Association prediction method and device of omics data and disease data, electronic equipment and storage medium

Abstract

The invention provides an association prediction method and device of omics data and disease data, electronic equipment and a storage medium, and relates to the technical field of biological data processing, and the method comprises the steps: obtaining a plurality of omics data and at least one kind of disease data, constructing a first association matrix and a second association matrix, the first incidence matrix represents an incidence relation between elements in any two pieces of omics data, and the second incidence matrix represents an incidence relation between elements in each piece of omics data and elements in the disease data; a feature relation graph is constructed according to the first incidence matrix and the second incidence matrix, the feature relation graph comprises the interaction relation between the omics data, and the interaction relation between any two omics data is determined according to the correlation features of the elements in each of the two omics data and the elements in the at least one kind of disease data. The problem that an existing data analysis mode is not high in accuracy can be solved in a targeted mode.

Description

Technical Field

[0001] The present application relates to the technical field of biological data processing, and in particular to a method, device, electronic device and storage medium for predicting the association between omics data and disease data. Background Art

[0002] Omics data include genomics, transcriptomics, proteomics, metabolomics, microbiome, exposure genomics, etc. By using multi-omics to quantify and analyze biomolecular characteristics at multiple levels, it can help understand the molecular relationships and functions in biological systems, thus playing a vital role in pathological research.

[0003] In fact, the occurrence of diseases involves many omics. Among the related technologies, most of them use gene co-expression analysis (WGCNA) to calculate the correlation between multi-omics data. However, this method only studies the relationship between different omics, but does not study the association between omics data and disease data, nor can it know the relationship between omics data and disease data. Not only is there a problem of low data utilization efficiency, but it is also not conducive to understanding the pathogenesis of the disease at the molecular level, and the analysis accuracy is not high.

[0004] It should be noted that the above statements are only used to provide background technical information related to the present application and do not necessarily constitute prior art. Summary of the invention

[0005] In view of this, the purpose of the present application is to propose a method, device, electronic device and storage medium for predicting the association between omics data and disease data. The present application can specifically solve the problem of low accuracy of existing data analysis methods.

[0006] Based on the above-mentioned purpose, in a first aspect, the present application proposes a method for predicting the association between omics data and disease data, the method comprising: obtaining multiple omics data and at least one disease data, constructing a first association matrix and a second association matrix, the first association matrix representing the association relationship between each element in any two omics data, and the second association matrix representing the association relationship between each element in each omics data and each element in the disease data; constructing a characteristic relationship graph according to the first association matrix and the second association matrix, the characteristic relationship graph including the interactive relationship between each omics data, wherein the interactive relationship between any two omics data is determined according to the association characteristics of each element in each omics data in the two omics data and each element in the at least one disease data; predicting the association relationship between the multiple omics data and the at least one disease data based on the characteristic relationship graph.

[0007] In some embodiments, constructing a characteristic relationship graph based on the first association matrix and the second association matrix includes: extracting a first association feature between each element of each omics data from the first association matrix; for any omics data, extracting a second association feature between each element in the omics data and each element in at least one disease data from the second association matrix; after determining each omics data in the multiple omics data and its corresponding second association feature, determining the interactive relationship between each omics data based on the first association feature and the second association feature of each omics data, so as to construct the characteristic relationship graph based on the interactive relationship.

[0008] In some embodiments, constructing the feature relationship graph according to the interactive relationship includes: taking each omics data in the multiple omics data as a node of the feature relationship graph, and taking each omics data and its corresponding second associated feature as the initial feature of the corresponding omics node; obtaining the edge of the feature relationship graph according to the first associated features of each omics data, and constructing the feature relationship graph.

[0009] In some embodiments, after acquiring multiple omics data and at least one disease data, the method further includes: extracting omics element names and disease element names from a target database; obtaining a disease name set based on the presence of the disease element names in a preset disease vocabulary; obtaining an omics name set based on the presence of the omics element names in a preset omics vocabulary; and initializing the association relationship between each omics element of the multiple omics data and between each omics element and a disease element of at least one disease data based on each disease element name in the disease name set and each omics element name in the omics name set.

[0010] In some embodiments, based on each disease element name in the disease name set and each omics element name in the omics name set, the association relationship between each omics element of the multiple omics data and between each omics element and the disease element of at least one disease data is initialized, including: based on the presence of a disease element name in the disease name set in any of the omics data, determining that there is a second association relationship between the disease element corresponding to the disease element name and the omics element of the omics data; based on the absence of a disease element name in the disease name set in any of the omics data, determining that the second association relationship between the disease element corresponding to the disease element name and the omics data is an unknown relationship; based on the presence of an omics element name in the omics name set in at least two of the omics data, determining that the omics elements in the at least two of the omics data have a first association relationship.

[0011] In some embodiments, the method further includes: constructing a disease similarity matrix based on the similarity between each disease element in the disease name set; and updating the initial features of each node in the feature relationship graph based on the disease similarity matrix.

[0012] In some embodiments, predicting the association between the multiple omics data and the at least one disease data based on the feature relationship graph includes: updating the node features and edge depth features of the feature relationship graph to obtain an updated second association matrix; calculating the importance weights of different omics nodes in each edge of the feature relationship graph; and predicting the association between the multiple omics data and the at least one disease data based on the importance weights and the updated second association matrix.

