiments of the present disclosure. Exemplarily, the graph construction method 210 may be: Figure 1 The server 151 is shown as being implemented, although this is not restrictive.

[0053] Specifically, in step 211, multiple entity types and multiple relationship types in the cell-related field are determined. It should be noted that in the present disclosure, the term "cell-related field" indicates multiple fields related to the structure, components and functions of cells in the biomedical field. Exemplarily, multiple entity types in the cell-related field include but are not limited to cells, genes, and gene ontology terms. Each relationship type represents the relationship between two entities of any entity type in the multiple entity types, and the relationship type includes but is not limited to a cell-cell relationship type representing the relationship between a cell entity and a cell entity, a cell-gene relationship type representing the relationship between a cell entity and a gene entity, a gene entity-gene ontology term relationship type representing the relationship between a gene entity and a gene ontology term entity, and a gene ontology term-gene ontology term relationship type representing the relationship between a gene ontology term entity and a gene ontology term entity. Exemplarily, lung ciliated cells belong to ciliated cells, and therefore, this subordinate relationship between the lung ciliated cell entity and the ciliated cell entity can be classified as the above-mentioned cell-cell relationship type.

[0054] In step 212, a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities are obtained from a first data set, the entity type of each first entity is one of the plurality of entity types, and the relationship type of each first relationship is one of the plurality of relationship types, wherein the first data set includes a plurality of preset entities and a plurality of relationships corresponding to the plurality of entities. As described above, the first data set may include a plurality of biomedical databases such as CellOntology, CellMarker, HGNC, GeneOntology, etc. Through this step, entities having one of the entity types such as cell, gene, and gene ontology term can be obtained from these databases as first entities, and relationships having the corresponding relationship types can be obtained from these databases as first relationships.

[0055] For example, one gene product of the immediate early gene (IEGs) is cytochrome C. In the GeneOntology database, functional annotations are provided for this gene product of the immediate early gene from three aspects: molecular function, cellular component, and biological process based on gene ontology, namely, molecular function: oxidoreductase activity; cellular component: mitochondrial matrix; biological process: oxidative phosphorylation. That is, through the GeneOntology database, the gene "immediate early gene" and the gene ontology terms "oxidoreductase activity", "mitochondrial matrix" and "oxidative phosphorylation" can be obtained as the first entity, and the annotation relationship between these gene ontology terms and the immediate early gene is taken as the first relationship. It should be noted that there are also various relationships between gene ontology terms. For example, the molecular function term "cyclin-dependent protein kinase activity" is part of the biological process "cell cycle". In gene ontology, this relationship is defined as "part of..." (part of, P). Therefore, through the GeneOntology database, the relationship between gene ontology terms can also be obtained as the first relationship.

[0056] In step 213, the cell knowledge graph corresponding to the cell-related field is constructed with the multiple first entities as nodes and the multiple first relationships as edges. Exemplarily, from the perspective of visually displaying the cell knowledge graph, different colors or thicknesses can be set for different edges according to the relationship type of the corresponding first relationship. For example, the edge corresponding to the first relationship with a cell-cell relationship type can be red, the edge corresponding to the first relationship with a cell-gene relationship type can be green, the edge corresponding to the first relationship with a cell-gene ontology term relationship type can be blue, and the edge corresponding to the first relationship with a gene ontology term-gene ontology term relationship type can be yellow. By setting different formats for different types of edges, the different relationships between nodes in the cell knowledge graph are made more obvious.

[0057] In step 214, a plurality of second entities and second relationships corresponding to the plurality of second entities are obtained from a second data set, the entity type of each second entity is one of the plurality of entity types, and the relationship type of the second relationship is one of the plurality of relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the plurality of second entities and the second relationship. The second data set may be a collection of research literature (e.g., papers, patents, etc.) in the cell-related field. For example, a large amount of literature data published in academic journals and conferences provides a large amount of single-cell sequencing analysis results data, which often contain the latest research results on cells, genes, and gene functions. Obtaining a plurality of second entities and second relationships corresponding to the plurality of second entities from these documents helps to make better use of existing research results, thereby supplementing the relevant entities and relationships into the constructed cell knowledge graph, so that the cell knowledge graph can be further improved.

[0058] The process of obtaining multiple second entities and second relationships corresponding to the multiple second entities from these documents can be done manually (for example, obtaining the multiple second entities and second relationships from the documents by professionals familiar with cell-related knowledge). Alternatively, natural language processing technology (for example, BERT (Bidirectional Encoder Representations from Transformers) model) can also be used to obtain the second entity and the second relationship from the documents. The present invention does not limit the specific natural language processing technology used. It should be noted that the entity type of each second entity is one of the multiple entity types, and the relationship type of the second relationship is one of the multiple relationship types. Therefore, the corresponding natural language processing technology needs to meet the above conditions when extracting entities and relationships from the documents. Taking the use of the BERT model for entity and relationship extraction as an example, when training the BERT model, the corresponding training sample can be the literature data in the cell-related field, and the label corresponding to the training sample needs to correspond to the multiple entity types and the multiple relationship types. Exemplarily, when the document is annotated with entities and relationships to determine the training samples and labels, the entities corresponding to the multiple entity types and the relationships corresponding to the multiple relationship types need to be annotated as labels.

[0059] In step 215, the cell knowledge graph is updated according to the multiple second entities and the second relationships. Specifically, the cell knowledge graph can be updated with the multiple second entities as nodes and the multiple second relationships as edges. This process can be regarded as a supplement to the cell knowledge graph constructed in step 213, that is, entities and relationships that do not exist in the cell knowledge graph constructed in step 213 are supplemented to improve the cell knowledge graph. Similar to the above description of step 213, from the perspective of visual display of the cell knowledge graph, different colors or thicknesses can be set for different edges according to the relationship type of the corresponding second relationship to make the different relationships between nodes in the cell knowledge graph more prominent.

[0060] pass Figure 2A The graph construction method 210 shown is based on a first data set in a cell-related field (for example, multiple biomedical databases) and literature data. The multiple entities and corresponding multiple relationships in the constructed cell knowledge graph can contain cell-related information with multiple biological levels (for example, information related to cell types, marker genes, cell functions, etc.), thereby facilitating efficient processing and analysis of entity and relationship information in the cell knowledge graph, and allowing the knowledge graph to include the latest and authoritative research results in the cell-related field, thereby improving the accuracy of the information in the knowledge graph.

[0061] Figure 2B Schematically shows Figure 2A An exemplary schematic diagram of the graph construction method 210 in FIG. Figure 2B As shown, reference numeral 220 indicates the basic architecture of the cell knowledge graph constructed based on the first data set and updated based on the second data set (hereinafter referred to as cell knowledge graph 220 for simplicity). In the cell knowledge graph 220, three entity types are included: cell, gene, and gene ontology term (GO term), and four relationship types are included: cell-cell, cell-gene, cell-GO term, GO term-GO term. The various entities corresponding to these entity types and the various relationships between the entities corresponding to these relationship types are obtained through the above-mentioned databases (CellOntology, CellMarker, HGNC, GeneOntology) and literature data in cell-related fields.

[0062] It should be noted that in the present disclosure, the entity type of "cell" can also be expressed as "cell type", such as T lymphocytes, B lymphocytes, etc. For example, the CellOntology database defines different cell types based on cell ontology and provides relationships between them. Cell ontology provides a standard nomenclature for defined cell types for comparative analysis and biomarker discovery. Different cell types can not only characterize the normal functions of different organs and tissues, but can also serve as cellular biomarkers for a variety of different disease states and treatment / vaccine responses.

[0063] A user (e.g., user 110 in example scenario 100) may obtain information or knowledge in a cell-related field from the cell knowledge graph 220. For example, a user may use T lymphocytes as query cells and obtain information related to T lymphocytes from the cell knowledge graph 220, including but not limited to the relationship between T lymphocytes and other types of cells, marker genes of T lymphocytes, GO terms describing the functions of marker genes of T lymphocytes, and the like.

[0064] In addition, when obtaining the first entity and the first relationship from the first data set, it involves fusing data from multiple databases containing corresponding entities and relationships; when obtaining the second entity and the second relationship from the second data set, it involves extracting corresponding entities and relationships from literature in cell-related fields. In the above process, various means of constructing knowledge graphs in related technologies can be applied, such as entity recognition, entity disambiguation, entity unification, reference resolution, relationship extraction, etc. Exemplarily, when obtaining the second entity and the second relationship from the second data set, a variety of natural language processing technologies or combinations thereof can be applied to perform entity recognition (extraction), and the multiple natural language processing technologies or combinations thereof include but are not limited to a combination of a long short-term memory network (LSTM) and a conditional random field algorithm (CRF), a combination of a convolutional neural network (CNN) and CRF, a combination of BERT and CRF, a combination of BERT and LSTM and CRF, and the like.

[0065] In some embodiments, step 214 may include: using a trained graph element extraction model to obtain the plurality of second entities and the second relationship from the second data set, wherein an example flowchart of a graph element extraction model training method (hereinafter referred to as model training method 310 for brevity) is shown in Figure 3A The following is a detailed description of each step of the model training method 310.

[0066] In step 311, a training data set is obtained, the training data set including sample literature data, and sample entity data and sample relationship data as labels of the sample literature data. Papers (text data) published in journals and conferences in cell-related fields (e.g., "CELL" journal) can be used as the sample literature data, and then professionals with knowledge in cell-related fields manually annotate the sample entity data and sample relationship data in the sample literature data as labels. In this process, the relevant professionals need to annotate the sample literature data according to the multiple entity types (including but not limited to cells, genes, GO terms) and multiple relationship types (including but not limited to cell-cell, cell-gene, cell-GO term, GO term-GO term) mentioned above.

[0067] For example, here is a data format of the training dataset (entity types include diseases):

[0068] { "text": "SAMD11 is used to treat diabetes", "spo_list": [{"predicate": "treat", "object_type": "GENE", "subject_type": "DISEASE", "object": "SAMD11", "subject": "diabetes"}]

[0069] Among them, "text" represents a sentence in the sample document data, and the graph element extraction model extracts entities and the relationships between entities based on this sentence; "spo_list" stores the sample entity data and sample relationship data as the labels corresponding to this sentence, as well as the corresponding entity types and relationship types. Among them, "predicate" represents the sample relationship data, which is "treatment" ("treat") here, "object_type" and "subject_type" represent the two entity types corresponding to the relationship, which are "gene" and "disease" ("GENE", "DIEASE") here, and "object" and "subject" represent the sample entity data corresponding to the relationship in this sentence, which are "SAMD11" and "diabetes" respectively.

[0070] In step 312, the sample document data is input into the graph element extraction model (e.g., BERT model) to obtain predicted entity data and predicted relationship data corresponding to the sample document data. It should be noted that the graph element extraction model may include a feature extraction model (whose output is a corresponding feature vector) for extracting features of the sample document data, and may also include a classifier (e.g., a fully connected neural network or a sigmoid activation function, whose input is a feature vector output by the feature extraction model) to obtain predicted entity data and predicted relationship data corresponding to the sample document data. Predicting entities and relationships in the sample document data can essentially be regarded as a multi-classification problem, so it is only necessary to obtain the feature vector output by the feature extraction model and use the classifier corresponding to entity prediction and the classifier corresponding to relationship prediction to map it into the probability value of the label, and then use the entity corresponding to the label with the largest probability value as the predicted entity data, and use the relationship corresponding to the label with the largest probability value as the predicted relationship data.

[0071] In step 313, a loss function is determined based on the sample entity data, the predicted entity data, the sample relationship data, and the predicted relationship data. The loss function is used to measure the difference between the predicted entity data and predicted relationship data output by the graph element extraction model and the sample entity data and sample relationship data corresponding to the sample document data. Through this loss function, the supervisory information provided by the sample entity data and the sample relationship data can be utilized in the process of model training. The embodiment of the present application does not limit the manner in which the loss function corresponding to the graph element extraction model is obtained. Exemplarily, the cross entropy loss function of the sample entity data, the predicted entity data, the sample relationship data, and the predicted relationship data can be obtained, and the cross entropy loss function is used as the loss function corresponding to the graph element extraction model.

[0072] In step 314, the parameters of the graph element extraction model are adjusted so that the loss function is minimized. It should be noted that the expression "minimizing the loss function" in this article may refer to the minimization of the loss function value when the number of model training times reaches a times threshold, or may refer to the minimization of the loss function value caused by the loss function value being less than the loss function threshold and triggering the termination of training, or may refer to obtaining an ideal or very close to the ideal global minimum of the loss function. In addition, alternative expressions for "minimizing the loss function" may be "making the loss function converge", or "making the loss function reach a maximum value", etc. The times threshold and the loss function threshold are set based on experience, or flexibly adjusted according to the application scenario, and the embodiments of the present application do not limit this.

[0073] Figure 3B Schematic diagram showing the use of Figure 3AThe graph element extraction model in is used to obtain an exemplary schematic diagram of entities and relationships. Figure 3B As shown in the figure, the graph element extraction model is Figure 3B Multiple entities and the relationships between entities are extracted from the literature data above. Figure 3B The entities corresponding to the entity type "cell" extracted from the literature data above include proliferating T cells, CD8 + T cells (CD8 + T cell), Th1 effector cell, regulatory T cell (Treg cell); the atlas element extraction model is from Figure 3B The entities corresponding to the entity type "gene" extracted from the literature data above include MKI67, GZMB, HLA-DRA, CD74, GBP1, GBP5, STAT1, GZMA, and GZMH.