[0013] In some embodiments, the node features and the deep features of the edges of the feature relationship graph are updated to obtain an updated second association matrix, including: aggregating the deep features of the edges of the feature relationship graph to the nodes of the feature relationship graph, and updating the node features of the feature relationship graph; calculating the importance weights of different omics nodes in each edge of the feature relationship graph, and updating the deep features of the edges according to the importance weights; decoding the updated node features and the updated deep features of the edges to obtain an updated second association matrix.

[0014] In some embodiments, the method further includes: training a prediction network according to a training sample set, wherein the prediction network is used to predict the association relationship between the multiple omics data and the at least one disease data based on the feature relationship graph, and the training sample set includes: the association relationship between multiple omics data samples and multiple disease data samples respectively, and the association relationship includes the association relationship between each element in any two omics data, and the association relationship between each element in each omics data and each element in the disease data.

[0015] In the second aspect, a device for predicting the association between omics data and disease data is also provided, the device comprising: a data processing module, used to acquire multiple omics data and at least one disease data, and construct a first association matrix and a second association matrix, wherein the first association matrix represents the association relationship between each element in any two omics data, and the second association matrix represents the association relationship between each element in each omics data and each element in the disease data; a relationship graph construction module, used to construct a characteristic relationship graph according to the first association matrix and the second association matrix, the characteristic relationship graph includes the interactive relationship between each omics data, wherein the interactive relationship between any two omics data is determined according to the association characteristics of each element in each omics data in the two omics data and each element in the at least one disease data; an association prediction module, used to predict the association relationship between the multiple omics data and the at least one disease data based on the characteristic relationship graph.

[0016] According to a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described in the first aspect.

[0017] According to a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, and the program is executed by a processor to implement any method described in the first aspect.

[0018] In general, this application has at least the following beneficial effects:

[0019] A method for predicting the association between omics data and disease data is provided. The correlation between omics elements and the known association between omics elements and disease elements is used to construct a characteristic relationship graph, so as to fully explore the implicit high-order relationship between each omics element and omics element, and between omics elements and disease elements, and realize the prediction of potential associations between multi-omics data and disease data through characteristic relationship graphs.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the multiple drawings represent the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings only depict some embodiments disclosed in the present application and should not be regarded as limiting the scope of the present application. In addition, the same reference numerals are used to represent the same parts in all the drawings.

[0022] Figure 1 A flowchart showing a method for predicting the association between omics data and disease data provided in an embodiment of the present application is shown;

[0023] Figure 2 A schematic diagram showing a process of a method for predicting association between omics data and disease data in one example;

[0024] Figure 3 A schematic diagram showing the structure of a device for predicting association between omics data and disease data provided in an embodiment of the present application is shown;

[0025] Figure 4 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0026] Figure 5 A schematic diagram of a storage medium provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] The following embodiments of the technical solution of the present application are described in detail in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which this application belongs; the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" in the specification and claims of this application and the above-mentioned figure descriptions and any variations thereof are intended to cover non-exclusive inclusions.

[0029] In the description of the embodiments of the present application, the technical terms "first", "second", etc. are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise clearly and specifically defined.

[0030] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0031] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0032] Omics data include but are not limited to genomics, transcriptomics, proteomics, metabolomics, microbiome, exposure genomics, etc. By using multi-omics to quantify and analyze biomolecular features at multiple levels, it can help understand the molecular relationships and functions in biological systems, thus playing a vital role in pathological research.

[0033] Studying the relationship between multi-omics and diseases is conducive to understanding biological processes, discovering new therapeutic targets and biomarkers, and thus providing key scientific support for disease prevention and diagnosis. In related technologies, gene co-expression analysis is used to calculate the correlation between multi-omics data. However, this method only studies the relationship between different omics, but does not study the association between omics data and disease data. Not only is there a problem of low data utilization efficiency, but it is also not conducive to understanding the pathogenesis of diseases at the molecular level, and the analysis accuracy is not high.

[0034] In order to solve the above problems, an embodiment of the present application provides a method for predicting the association between omics data and disease data, which constructs a feature relationship graph by utilizing the correlation between omics and the known association between omics and diseases, thereby fully exploring the implicit high-order relationships between omics and omics, and between omics and diseases, and realizing the prediction of potential associations between multi-omics data and disease data through feature relationship graphs.

[0035] Figure 1 A flowchart showing a method for predicting the association between omics data and disease data provided in an embodiment of the present application is shown. Figure 1 The method comprises the following steps S101 to S103:

[0036] S101, acquiring a plurality of omics data and at least one disease data, and constructing a first association matrix and a second association matrix.

[0037] In one example, the multiple omics data may include genomics, transcriptomics, proteomics, metabolomics, epigenetics, miRNA, lncRNA, mRNA, etc., which are closely related to the pathogenesis of the disease. Disease data may be any data describing the clinical characteristics, pathological type, disease subtype, disease progression stage, and other detailed information of the disease.

[0038] In one example, multiple omics data and at least one disease data can be obtained through existing databases. For example, for miRNA, it can be obtained from the HMDD v2.0 (Human microRNA Disease Database) database, which is a database that specifically collects and organizes the association between human microRNA (miRNA) and diseases. Similarly, for lncRNA, lncRNA and disease association data can be downloaded from LncRNADisease v2.0 and Lnc2Cancer 3.0. LncRNADisease v2.0 specifically collects and organizes the association data between long non-coding RNA (lncRNA) and human diseases. This database not only contains experimentally supported associations between lncRNA and diseases, but also provides tools for predicting the association between new lncRNA and diseases. Lnc2Cancer 3.0 is a database that specifically collects and organizes the association between long non-coding RNA (lncRNA) and cancer. This database contains experimentally supported association data between lncRNA and various cancer subtypes, as well as regulatory mechanisms, biological functions and clinical application information of cancer-related lncRNA. lncRNA-miRNA association data can also be obtained from the Starbase v2.0 database, which is a database specifically used to decode miRNA-ceRNA (competing endogenous RNA), miRNA-ncRNA (non-coding RNA) and protein-RNA interaction networks. That is, this embodiment can extract the omics data in the database, the relationship between the elements of the omics data and the elements of the omics data, and the relationship between the elements of the omics data and the disease elements, and then construct the first association matrix and the second association matrix.