[0074] Figure 3B The dotted area below represents a part of the cell knowledge graph constructed based on the database data in the cell-related field, which shows Treg cells, CD8 + T cell l, Th1 effector cell, GBP1, GBP5, STAT1, GZMA, GZMH and other entities and their relationships, and according to the atlas element extraction model from Figure 3B The entities and corresponding relationships extracted from the literature data above can update the cell knowledge graph, that is, make the cell knowledge graph contain more information (such as Figure 3B As shown, proliferating T cells, CD8 + Entities such as T cells, MKI67, GZMB, HLA-DRA, CD74 and the relationships between them are added to the cell knowledge graph).

[0075] In the process of entity recognition on literature data, we often encounter situations where entity abbreviations are connected together. Figure 4 Schematic diagram showing the use of Figure 3A An exemplary schematic diagram of the graph element extraction model in to handle the situation where multiple entity abbreviations are connected together. The graph element extraction model is derived from Figure 4 The entities corresponding to the entity type "cell" extracted from the literature data above include proliferating T cells, CD8 + T cells (CD8 + T cell), Th1 effector cell, regulatory T cell (Treg cell); the atlas element extraction model is from Figure 4The entities corresponding to the entity type "gene" extracted from the literature data above include MKI67, GZMB, HLA-DRA, CD74, GBP1, STAT1, GZMA, and GZMH.

[0076] like Figure 4 As shown, GBP1 / 5 in the literature actually represents two gene entities: GBP1 and GBP5. Exemplarily, in this case, the first gene entity can be retained to simplify the entity recognition process (refer to the training steps of the graph element extraction model above, and only the corresponding sample entity data needs to be set for this case, for example, the sample entity data corresponding to GBP1 / 5 in the literature data is set to GBP1). Of course, this processing method is only exemplary, and this situation can also be processed based on certain preset rules. Taking GBP1 / 5 as an example, after the separator " / " is identified, the number "5" on the right side of the separator can be replaced with the number "1" in the gene entity GBP1 on the left side of the separator to obtain GBP5 as another gene entity (this process can be implemented by a corresponding algorithm module, which can be independent of the graph element extraction model to perform entity and relationship extraction steps in series with the graph element extraction model, or can be incorporated into the graph element extraction model to perform entity and relationship extraction steps in parallel with the graph element extraction model).

[0077] In some embodiments, the literature data in the cell-related field can be obtained by the following steps: using data related to literature in the cell-related field that meets the screening criteria as the literature data, wherein the screening criteria include at least one of the following: publication date of the literature, number of citations of the literature, impact factor of the literature publishing platform, and user rating of the literature on the corresponding publishing platform. Considering that the literature data in the cell-related field (e.g., journal articles, conference papers) is very large in quantity and the quality may vary, it is necessary to obtain literature data that meets specific standards based on certain screening criteria. The publication date of a document may affect the quality of the document data. For example, compared with documents published ten years ago, documents published in the relevant field in the past three years can provide newer research results (even thanks to the advancement of experimental equipment and the development of analytical tools, they can provide more accurate analysis results); the number of citations of a document can reflect the academic influence of the document in the relevant field, and documents with more citations often have more accurate and cutting-edge research results; the impact factor of a document publishing platform (for example, a journal) refers to the frequency of citations of documents published on the platform in a specific year or period, which is an important indicator for measuring the academic influence of the platform. To a certain extent, it can be considered that documents published on platforms with higher impact factors have higher academic value; the user rating of a document on the corresponding publishing platform can reflect the evaluation of the document by researchers in the same or similar fields, and documents with higher user ratings often have greater academic influence and more authoritative research results. One or more of the screening indicators such as the publication date of the document, the number of citations of the document, the impact factor of the document publishing platform, and the user rating of the document on the corresponding publishing platform can be comprehensively considered to screen out more accurate and cutting-edge document data in the cell-related field.

[0078] Figure 5 The following schematically shows an exemplary principle diagram of a method for constructing a cell knowledge graph according to some other embodiments of the present disclosure (hereinafter referred to as graph construction method 500 for brevity). In graph construction method 500, the multiple entity types also include at least one of the following: protein, drug, pathway, disease, phenotype, and tissue, wherein the pathway is used to describe the interaction between molecules in a cell, the phenotype is used to describe the symptoms produced by a patient, and the tissue is composed of cells and intercellular matrix with similar morphology and function. Exemplarily, graph construction method 500 can be composed of Figure 1 The server 151 is shown as being implemented, although this is not restrictive.

[0079] like Figure 5As shown, reference numeral 520 indicates the basic architecture of the cell knowledge graph constructed by the graph construction method 500 based on the first data set and updated based on the second data set (hereinafter referred to as the cell knowledge graph 520 for simplicity). In the cell knowledge graph 520, nine entity types are included: cell, gene, gene ontology term (GO term), tissue, protein, drug, pathway, disease, and phenotype, and twelve relationship types are included: cell-cell, cell-gene, cell-GO term, GO term-GO term, cell-tissue (which represents the relationship between cell entity and tissue entity), cell-protein (which represents the relationship between cell entity and protein entity), gene-protein (which represents the relationship between gene entity and protein entity), drug-protein (which represents the relationship between drug entity and protein entity), pathway-gene (which represents the relationship between pathway entity and gene entity), gene-disease (which represents the relationship between gene entity and disease entity), disease-disease (which represents the relationship between disease entity and disease entity), disease-phenotype (which represents the relationship between disease entity and phenotype entity). Various entities corresponding to these entity types and various relationships between entities corresponding to these relationship types can be obtained through databases (including but not limited to CellOntology, CellMarker, HGNC, GeneOntology, DrugBank, KEGG, OMIM, DiseaseOntology, HPO, TissueNexus) and literature data in cell-related fields.

[0080] A user (e.g., user 110 in example scenario 100) can obtain information or knowledge in the cell-related field from the cell knowledge graph 520. Exemplarily, as described above, a user can use T lymphocytes as query cells to obtain information related to T lymphocytes from the cell knowledge graph 520, including but not limited to the relationship between T lymphocytes and other types of cells, marker genes of T lymphocytes, GO terms describing the functions of marker genes of T lymphocytes, etc. In addition, since the cell knowledge graph 520 contains more entity types and relationship types than the cell knowledge graph 220, the user can obtain more types of information from the cell knowledge graph 520, for example, information on the relationship between gene entities and pathway entities can be obtained.

[0081] Figure 6 The following schematically shows the visualization query results of the cell knowledge graph constructed according to some embodiments of the present disclosure. Figure 6As shown in the figure, a cell entity (here, human T cells) and its marker gene entity, as well as the relationship between them, are visualized. Specifically, the two types of nodes in the knowledge graph represent cell entities and marker gene entities, respectively, and the edges between these nodes indicate the gene-cell relationship type. Figure 6 It can be seen that there are multiple marker genes for human T cells, including LY9, IL6ST, CD48, SPN, CD160, ZAP70, CD2, CD69, CD1A, MME, MS4A1, etc.

[0082] According to another aspect of the present disclosure, a method for predicting relationships between entities (hereinafter referred to as a relationship prediction method for simplicity) is provided, the relationship prediction method comprising: obtaining two entities from a cell knowledge graph constructed by the graph construction method 210 as a first prediction entity and a second prediction entity, wherein there is no directly connected edge between a node corresponding to the first prediction entity in the cell knowledge graph and a node corresponding to the second prediction entity in the cell knowledge graph; performing relationship prediction on the first prediction entity and the second prediction entity based on the cell knowledge graph to obtain a relationship prediction result, the relationship prediction result being used to characterize the degree of association between the first prediction entity and the second prediction entity.

[0083] Combined with the following Figure 7 To better illustrate the relationship prediction method. Figure 7As shown, reference numeral 700 indicates the basic architecture of the cell knowledge graph constructed by the graph construction method 210 (hereinafter referred to as cell knowledge graph 700 for simplicity). In the cell knowledge graph 700, three entity types are included: cell, gene, and gene ontology term (GO term), and four relationship types are included: cell-cell, cell-gene, cell-GO term, GO term-GO term. The various entities corresponding to these entity types and the various relationships between the entities corresponding to these relationship types are obtained through databases (CellOntology, CellMarker, HGNC, GeneOntology) and literature data in cell-related fields. Exemplarily, the entity type of the first predicted entity is cell (as indicated by reference numeral 710), and the entity type of the second predicted entity is GO term (as indicated by reference numeral 720). There is no directly connected edge between the node corresponding to the first prediction entity in the cell knowledge graph and the node corresponding to the second prediction entity in the cell knowledge graph. In the cell knowledge graph 700, this feature is manifested as: there is no solid line connected edge between the cell 710 and the GO term 720, that is, there is no data related to the relationship between them in the first data set and the second data set used in the graph construction method 210. It should be noted that the first prediction entity and the second prediction entity can have entities of any entity type (for example, they can have a cell entity type and a gene entity type, respectively, and even they can both have a GO term entity type). Here, the first prediction entity and the second prediction entity are respectively cell 710 and GO term 720. The two entity types are used as an example to illustrate the principle of the relationship prediction method.

[0084] According to the entities and relationships contained in the cell knowledge graph 700, the first prediction entity and the second prediction entity can be predicted to obtain a relationship prediction result, and the relationship prediction result is used to characterize the degree of association between the first prediction entity and the second prediction entity. This process can be carried out by analyzing and processing the relevant entities and relationships in the cell knowledge graph 700. For example, the cell knowledge graph 700 can be used to analyze and process to determine that a marker gene of human liver cells is an immediate early gene (IEGs), and the gene product of the immediate early gene, cytochrome C, is annotated by the following GO terms: molecular function: oxidoreductase activity; cell component: mitochondrial matrix; biological process: oxidative phosphorylation, so these GO terms can also be used as functional annotations of human liver cells. Alternatively, this process can be carried out by knowledge graph representation, that is, mapping the entities and relationships in the cell knowledge graph 700 to low-dimensional vectors. Entities and relationships can be encoded into low-dimensional vectors (which may also be referred to as feature vectors) through factor decomposition models, neural network models (e.g., CNN, graph neural network (GNN), recurrent neural network (RNN), transformer model (Transformer)) and other models in related technologies, or a combination thereof, and the present disclosure does not limit this.

[0085] Through this relationship prediction method, the cell knowledge graph can be further expanded or derived based on the entities and relationships in the cell knowledge graph, thereby further enriching the cell knowledge graph or predicting or deducing the relationships between entities that are not directly related in the cell knowledge graph, so as to expand the application of the cell knowledge graph (for example, for providing functional annotations for cells).

[0086] In some embodiments, the relationship prediction of the first prediction entity and the first prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: based on multiple entities and multiple relationships in the cell knowledge graph, respectively determining the first feature vector corresponding to the first prediction entity and the second feature vector corresponding to the second prediction entity; calculating the similarity between the first feature vector and the second feature vector, and using the similarity as the relationship prediction result. Exemplarily, the first feature vector corresponding to the first prediction entity and the second feature vector corresponding to the second prediction entity can be obtained by a neural network model such as a Transformer model, and then the similarity (for example, cosine similarity) between the first feature vector and the second feature vector is used as the relationship prediction result. Determining the degree of association between entities by the similarity between feature vectors simplifies the process of entity relationship prediction, helps to enrich the cell knowledge graph in a convenient way or predict or derive the relationship between entities that are not directly connected in the cell knowledge graph, thereby expanding the application of the cell knowledge graph.

[0087] In some embodiments, the relationship prediction of the first prediction entity and the second prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: obtaining a reference entity from the cell knowledge graph, wherein the node corresponding to the reference entity in the cell knowledge graph and the node corresponding to the first prediction entity in the cell knowledge graph have a directly connected edge, and the node corresponding to the reference entity in the cell knowledge graph and the node corresponding to the second prediction entity in the cell knowledge graph have a directly connected edge, and the relationship between the reference entity and the first prediction entity, and the relationship between the reference entity and the second prediction entity, The relationship prediction result is determined. The following is combined with Figure 8 To better illustrate the above process.

[0088] like Figure 8 As shown, reference numeral 800 indicates the basic architecture of the cell knowledge graph constructed by the graph construction method 210 (hereinafter referred to as the cell knowledge graph 800 for simplicity). In the cell knowledge graph 800, three entity types are included: cell, gene, and gene ontology term (GO term), and four relationship types are included: cell-cell, cell-gene, cell-GO term, GO term-GO term. Various entities corresponding to these entity types and various relationships between entities corresponding to these relationship types are obtained through databases (CellOntology, CellMarker, HGNC, GeneOntology) and literature data in cell-related fields.

[0089] Exemplarily, the entity type of the first prediction entity is cell (as indicated by reference numeral 810), the entity type of the second prediction entity is GO term (as indicated by reference numeral 820), and the entity type of the reference entity is gene (as indicated by reference numeral 830). There is no directly connected edge between the node corresponding to the first prediction entity in the cell knowledge graph and the node corresponding to the second prediction entity in the cell knowledge graph. In the cell knowledge graph 800, this feature is manifested as: there is no solid line connected edge between cell 810 and GO term 820, that is, there is no data related to the relationship between them in the first data set and the second data set used in the graph construction method 210. There is a directly connected edge between the node corresponding to the reference entity in the cell knowledge graph and the node corresponding to the first prediction entity in the cell knowledge graph. In the cell knowledge graph 800, this feature is manifested as: there is a solid line connected edge between cell 810 and gene 830, that is, there is data related to the relationship between them in the first data set and the second data set used in the graph construction method 210. There is a directly connected edge between the node corresponding to the reference entity in the cell knowledge graph and the node corresponding to the second prediction entity in the cell knowledge graph. In the cell knowledge graph 800, this feature is manifested as: there is a solid line connected edge between the gene 830 and the GO term 820, that is, the first data set and the second data set used in the graph construction method 210 have data related to the relationship between them.