[0039] Among them, the omics element refers to the smallest unit in the omics data, for example, the omics data includes the expression level, subcellular localization, and interaction with other molecules of lncRNA. The corresponding omics element is the lncRNA molecule. That is to say, for omics data, the omics element of this embodiment refers to the molecular level information of omics data.

[0040] For example, if the omics data include DNA methylation levels, histone modifications, and chromatin accessibility, the corresponding omics elements are methylation sites, histone modification sites, open chromatin regions, etc.

[0041] In this embodiment, the elements of the disease data may be basic units formed by disease data such as genetic diseases, biochemical diseases, physiological diseases, and pathological diseases.

[0042] In this embodiment, the first association matrix represents the association relationship between each element in any two omics data, and the second association matrix represents the association relationship between each element in each omics data and each element in the disease data. Continuing the above example, the first association matrix constructed may include a miRNA-lncRNA association matrix, that is, the association relationship between each miRNA molecule of human microRNA omics and the lncRNA molecule of long non-coding RNA omics. The second association matrix constructed may include a miRNA-disease association matrix, that is, the association relationship between the miRNA molecules of human microRNA omics and disease elements, and the second association matrix constructed may also include a lncRNA-disease association matrix, that is, the association relationship between the long non-coding RNA omics tuple and the disease element.

[0043] S102, constructing a characteristic relationship diagram according to the first correlation matrix and the second correlation matrix, wherein the characteristic relationship diagram includes the interactive relationship between each omics data.

[0044] In this embodiment, the interactive relationship can be an interaction and influence relationship between biological molecules, such as an interaction relationship, a causal relationship, a similarity relationship, and an association strength. For example, if the correlation between two omics data is large, it is considered that the two omics data have an interactive relationship. For example, if a disease is greatly affected by an omics element in a set of omics data, it is considered that the omics data and the disease data have an interactive relationship. For another example, if the omics elements in two omics data are strongly associated with the same disease element, there may be a direct biological interaction between them or they may participate in the same biological process together.

[0045] Among them, the interactive relationship between any two omics data is determined according to the correlation characteristics of each element in each omics data of the two omics data and each element in at least one disease data. That is to say, the correlation characteristics of each element in each omics data of the two omics data and each element in at least one disease data are used as factors to determine the relationship between each omics data, so that the created feature relationship diagram not only contains the relationship between each element of the omics data, but also integrates the relationship characteristics between each element of each omics data and each element of the disease data. In this way, the implicit high-order relationship between omics elements and between omics elements and disease elements can be fully explored, and the potential association between multi-omics and diseases can be predicted.

[0046] For example, two omics data include omics data A and omics data B, and omics data A and omics data B are correlated in omics element characteristics, and omics data A and disease C have correlation characteristics. Then, the correlation between omics data B and disease C can be predicted based on the relationship between omics data A and omics data B and the correlation characteristics between omics data A and disease C.

[0047] S103: predicting the association between the plurality of omics data and at least one disease data based on the feature relationship graph.

[0048] After the feature relationship graph is constructed in the above step S102, since the feature relationship graph not only includes the association relationship between the elements in any two omics data, but also includes the association relationship between the elements in each omics data and the elements in the disease data, the association relationship between multiple omics data and at least one disease data can be predicted based on the feature relationship graph through a prediction network. In one example, the prediction network can be a neural network.

[0049] The association relationship between the multiple omics data and at least one disease data includes the association relationship between the elements in any two omics data, and the association relationship between the elements in each omics data and the elements in the disease data.

[0050] The above embodiment constructs a first correlation matrix that represents the correlation between each element in any two omics data, and constructs a second correlation matrix that represents the correlation between each element in each omics data and each element in the disease data, and then constructs a feature relationship graph through the first correlation matrix and the second correlation matrix, so that the created feature relationship graph not only includes the relationship between omics elements and omics elements, but also integrates the relationship characteristics between each omics element and the disease element. In this way, the implicit high-order relationship between omics and omics and diseases can be fully explored, and the potential correlation between multiple omics and diseases can be predicted.

[0051] In an embodiment of the present application, a feature relationship diagram is constructed based on a first association matrix and a second association matrix, including: extracting a first association feature between each element of each omics data from the first association matrix; for any omics data, extracting a second association feature between each element of the omics data and each element of at least one disease data from the second association matrix; after determining each omics data in a plurality of omics data and its corresponding second association feature, determining the interactive relationship between each omics data based on the first association feature and the second association feature of each omics data, so as to construct a feature relationship diagram based on the interactive relationship.

[0052] In one example, since the first association matrix describes the association relationship between the omics elements of different omics data, the first association features characterize the features of the association between the omics elements of each omics data and the omics elements of other omics data. These features may include correlation coefficients, statistically significant features reflecting the strength and direction of the interaction between omics data.

[0053] Similarly, the second association matrix describes the association between each element in the omics data and each element in at least one disease data. The second association feature characterizes the characteristics of the association between each element in each omics data and each element in the disease data. For example, these features can show the association between the expression level of a gene and the risk of a specific disease, or the association between the abundance of a protein and the disease process.