[0090] It should be noted that the first prediction entity, the second prediction entity and the reference entity may have entities of any entity type (for example, they may have a cell entity type, a GO term entity type, and a gene entity type, respectively). Here, the first prediction entity, the second prediction entity and the reference entity respectively have three entity types of cell 810, GO term 820 and gene 830 as examples to illustrate the principle of the corresponding relationship prediction method.

[0091] Specifically, in the cell knowledge graph 800, w_1 can be used to represent the weight value of the edge between the node corresponding to the reference entity and the node corresponding to the first prediction entity. For example, the expression value of the marker gene (here, the reference entity) corresponding to the cell (here, the first prediction entity) obtained from the CellMarker database can be used as w_1. w_2 can be used to represent the weight value of the edge between the node corresponding to the reference entity and the node corresponding to the second prediction entity. For example, the information content (IC) of the GO term (here, the second prediction entity) can be determined through the GeneOntology database and the information content can be used as w_2. The specific steps for calculating IC are known in the art and will not be repeated here.

[0092] Exemplarily, the product of w_1 and w_2 obtained in the above steps can be used as the relationship prediction result. When the product of w_1 and w_2 is higher than a preset threshold (for example, 0.5), it can be considered that the degree of association between the first prediction entity and the second prediction entity is high, or it can be considered that there is an association between them, which helps to expand the cell knowledge graph 800 or expand its application (for example, for providing functional annotations for cells).

[0093] It should be noted that in the present disclosure, other methods different from the above methods can be used to mine the relationship between the first prediction entity and the second prediction entity. The other methods include but are not limited to link prediction algorithms, PageRank algorithms, community search algorithms, association pattern mining algorithms, and the like.

[0094] Depending on the specific application scenario, the entity types corresponding to the first prediction entity and the second prediction entity may be various, and several examples are given below:

[0095] Example 1: The entity type corresponding to the first prediction entity is gene, and the entity type corresponding to the second prediction entity is disease. When the relationship prediction result indicates that the degree of association between the first prediction entity and the second prediction entity is greater than a preset threshold (e.g., 0.4), it can be determined that the disease corresponding to the second prediction entity is associated with the gene corresponding to the first prediction entity. This example illustrates an application of the cell knowledge graph provided by the present disclosure (e.g., the cell knowledge graph 520 described above), that is, the cell knowledge graph provided by the present disclosure can be used to analyze the association between genes and diseases, thereby facilitating the analysis of the causes and mechanisms of the disease, and thereby facilitating the proposal of more scientific response plans.

[0096] Example 2: The entity type corresponding to the first prediction entity is a cell, and the entity type corresponding to the second prediction entity is a drug. When the relationship prediction result indicates that the degree of association between the first prediction entity and the second prediction entity is greater than a preset threshold (e.g., 0.6), it can be determined that the drug corresponding to the second prediction entity has an effect on the cell corresponding to the first prediction entity. This example illustrates another application of the cell knowledge graph provided by the present disclosure (e.g., the cell knowledge graph 520 described above), that is, the cell knowledge graph provided by the present disclosure can be used to analyze the association between cells and drugs, thereby facilitating the analysis of the causes and mechanisms of the disease, and further facilitating the development of drugs that may have an effect on cells.

[0097] Fig. 9 An exemplary principle diagram of a cell clustering method according to some embodiments of the present disclosure is schematically shown. In the cell clustering method, based on multiple entities and multiple relationships in the above-mentioned cell knowledge graph (for example, the cell knowledge graph constructed by the graph construction method 210), a feature vector corresponding to a node indicating a cell in the cell knowledge graph can be determined (for example, through a BERT model); and then the cells (cell entities) in the cell knowledge graph are clustered according to the feature vector to obtain a clustering result of the cells corresponding to the feature vector. Exemplarily, a K-mean clustering method can be used for clustering.

[0098] The following describes a method of using subspace vectors to represent feature vectors, taking the example of a cell knowledge graph with N cell entities, each cell entity corresponding to a 128-dimensional feature vector, artificially dividing the feature vectors into 4 subspaces (the number of subspaces is not restrictive), and clustering the subspace vectors into 256 categories.

[0099] Since the dimension of each feature vector is 128, and it is divided into, for example, 4 subspaces, it can be seen that the dimension of each subspace is 32. Then cluster the subspace vectors of each subspace. The subspace vectors can be clustered using the K-Means clustering method. For example, the subspace vectors can be clustered into 256 categories. In this way, each subspace vector will get a class code between 0 and 255. Since the feature vector is divided into 4 subspaces, the original 128-dimensional feature vector can be represented by a vector of a 4-dimensional class code.

[0100] After the above operations, the high-dimensional feature vector of the cell entity will be represented by a very short subspace class code vector. When analyzing a new cell entity, its category can be determined according to its corresponding subspace class code vector, and then the new cell entity can be analyzed according to the various characteristics of other cell entities corresponding to the category.

[0101] It should be noted that, in the present disclosure, other clustering algorithms different from the above-mentioned clustering algorithms can be used to cluster cell entities in the cell knowledge graph, and the other clustering algorithms include but are not limited to the Leiden clustering algorithm, the Louvain clustering algorithm, and the like. Clustering cells by using the cell knowledge graph constructed based on the method provided by the present disclosure helps to infer the properties of other cells belonging to the same category as the cell based on the information associated with a certain cell. Exemplarily, assuming that in the cell knowledge graph constructed based on the method provided by the present disclosure, a certain cell has a relationship with a certain disease, then other cells belonging to the same category as the cell may have a certain association with the disease, and the other cells (and other related information, for example, the relationship between the other cells and a certain drug) may be helpful for research on the formation mechanism of the disease or even drug development.

[0102] Fig.10 The following schematically shows an example block diagram of an apparatus for constructing a cell knowledge graph according to some embodiments of the present disclosure (hereinafter referred to as the cell knowledge graph construction apparatus 1000 for simplicity). For example, the cell knowledge graph construction apparatus 1000 can be deployed in Figure 1 On server 151 in. Fig.10 As shown, the cell knowledge graph construction device 1000 includes a type determination module 1010, a first data acquisition module 1020, a graph construction module 1030, a second data acquisition module 1040, and a graph update module 1050.

[0103] Specifically, the type determination module 1010 can be configured to determine multiple entity types and multiple relationship types in the cell-related field, each relationship type representing the relationship between two entities having any entity type among the multiple entity types, wherein the multiple entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe the functions of genes and their products; the first data acquisition module 1020 can be configured to acquire multiple first entities and multiple first relationships corresponding to the multiple first entities from a first data set, the entity type of each first entity is one of the multiple entity types, and the relationship type of each first relationship is one of the multiple relationship types, wherein the first data set includes multiple entities having the multiple entity types, and includes multiple entities having the multiple relationships multiple relationships of the type; the graph construction module 1030 can be configured to construct a cell knowledge graph corresponding to the cell-related field with the multiple first entities as nodes and the multiple first relationships as edges; the second data acquisition module 1040 can be configured to obtain multiple second entities and second relationships corresponding to the multiple second entities from a second data set, the entity type of each second entity is one of the multiple entity types, and the relationship type of the second relationship is one of the multiple relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the multiple second entities and the second relationships; the graph update module 1050 can be configured to update the cell knowledge graph according to the multiple second entities and the second relationships.

[0104] It should be understood that the cell knowledge graph construction device 1000 can be implemented in software, hardware, or a combination of software and hardware. Multiple different modules can be implemented in the same software or hardware structure, or one module can be implemented by multiple different software or hardware structures.

[0105] In addition, the cell knowledge graph construction device 1000 can be used to implement the graph construction method 210 described above, and the relevant details have been described in detail above, and for the sake of brevity, they will not be repeated here. The cell knowledge graph construction device 1000 can have the same features and advantages as described about the aforementioned graph construction method 210.

[0106] According to another aspect of the present disclosure, a device for predicting the relationship between entities (hereinafter referred to as the entity relationship prediction device for simplicity) is provided, and the entity relationship prediction device includes an entity acquisition module and an entity relationship prediction module. The entity acquisition module can be configured to acquire two entities from the cell knowledge graph constructed by the cell knowledge graph construction device 1000 as a first predicted entity and a second predicted entity, wherein there is no directly connected edge between the node corresponding to the first predicted entity in the cell knowledge graph and the node corresponding to the second predicted entity in the cell knowledge graph; the entity relationship prediction module can be configured to perform relationship prediction on the first predicted entity and the second predicted entity based on the cell knowledge graph to obtain a relationship prediction result, and the relationship prediction result is used to characterize the degree of association between the first predicted entity and the second predicted entity.

[0107] According to another aspect of the present disclosure, a device for clustering cells is provided (hereinafter referred to as a cell clustering device for simplicity). The cell clustering device includes a cell feature determination module and a cell clustering module. The cell feature determination module can be configured to determine a feature vector corresponding to a node indicating a cell in the cell knowledge graph based on multiple entities and multiple relationships in the cell knowledge graph constructed by the cell knowledge graph construction device 1000; the cell clustering module can be configured to cluster the cells in the cell knowledge graph according to the feature vector to obtain a clustering result of the cell corresponding to the feature vector.

[0108] Fig.11 An example system is illustrated, which includes an example computing device 1100 representing one or more systems and / or devices that can implement the various techniques described herein. The computing device 1100 can be, for example, a server of a service provider, a device associated with a server, a system on a chip, and / or any other suitable computing device or computing system. Any one or more of the above-mentioned cell knowledge graph construction device 1000, entity relationship prediction device, and cell clustering device can take the form of computing device 1100. Alternatively, any one or more of the cell knowledge graph construction device 1000, entity relationship prediction device, and cell clustering device can be implemented as a computer program in the form of an application 1116.

[0109] like Fig.11The illustrated example computing device 1100 includes a processing system 1111, one or more computer-readable media 1112, and one or more I / O interfaces 1113 that are communicatively coupled to each other. Although not shown, the computing device 1100 may also include a system bus or other data and command transmission system that couples the various components to each other. The system bus may include any one or combination of different bus structures, such as a memory bus or memory controller, a peripheral bus, a universal serial bus, and / or a processor or local bus utilizing any of a variety of bus architectures. Various other examples are also contemplated, such as control and data lines.

[0110] Processing system 1111 represents the functionality of performing one or more operations using hardware. Thus, processing system 1111 is illustrated as including hardware elements 1114 that may be configured as processors, functional blocks, and the like. This may include other logic devices implemented in hardware as application specific integrated circuits or formed using one or more semiconductors. Hardware elements 1114 are not limited by the materials from which they are formed or the processing mechanisms employed therein. For example, a processor may be composed of (multiple) semiconductors and / or transistors (e.g., electronic integrated circuits (ICs)). In such a context, processor executable instructions may be electronic executable instructions.

[0111] Computer-readable media 1112 is illustrated as including memory / storage 1115. Memory / storage 1115 represents memory / storage capacity associated with one or more computer-readable media. Memory / storage 1115 may include volatile media (such as random access memory (RAM)) and / or non-volatile media (such as read-only memory (ROM), flash memory, optical disks, magnetic disks, etc.). Memory / storage 1115 may include fixed media (e.g., RAM, ROM, fixed hard drives, etc.) and removable media (e.g., flash memory, removable hard drives, optical disks, etc.). Computer-readable media 1112 may be configured in various other ways as further described below.

[0112] One or more I / O interfaces 1113 represent functionality that allows a user to input commands and information to the computing device 1100 using various input devices and optionally also allows information to be presented to the user and / or other components or devices using various output devices. Examples of input devices include a keyboard, a cursor control device (e.g., a mouse), a microphone (e.g., for voice input), a scanner, a touch function (e.g., a capacitive or other sensor configured to detect physical touch), a camera (e.g., a gesture that can detect motion that does not involve touch using visible or invisible wavelengths (such as infrared frequencies), etc.). Examples of output devices include a display device (e.g., a monitor or projector), a speaker, a printer, a network card, a tactile response device, etc. Therefore, the computing device 1100 can be configured in various ways as further described below to support user interaction.

[0113] The computing device 1100 also includes an application 1116. The application 1116 may be, for example, a software instance of any one or more of the cell knowledge graph construction apparatus 1000, the entity relationship prediction apparatus, and the cell clustering apparatus, and implements the technology described herein in combination with other elements in the computing device 1100.

[0114] Various techniques may be described herein in the general context of software hardware elements or program modules. Generally, these modules include routines, programs, elements, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The terms "module", "function" and "component" as used herein generally represent software, firmware, hardware or a combination thereof. The features of the techniques described herein are platform-independent, meaning that these techniques can be implemented on a variety of computing platforms with a variety of processors.

[0115] An implementation of the described modules and techniques may be stored on or transmitted across some form of computer-readable media. Computer-readable media may include various media accessible by computing device 1100. By way of example and not limitation, computer-readable media may include “computer-readable storage media” and “computer-readable signal media”.

[0116] As opposed to a simple signal transmission, carrier wave, or signal itself, "computer-readable storage media" refers to media and / or devices that are capable of persistently storing information, and / or tangible storage devices. Therefore, computer-readable storage media refers to non-signal bearing media. Computer-readable storage media include hardware such as volatile and non-volatile, removable and non-removable media and / or storage devices implemented in a method or technology suitable for storing information (such as computer-readable instructions, data structures, program modules, logic elements / circuits, or other data). Examples of computer-readable storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical storage devices, hard disks, cassettes, magnetic tape, magnetic disk storage devices or other magnetic storage devices, or other storage devices, tangible media, or articles of manufacture suitable for storing the desired information and accessible by a computer.