[0054] In one example, 0 to 1 can be used to represent the strength of association between different omics elements, and the strength of association between omics elements and disease elements. For example, 1 can be used to represent that there is a strong first association feature between two omics elements, or that there is a strong second association relationship between an omics element and a disease element, and 0 can be used to represent that there is no first association feature between two omics elements, or that there is no second association relationship between an omics element and a disease element, so as to simplify the process of constructing the feature relationship diagram.

[0055] In one example, taking genomic data A and proteomics data B as examples, the first association matrix can be an m*n matrix, where m genes, n proteins, and Aij represents the relationship between the ith gene and the jth protein. Taking genomic data A and disease omics data C as examples, the second association matrix can be an x*y matrix, where x genes, y diseases, and Aij represents the relationship between the ith gene and the jth disease.

[0056] Combining this information, a feature relationship graph can be constructed, where the edge between A and B represents their association in gene expression and protein abundance, and the edge between A and C represents their association in gene expression and disease. Constructing the feature relationship graph in this way enables visualization of the relationship graph, as well as the complex relationships between expression omics data and their connection to diseases.

[0057] In an embodiment of the present application, a feature relationship graph is constructed based on the interactive relationship, including: taking each omics data in a plurality of omics data as a node of the feature relationship graph, taking each omics data and its corresponding second associated feature as the initial feature of the corresponding omics node; obtaining the edge of the feature relationship graph based on the first associated features of each omics data, and constructing the feature relationship graph.

[0058] In this embodiment, the type of feature relationship graph includes a heterogeneous hypergraph, which is a hypergraph with multiple types of nodes or hyperedges. This heterogeneous hypergraph composed of multiple types of nodes and hyperedges can represent more information. Different from the general graph structure in which each edge only connects two nodes, each hyperedge in the heterogeneous hypergraph can connect more than two nodes, so it is more flexible and can model high-order relationships between targets.

[0059] This embodiment uses To represent a hypergraph, where V is the set of nodes v, ε is the set of hyperedges e, and the positive weight of each hyperedge e is w(e). It can be represented by the association matrix H of |V|×|ε|, which is defined as:

[0060]

[0061] In this embodiment, V={V1,...,V O} is a node set containing O kinds of omics. Each omics data in multiple omics data is used as a node in the feature relationship graph. Each omics data o and its corresponding second associated feature are used as the initial feature of the corresponding node of the omics. The initial feature can be expressed as represents the number of nodes contained in the omics o, M represents the number of all nodes, and D represents the node feature dimension.

[0062] Compared with directly taking a large omics category as a node, this embodiment takes each omics in the omics data as a unit node, which can be described more clearly when constructing hyperedge relationships, and can better utilize each omics element. For example, it can describe which miRNAs are associated with which lncRNAs, and which miRNAs are not associated with which lncRNAs, and can make full use of the known associations between the above omics elements and disease elements, which is conducive to the subsequent inference of the potential associations between them, and at the same time can ensure the consistency of the characteristic dimensions of the nodes.

[0063] For example, for the miRNA-disease association matrix, each row represents a miRNA, each column represents a disease, and the initial node feature of the miRNA is the i-th row in the miRNA-disease association matrix. Similarly, for all omics, the initial node features corresponding to the omics can be found in the omics-disease association matrix (the second association matrix).

[0064] Optionally, the weighted k-neighborhood profile (WKNNP) algorithm can be used to smooth each omics-disease association matrix to compensate for the sparsity of the association between each omics and the disease, and then the corresponding initial node features are found in each smoothed omics-disease association matrix.

[0065] In the embodiment of the present application, the edge of the feature relationship graph is obtained according to the first correlation feature of each omics data. As long as there is a relationship between the omics data, they are constructed into the same hyperedge to construct the feature relationship graph.

[0066] For example, miRNA m i Gene g i 、lncRNAl i Related, lncRNAl i mRNA r i Related, mRNA i miRNA m j If miRNA m i 、miRNA m j , gene g i 、lncRNA l i 、mRNA i are contained in the same hyperedge.

[0067] In the related art, if a model is constructed based on the above relationship, at least four steps are required to establish four edges, respectively connecting miRNA with gene, miRNA with lncRNA, lncRNA with mRNA, and mRNA with miRNA, so that information can be transmitted between miRNA and miRNA. This has high computational cost and will bring too much noise. However, the embodiment of the present application can directly characterize the complex multivariate nonlinear implicit high-order relationship between omics elements and omics elements.

[0068] In one example, if disease data is used as a hyperedge, it is impossible to ensure that the node feature dimensions of each omics data are consistent. For example, there are only three omics, miRNA, lncRNA, and gene. Taking miRNA as an example, its node feature can be composed of its association with all lncRNAs and genes. The associated position is 1, and the unassociated position is 0. This will result in: there will be a lot of 0s, making the matrix sparse, affecting model learning; for miRNA, its feature dimension is the number of lncRNAs + the number of genes, while for lncRNA, the feature dimension is the number of miRNAs + the number of genes, which will cause the feature dimensions corresponding to each omics to be different in length and cannot be input into the neural network for processing. If disease data is used as a hyperedge, using a multi-layer perceptron MLP to compress the node feature dimensions to the same dimension will result in information loss, and since the MLP layer cannot share parameters (the input feature dimensions are different, so there can only be a corresponding MLP layer for each omics), this will increase the complexity of the model and may reduce the training speed and generalization ability.

[0069] Therefore, this embodiment uses the association matrix between omics and diseases as the omics node feature, and the relationship between omics elements as the hyperedge. This ensures that the node feature dimensions are consistent and the matrix sparsity problem does not occur. At the same time, the correlation between omics elements and the known association between omics elements and disease elements can be used to fully explore the multivariate implicit high-order relationships between omics and between omics elements and disease elements.

[0070] In an embodiment of the present application, after obtaining multiple omics data and at least one disease data, the method of this embodiment also includes: extracting omics element names and disease element names from a target database; obtaining a disease name set based on the presence of the disease element names in a preset disease vocabulary; obtaining an omics name set based on the presence of the omics element names in a preset omics vocabulary; and initializing the association relationship between each omics element of the multiple omics data and between each omics element and a disease element of at least one disease data based on each disease element name in the disease name set and each omics element name in the omics name set.

[0071] In this embodiment, the target database includes all omics databases or disease databases. For example, the target database includes the above-mentioned HMDD v2.0 database, LncRNADisease v2.0 and Lnc2Cancer 3.0, Starbase v2.0, etc.

[0072] The preset disease word library may be a Medical Subject Headings (MeSH) database, and the preset omics word library may be, for example, a NCBI (National Center for Biotechnology Information) database.

[0073] The omics element names and disease element names are extracted from the target database, and the disease element names are obtained according to the presence of the disease element names in the preset disease word library, so as to obtain the disease name set, and the disease names recognized and included in the MeSH database can be determined, so as to achieve the unification of the disease names. The omics element names are obtained according to the presence of the omics element names in the preset omics word library, so as to determine the omics names recognized and included in the NCBI database, so as to achieve the unification of the omics names.

[0074] Based on the names of each disease element in the disease name set and the names of each omics element in the omics name set, the association between multiple omics data and at least one disease data is initialized. That is to say, the relationship between omics data and a certain disease, or the relationship between omics data and omics data can be preliminarily judged based on whether the disease name and omics name exist in the omics data or the disease data.

[0075] In an embodiment of the present application, based on each disease element name in the disease name set and each omics element name in the omics name set, the association relationship between each omics element of the multiple omics data and between each omics element and the disease element of at least one disease data is initialized, including: based on the presence of the disease element name in the disease name set in any of the omics data, determining that the disease element corresponding to the disease element name has a second association relationship with the omics element of the omics data; based on the absence of the disease element name in the disease name set in any of the omics data, determining that the second association relationship between the disease element corresponding to the disease element name and the omics data is an unknown relationship; based on the presence of the omics element name in the omics name set in at least two of the omics data, determining that the omics elements in the at least two of the omics data have a first association relationship.

[0076] Specifically, after obtaining the disease name set and the omics name set, a name query can be performed in the multiple omics data and at least one disease data obtained in step S101. For example, when the disease element name M in the disease name set exists in the omics data A, it is determined that the omics data A has a second association relationship with the disease M. When the disease element name M in the disease name set does not exist in the omics data A, it is determined that the omics data A does not have a second association relationship with the disease M. That is, the second association relationship is an unknown relationship, wherein a known association relationship can be represented as 1, and an unknown association relationship can be represented as 0.

[0077] The omics data include omics data A, omics data B, and omics data C. When the disease name M cannot be found in any of the omics data A, B, and C, it is considered that there is no second association relationship between the omics data A, B, and C and the disease M.

[0078] At the same time, based on the fact that the omics element names in the omics name set exist in at least two omics data, it is determined that at least two omics data have a first association relationship. That is, when the same omics element noun appears in multiple omics data, it can be represented that the multiple omics data have an association relationship.

[0079] By unifying the names of omics elements and disease elements, and based on the names of each disease element in the disease name set and each omics element in the omics name set, the association relationship between each omics element of multiple omics data and between each omics element and the disease element of at least one disease data, data standardization can be achieved, and known association relationships and unknown association relationships can be quickly identified.

[0080] In an embodiment of the present application, the above method also includes: constructing a disease similarity matrix based on the similarity between each disease element in the disease name set; and updating the initial features of each node in the feature relationship graph based on the disease similarity matrix.

[0081] In the above embodiment, a set of disease names is obtained, and then the similarity between each disease in the disease name set can be judged based on the disease database to construct a disease similarity matrix. For example, the disease-disease similarity matrix can be calculated through the directed acyclic graph on the MeSH database. The initial features of each node in the feature relationship diagram are updated according to the disease similarity matrix. Each omics-disease association matrix can be multiplied with the disease-disease similarity matrix to obtain an updated omics-disease association matrix, and then the updated association features of each omics and disease are extracted as the updated initial node features. In this way, the constructed feature relationship diagram not only includes the direct association between omics data and diseases, but also takes into account the similarity between diseases.

[0082] After obtaining the feature relationship graph, this embodiment uses a prediction network to learn the association relationships in the feature relationship graph, and then predicts the multivariate implicit high-order relationships between the omics elements and the omics elements-disease elements.

[0083] Specifically, in an embodiment of the present application, the association relationship between multiple omics data and at least one disease data is predicted based on a feature relationship graph, including: updating the node features and edge depth features of the feature relationship graph to obtain an updated second association matrix; calculating the importance weights of different omics nodes in each edge of the feature relationship graph; and predicting the association relationship between multiple omics data and at least one disease data based on the importance weights and the updated second association matrix.