[0117] "Computer-readable signal media" refers to signal-bearing media that is configured to send instructions to the hardware of the computing device 1100, such as via a network. Signal media typically can embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, data signal, or other transport mechanism. Signal media also include any information delivery media. The term "modulated data signal" refers to a signal that has one or more of its characteristics set or changed so as to encode information into the signal. By way of example and not limitation, communication media include wired media such as a wired network or direct connection and wireless media such as acoustic, RF, infrared, and other wireless media.

[0118] As previously described, hardware elements 1114 and computer-readable media 1112 represent instructions, modules, programmable device logic, and / or fixed device logic implemented in hardware form, which in some embodiments can be used to implement at least some aspects of the technology described herein. Hardware elements can include integrated circuits or systems on chips, application-specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), complex programmable logic devices (CPLDs), and other implementations in silicon or components of other hardware devices. In this context, hardware elements can be used as processing devices that perform program tasks defined by instructions, modules, and / or logic embodied by hardware elements, as well as hardware devices for storing instructions for execution, such as the computer-readable storage media described previously.

[0119] The foregoing combinations may also be used to implement the various techniques and modules described herein. Thus, software, hardware or program modules and other program modules may be implemented as one or more instructions and / or logic embodied on some form of computer-readable storage medium and / or by one or more hardware elements 1114. The computing device 1100 may be configured to implement specific instructions and / or functions corresponding to the software and / or hardware modules. Thus, for example, by using a computer-readable storage medium and / or hardware element 1114 of a processing system, the module may be implemented as a module executable by the computing device 1100 as software, at least in part, in hardware. Instructions and / or functions may be executable / operable by one or more articles (e.g., one or more computing devices 1100 and / or processing systems 1111) to implement the techniques, modules, and examples described herein.

[0120] In various embodiments, computing device 1100 can be implemented in various configurations. For example, computing device 1100 can be implemented as a computer-type device including a personal computer, a desktop computer, a multi-screen computer, a laptop computer, a netbook, etc. Computing device 1100 can also be implemented as a mobile device-type device including mobile devices such as mobile phones, portable music players, portable game devices, tablet computers, multi-screen computers, etc. Computing device 1100 can also be implemented as a television-type device, which includes a device with or connected to a generally larger screen in a casual viewing environment. These devices include televisions, set-top boxes, game consoles, etc.

[0121] The techniques described herein can be supported by these various configurations of computing device 1100 and are not limited to the specific examples of the techniques described herein. Functionality can also be implemented in whole or in part on the "cloud" 1120 by using a distributed system, such as through platform 1122 as described below.

[0122] Cloud 1120 includes and / or represents a platform 1122 for resources 1124. Platform 1122 abstracts the underlying functionality of the hardware (e.g., servers) and software resources of cloud 1120. Resources 1124 may include applications and / or data that may be used when performing computer processing on a server remote from computing device 1100. Resources 1124 may also include services provided over the Internet and / or over a subscriber network such as a cellular or Wi-Fi network.

[0123] The platform 1122 may abstract resources and functionality to connect the computing device 1100 with other computing devices. The platform 1122 may also be used to abstract hierarchies of resources to provide hierarchies of corresponding levels of demand encountered for resources 1124 implemented via the platform 1122. Thus, in an interconnected device embodiment, the implementation of the functionality described herein may be distributed throughout the system 1600. For example, functionality may be implemented in part on the computing device 1100 and through the platform 1122 that abstracts the functionality of the cloud 1120.

[0124] It should be understood that, for the sake of clarity, embodiments of the present disclosure are described with reference to different functional units. However, it will be apparent that, without departing from the present disclosure, the functionality of each functional unit can be implemented in a single unit, in multiple units, or as a part of other functional units. For example, the functionality described as being performed by a single unit can be performed by multiple different units. Therefore, reference to a specific functional unit is only considered as a reference to a suitable unit for providing the described functionality, rather than indicating a strict logical or physical structure or organization. Therefore, the present disclosure can be implemented in a single unit, or can be physically and functionally distributed between different units and circuits.

[0125] It will be understood that although the first, second, third, etc. terms may be used to describe various devices, elements, parts or portions in this article, these devices, elements, parts or portions should not be limited by these terms. These terms are only used to distinguish one device, element, part or portion from another device, element, part or portion.

[0126] Although the present disclosure has been described in conjunction with some embodiments, it is not intended to be limited to the specific forms set forth herein. On the contrary, the scope of the present disclosure is limited only by the appended claims. Additionally, although individual features may be included in different claims, these may possibly be advantageously combined, and inclusion in different claims does not imply that the combination of features is not feasible and / or advantageous. The order of features in the claims does not imply any specific order in which the features must work. Furthermore, in the claims, the word "comprising" does not exclude other elements, and the term "one" or "an" does not exclude a plurality. The figure marks in the claims are provided only as clear examples and should not be interpreted as limiting the scope of the claims in any way.

[0127] It should be understood that, for the sake of clarity, embodiments of the present disclosure are described with reference to different functional units. However, it will be apparent that, without departing from the present disclosure, the functionality of each functional unit can be implemented in a single unit, in multiple units, or as a part of other functional units. For example, the functionality described as being performed by a single unit can be performed by multiple different units. Therefore, reference to a specific functional unit is only considered as a reference to a suitable unit for providing the described functionality, rather than indicating a strict logical or physical structure or organization. Therefore, the present disclosure can be implemented in a single unit, or can be physically and functionally distributed between different units and circuits.

[0128] The present disclosure provides a computer-readable storage medium having computer-readable instructions stored thereon, which, when executed, implement the above-mentioned method for constructing a cell knowledge graph, the method for predicting the relationship between entities, and the cell clustering method.

[0129] The present disclosure provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computing device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computing device executes the method for constructing a cell knowledge graph, the method for predicting the relationship between entities, and the cell clustering method provided in the above-mentioned various optional implementations.