[0084] Specifically, the node features and edge deep features of the feature relationship graph are updated through the heterogeneous node encoder to obtain an updated second association matrix, and the importance weights of different omics nodes in each edge of the feature relationship graph are calculated through the heterogeneous hyperedge encoder.

[0085] In an embodiment of the present application, the node features and the deep features of the edges of the feature relationship graph are updated to obtain the updated second association matrix, including: aggregating the deep features of the edges of the feature relationship graph to the nodes of the feature relationship graph, and updating the node features of the feature relationship graph; calculating the importance weights of different omics nodes in each edge of the feature relationship graph, and updating the deep features of the edges according to the importance weights; decoding the updated node features and the updated deep features of the edges to obtain an updated second association matrix.

[0086] This embodiment uses a variational autoencoder to obtain an updated second association matrix. The variational autoencoder can learn a heterogeneous hypergraph embedded in a low-dimensional vector space. The variational autoencoder includes a heterogeneous hypergraph encoder and a hypergraph decoder. The heterogeneous hypergraph encoder is used to infer the required embedded reasoning model. The heterogeneous hypergraph encoder specifically includes a heterogeneous node encoder and a heterogeneous hyperedge encoder.

[0087] Figure 2 A schematic diagram showing a process of predicting the association between omics data and disease data in an example is shown. Figure 2As shown, the miRNA-lncRNA association matrix, miRNA-gene association matrix, and lncRNA-gene association matrix are obtained by acquiring multiple omics data, and the genes in them represent disease genes. A heterogeneous hypergraph is constructed based on the miRNA-lncRNA association matrix, the miRNA-gene association matrix, and the lncRNA-gene association matrix to obtain a hypergraph association matrix, which contains nodes and hyperedges. The node features are then spliced ​​and updated through the heterogeneous node encoder to obtain an updated multi-omics-disease hypergraph. At the same time, the hyperedge features are updated through the hyperedge encoder to obtain the updated hyperedge features, and then the final multi-omics-disease association matrix is ​​obtained based on the heterogeneous hypergraph decoder.

[0088] Specifically, through the heterogeneous node encoder, the heterogeneous space H is first nonlinearly projected into the common space using the hidden layer The hyperedge feature F ε Aggregate to the nodes and update the node features of the feature relationship graph. Optionally, there can be multiple hidden layers.

[0089] Two separate fully connected layers are then used to estimate the variational distribution The mean and variance

[0090]

[0091]

[0092] W and b are learnable weights and biases, respectively. The node representation obtained after the heterogeneous node encoder It can be calculated as:

[0093]

[0094] in ⊙ represents element-wise product.

[0095] This embodiment also uses a hidden layer to transform the heterogeneous space H T Non-linear projection into common space Node Features Aggregate onto hyperedges, optionally, the hidden layers can have multiple layers.

[0096] For diseases, each type of omics should have different degrees of influence, so adding an attention mechanism to the heterogeneous hyperedge decoder can learn the weights of different omics nodes in each hyperedge. The importance weight of the o-th type of omics node in all N hyperedges is expressed as The omics impact vector is represented by Indicates that K is randomly initialized.

[0097] Processed with a fully connected layer

[0098]

[0099] Then the importance weight It can be calculated as:

[0100]

[0101]

[0102] The hyperedge representation after adding the importance degree can be calculated as:

[0103]

[0104] Similarly, two separate fully connected layers are used to estimate the variational distribution The mean μ ε and variance σ ε The hyperedge representation obtained after the heterogeneous hyperedge encoder is It can be calculated as:

[0105] Z ε =μ ε +σ ε ⊙∈

[0106] After updating the nodes and edges in the above manner, the updated hypergraph is used as the original hypergraph topology using the heterogeneous hypergraph decoder. In the decoder, the node features are converted into and the hyperedge feature G ε The inner product is performed and then decoded into random variables, from which the original hypergraph structure is reconstructed. The loss function consists of the reconstruction error and the KL divergence. The reconstruction error reflects the performance of the model in reconstructing the input, and the KL divergence is a regularizer that measures the distance between the two distributions.

[0107] After training the variational autoencoder, the output of the heterogeneous node encoder As the updated association matrix of o omics and diseases, it is multiplied by the learned importance weight α and normalized to obtain the second association matrix, namely the multi-omics-disease association matrix:

[0108]

[0109]

[0110]

[0111] || indicates splicing. Y finalEach row represents each miRNA, mRNA, lncRNA, gene, protein, etc., and each column represents a disease. Each element in the matrix represents the potential association prediction score between miRNA / mRNA / lncRNA / gene / protein, etc. and the disease.

[0112] In the above embodiment, the known association between multi-omics data and disease data is used as the initial node feature. The prediction network will update the node feature during learning, which is equivalent to directly updating the association between multi-omics data and disease data. The step of calculating the association degree by using inner product or cosine similarity after learning the node features in the related technology is omitted, which saves calculation and improves efficiency. At the same time, the importance weight is learned, which enables the prediction network to better distinguish the impact of each type of omics on the disease.

[0113] The method of the embodiment of the present application also includes: obtaining a prediction network through training according to a training sample set, the prediction network is used to predict the association relationship between multiple omics data and at least one disease data based on a feature relationship graph, the training sample set includes: the association relationship between multiple omics data samples and multiple disease data samples respectively, the association relationship includes the association relationship between each element in any two omics data, and the association relationship between each element in each omics data and each element in the disease data.

[0114] The omics-disease association data set was used to construct the omics-disease association matrix, with 80% of the omics-disease associations used as the training set and 20% of the omics-disease associations used as the test set. During training, the values ​​of the corresponding positions in the association matrix for the test were set to 0.