[0130] Variations to the disclosed embodiments will be understood and effected by those skilled in the art in practicing the claimed subject matter by studying the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and "a" or "an" does not exclude a plurality. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. A method for constructing a cell knowledge graph, comprising: Determine a plurality of entity types and a plurality of relationship types in a cell-related field, each relationship type representing a relationship between two entities of any entity type in the plurality of entity types, wherein the plurality of entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe functions of genes and their products; Acquire a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities from a first data set, wherein the entity type of each first entity is one of the plurality of entity types, and the relationship type of each first relationship is one of the plurality of relationship types, wherein the first data set includes a plurality of preset entities and a plurality of relationships corresponding to the plurality of entities; Using the multiple first entities as nodes and the multiple first relationships as edges, constructing a cell knowledge graph corresponding to the cell-related field; Acquire a plurality of second entities and second relationships corresponding to the plurality of second entities from a second data set, wherein the entity type of each second entity is one of the plurality of entity types, and the relationship type of the second relationship is one of the plurality of relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the plurality of second entities and the second relationship; and updating the cell knowledge graph according to the plurality of second entities and the second relationship; The step of obtaining a plurality of second entities and second relationships corresponding to the plurality of second entities from the second data set comprises: obtaining the plurality of second entities and the second relationships from the second data set using a trained graph element extraction model, wherein the graph element extraction model is trained by the following steps: Acquire a training data set, the training data set comprising sample document data, and sample entity data and sample relationship data as labels of the sample document data; Inputting the sample document data into the graph element extraction model to obtain predicted entity data and predicted relationship data corresponding to the sample document data; Determining a loss function based on the sample entity data, the predicted entity data, the sample relationship data, and the predicted relationship data; The parameters of the graph element extraction model are adjusted to minimize the loss function.

2. The method according to claim 1, wherein: The literature data in the cell-related field is obtained by the following steps: The data related to the literature that meets the screening criteria in the cell-related field is used as the literature data, wherein the screening criteria include at least one of the following: The publication date of the document, the number of citations of the document, the impact factor of the document publishing platform, and the user rating of the document on the corresponding publishing platform.

3. The method according to claim 1, wherein: The plurality of entity types further comprises at least one of the following: Proteins, drugs, pathways, diseases, phenotypes, and tissues, wherein the pathway is used to describe the interaction between molecules within cells, the phenotype is used to describe the symptoms produced by patients, and the tissue is composed of cells and intercellular matrix with similar morphology and function.

4. A method for predicting relationships between entities, comprising: Obtaining two entities from the cell knowledge graph constructed by the method according to claim 1 as a first prediction entity and a second prediction entity, wherein there is no directly connected edge between a node corresponding to the first prediction entity in the cell knowledge graph and a node corresponding to the second prediction entity in the cell knowledge graph; Based on the cell knowledge graph, relationship prediction is performed on the first prediction entity and the second prediction entity to obtain a relationship prediction result, and the relationship prediction result is used to characterize the degree of association between the first prediction entity and the second prediction entity.

5. The method according to claim 4, wherein: The performing relationship prediction on the first prediction entity and the first prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: Based on a plurality of entities and a plurality of relationships in the cell knowledge graph, respectively determining a first feature vector corresponding to the first prediction entity and a second feature vector corresponding to the second prediction entity; The similarity between the first feature vector and the second feature vector is calculated, and the similarity is used as the relationship prediction result.

6. The method according to claim 4, wherein: The performing relationship prediction on the first prediction entity and the first prediction entity based on the cell knowledge graph to obtain a relationship prediction result includes: Acquire a reference entity from the cell knowledge graph, wherein a node corresponding to the reference entity in the cell knowledge graph and a node corresponding to the first prediction entity in the cell knowledge graph have a directly connected edge between them, and a node corresponding to the reference entity in the cell knowledge graph and a node corresponding to the second prediction entity in the cell knowledge graph have a directly connected edge between them, The relationship prediction result is determined according to the relationship between the reference entity and the first prediction entity, and the relationship between the reference entity and the second prediction entity.

7. The method according to claim 4, wherein: The entity type corresponding to the first predicted entity is a cell, the entity type corresponding to the second predicted entity is a gene ontology term, and the method further comprises: In response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a first preset threshold, a gene ontology term corresponding to the second prediction entity is used as a functional annotation of the cell corresponding to the first prediction entity.

8. The method according to claim 4, wherein: The entity type corresponding to the first prediction entity is gene, and the entity type corresponding to the second prediction entity is disease, and the method further includes: In response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a second preset threshold, it is determined that the disease corresponding to the second prediction entity is associated with the gene corresponding to the first prediction entity.

9. The method according to claim 4, wherein: The entity type corresponding to the first predicted entity is a cell, and the entity type corresponding to the second predicted entity is a drug, and the method further includes: In response to the relationship prediction result indicating that the degree of association between the first prediction entity and the second prediction entity is greater than a third preset threshold, it is determined that the drug corresponding to the second prediction entity has an effect on the cell corresponding to the first prediction entity.

10. A cell clustering method, comprising: Based on a plurality of entities and a plurality of relationships in the cell knowledge graph constructed according to the method of claim 1, determining a feature vector corresponding to a node indicating a cell in the cell knowledge graph; The cells in the cell knowledge graph are clustered according to the feature vector to obtain a clustering result of the cells corresponding to the feature vector.

11. A device for constructing a cell knowledge graph, comprising: A type determination module is configured to determine a plurality of entity types and a plurality of relationship types in a cell-related field, each relationship type representing a relationship between two entities having any entity type in the plurality of entity types, wherein the plurality of entity types include cells, genes, and gene ontology terms, and the gene ontology terms are used to describe functions of genes and their products; A first data acquisition module is configured to acquire a plurality of first entities and a plurality of first relationships corresponding to the plurality of first entities from a first data set, wherein the entity type of each first entity is one of the plurality of entity types, and the relationship type of each first relationship is one of the plurality of relationship types, wherein the first data set includes a plurality of entities having the plurality of entity types, and includes a plurality of relationships having the plurality of relationship types; A graph construction module is configured to construct a cell knowledge graph corresponding to the cell-related field using the multiple first entities as nodes and the multiple first relationships as edges; The second data acquisition module is configured to acquire a plurality of second entities and second relationships corresponding to the plurality of second entities from a second data set, wherein the entity type of each second entity is one of the plurality of entity types, and the relationship type of the second relationship is one of the plurality of relationship types, wherein the second data set includes literature data in the cell-related field, and the literature data includes data related to the plurality of second entities and the second relationships, and wherein acquiring the plurality of second entities and the second relationships corresponding to the plurality of second entities from the second data set comprises: using a trained graph element extraction model to extract the plurality of second entities from the second data set; Acquire the multiple second entities and the second relationships, wherein the graph element extraction model is trained by the following steps: acquire a training data set, the training data set comprising sample document data, and sample entity data and sample relationship data comprising labels of the sample document data, input the sample document data into the graph element extraction model, obtain predicted entity data and predicted relationship data corresponding to the sample document data, determine a loss function based on the sample entity data, the predicted entity data, the sample relationship data, and the predicted relationship data, and adjust the parameters of the graph element extraction model to minimize the loss function; and The graph updating module is configured to update the cell knowledge graph according to the multiple second entities and the second relationship.

12. A computing device comprising: a memory configured to store computer-executable instructions; A processor configured to perform the method according to any one of claims 1 to 10 when the computer executable instructions are executed by the processor.

13. A computer-readable storage medium storing computer-executable instructions, which, when executed, perform the method according to any one of claims 1 to 10.

14. A computer program product comprising computer executable instructions, wherein the computer executable instructions, when executed by a processor, perform the method according to any one of claims 1 to 10.

Images