[0115] The above-mentioned embodiment utilizes the correlation between omics elements and the known association between omics elements and disease elements to construct a feature relationship graph, thereby fully exploring the implicit high-order relationships between omics elements and omics elements, and between omics elements and disease elements, and realizing the prediction of potential associations between multi-omics data and disease data through the feature relationship graph.

[0116] In the embodiment of the present application, Figure 3 A schematic diagram of a device for predicting the association between omics data and disease data is shown in FIG. Figure 3 The device 300 for predicting association between omics data and disease data includes:

[0117] The data processing module 301 is used to obtain multiple omics data and at least one disease data, and construct a first correlation matrix and a second correlation matrix, wherein the first correlation matrix represents the correlation between each element in any two omics data, and the second correlation matrix represents the correlation between each element in each omics data and each element in the disease data;

[0118] A relationship graph construction module 302 is used to construct a characteristic relationship graph according to the first association matrix and the second association matrix, wherein the characteristic relationship graph includes the interactive relationship between each omics data, wherein the interactive relationship between any two omics data is determined according to the correlation characteristics of each element in each omics data of the two omics data and each element in the at least one disease data;

[0119] The association prediction module 303 is used to predict the association relationship between the multiple omics data and the at least one disease data based on the feature relationship graph.

[0120] Optionally, the data processing module 301 is also used to extract the first association feature between each element of each omics data from the first association matrix; for any omics data, extract the second association feature between each element in the omics data and each element in at least one disease data from the second association matrix; after determining each omics data in the multiple omics data and its corresponding second association feature, determine the interactive relationship between each omics data according to the first association feature and the second association feature of each omics data, so as to construct the feature relationship diagram according to the interactive relationship.

[0121] Optionally, the relationship graph construction module 302 is also used to use each of the multiple omics data as a node of the feature relationship graph, and use each omics data and its corresponding second associated feature as the initial feature of the corresponding omics node; obtain the edge of the feature relationship graph according to the first associated feature of each omics data, and construct the feature relationship graph.

[0122] Optionally, the data processing module 301 is also used to extract omics element names and disease element names from the target database; obtain a disease name set based on the disease element names existing in a preset disease vocabulary; obtain an omics name set based on the omics element names existing in a preset omics vocabulary; based on each disease element name in the disease name set and each omics element name in the omics name set, initialize the association relationship between each omics element of the multiple omics data and between it and the disease element of at least one disease data.

[0123] Optionally, the data processing module 301 is further used to determine that the disease element corresponding to the disease element name and the omics element of the omics data have a second association relationship based on the presence of the disease element name in the disease name set in any of the omics data; determine that the second association relationship between the disease element corresponding to the disease element name and the omics data is an unknown relationship based on the absence of the disease element name in the disease name set in any of the omics data; and determine that the omics elements in the at least two omics data have a first association relationship based on the presence of the omics element name in the omics name set in at least two of the omics data.

[0124] Optionally, the relationship graph construction module 302 is further used to construct a disease similarity matrix according to the similarity between each disease element in the disease name set; and update the initial feature of each node in the feature relationship graph according to the disease similarity matrix.

[0125] Optionally, the association prediction module 303 is used to update the node features and edge depth features of the feature relationship graph to obtain an updated second association matrix; calculate the importance weights of different omics nodes in each edge of the feature relationship graph; and predict the association relationship between the multiple omics data and the at least one disease data based on the importance weights and the updated second association matrix.

[0126] Optionally, the association prediction module 303 is also used to update the node features and the deep features of the edges of the feature relationship graph to obtain an updated second association matrix, including: aggregating the deep features of the edges of the feature relationship graph to the nodes of the feature relationship graph, and updating the node features of the feature relationship graph; calculating the importance weights of different omics nodes in each edge of the feature relationship graph, and updating the deep features of the edges according to the importance weights; decoding the updated node features and the updated deep features of the edges to obtain an updated second association matrix.

[0127] Optionally, the association prediction module 303 is further used to obtain a prediction network based on training sample sets, and the prediction network is used to predict the association relationship between the multiple omics data and the at least one disease data based on the feature relationship graph, and the training sample set includes: the association relationship between multiple omics data samples and multiple disease data samples respectively.

[0128] The above description of various embodiments tends to emphasize the differences between the various embodiments. The same or similar aspects can be referenced to each other, and for the sake of brevity, they will not be repeated herein.

[0129] The embodiments of the present application also provide an electronic device corresponding to the method for predicting the association between omics data and disease data provided in the aforementioned embodiments, so as to execute the aforementioned method for predicting the association between omics data and disease data.

[0130] Please refer to Figure 4 , which shows a schematic diagram of an electronic device provided by some embodiments of the present application. Figure 4 As shown, the electronic device 20 includes: a processor 200, a memory 201, a bus 202 and a communication interface 203, and the processor 200, the communication interface 203 and the memory 201 are connected via the bus 202; the memory 201 stores a computer program that can be run on the processor 200, and the processor 200 executes the method provided in any of the aforementioned embodiments of the present application when running the computer program.

[0131] The memory 201 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 203 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0132] The bus 202 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs, and the processor 200 executes the programs after receiving the execution instructions. The association prediction method between omics data and disease data disclosed in any implementation of the embodiments of the present application may be applied to the processor 200, or implemented by the processor 200.

[0133] The processor 200 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 200. The above processor 200 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor to be executed, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 201, and the processor 200 reads the information in the memory 201 and completes the steps of the above method in combination with its hardware.

[0134] The electronic device provided in the embodiment of the present application and the method for predicting the association between omics data and disease data provided in the embodiment of the present application are based on the same application concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0135] The present application also provides a computer-readable storage medium corresponding to the method for predicting the association between omics data and disease data provided in the above embodiment. Figure 5 The computer-readable storage medium shown is a CD 30 on which a computer program (i.e., a program product) is stored. When the computer program is executed by a processor, the association prediction method between omics data and disease data provided in any of the aforementioned embodiments will be executed.

[0136] It should be noted that examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0137] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the method for predicting the association between omics data and disease data provided in the embodiments of the present application are based on the same application concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0138] It should be noted that:

[0139] In the above text, the term "comprises", "includes" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be noted that the scope of the method and device in the embodiment of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0140] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0141] The embodiments of the present application are described above in conjunction with the accompanying drawings, which are only specific implementation modes of the present application. However, the present application is not limited to the above-mentioned specific implementation modes. The above-mentioned specific implementation modes are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the claims, all of which are within the protection of the present application.

Claims

1. A method for predicting the association between omics data and disease data, characterized in that: The method comprises: Acquire multiple omics data and at least one disease data, and construct a first association matrix and a second association matrix, wherein the first association matrix represents the association relationship between each element in any two omics data, and the second association matrix represents the association relationship between each element in each omics data and each element in the disease data; Constructing a characteristic relationship diagram according to the first association matrix and the second association matrix, wherein the characteristic relationship diagram includes the interactive relationship between each omics data, wherein the interactive relationship between any two omics data is determined according to the association characteristics of each element in each omics data of the two omics data and each element in the at least one disease data; The association relationship between the plurality of omics data and the at least one disease data is predicted based on the feature relationship graph.

2. The method according to claim 1, characterized in that The constructing a feature relationship graph according to the first association matrix and the second association matrix includes: Extracting first correlation features between elements of each omics data from the first correlation matrix; For any omics data, extracting second association features between each element in the omics data and each element in at least one disease data from the second association matrix; After determining each omics data and its corresponding second association feature in the plurality of omics data, the interactive relationship between each omics data is determined according to the first association feature and the second association feature of each omics data, so as to construct the feature relationship graph according to the interactive relationship.

3. The method according to claim 2, characterized in that Constructing the feature relationship graph according to the interactive relationship includes: Taking each omics data in the plurality of omics data as a node of the feature relationship graph, and taking each omics data and its corresponding second associated feature as the initial feature of the corresponding omics node; The edges of the feature relationship graph are obtained according to the first correlation features of each omics data, and the feature relationship graph is constructed.

4. The method according to claim 1, characterized in that After obtaining a plurality of omics data and at least one disease data, the method further comprises: Extract omics element names and disease element names from the target database; According to the names of the disease elements existing in the preset disease word library, a disease name set is obtained; According to the omics element names existing in the preset omics vocabulary, an omics name set is obtained; Based on each disease element name in the disease name set and each omics element name in the omics name set, the association relationship between each omics element of the plurality of omics data and between each omics element and a disease element of at least one disease data is initialized.

5. The method according to claim 4, characterized in that Initializing the association relationship between the omics elements of the plurality of omics data and between the omics elements and the disease elements of at least one disease data based on the names of the disease elements in the disease name set and the names of the omics elements in the omics name set, including: Based on the presence of a disease element name in the disease name set in any of the omics data, determining that a disease element corresponding to the disease element name has a second association relationship with an omics element of the omics data; Based on the fact that the disease element name in the disease name set does not exist in any of the omics data, determining that the second association relationship between the disease element corresponding to the disease element name and the omics data is an unknown relationship; Based on the fact that the omics element names in the omics name set exist in at least two of the omics data, it is determined that the omics elements in the at least two of the omics data have a first association relationship.

6. The method according to claim 4, characterized in that The method further comprises: Constructing a disease similarity matrix according to the similarity between each disease element in the disease name set; The initial features of each node in the feature relationship graph are updated according to the disease similarity matrix.

7. The method according to claim 1, characterized in that The predicting the association relationship between the plurality of omics data and the at least one disease data based on the characteristic relationship graph comprises: Updating the node features and edge depth features of the feature relationship graph to obtain an updated second association matrix; Calculate the importance weights of different omics nodes in each edge of the feature relationship graph; According to the importance weights and the updated second association matrix, the association relationship between the plurality of omics data and the at least one disease data is predicted.

8. The method according to claim 7, characterized in that Updating the node features and edge depth features of the feature relationship graph to obtain the updated second association matrix includes: Aggregating the deep features of the edges of the feature relationship graph onto the nodes of the feature relationship graph, and updating the node features of the feature relationship graph; Calculating the importance weights of different omics nodes in each edge of the feature relationship graph, and updating the deep features of the edge according to the importance weights; The updated node features and the updated edge depth features are decoded to obtain the updated second association matrix.

9. The method according to any one of claims 1 or 7-8, characterized in that: The method further comprises: A prediction network is obtained by training according to a training sample set, and the prediction network is used to predict the association relationship between the multiple omics data and the at least one disease data based on the feature relationship graph. The training sample set includes: the association relationship between multiple omics data samples and multiple disease data samples respectively, and the association relationship includes the association relationship between each element in any two omics data, and the association relationship between each element in each omics data and each element in the disease data.

10. A device for predicting the association between omics data and disease data, characterized in that: The device comprises: A data processing module, used to obtain multiple omics data and at least one disease data, and construct a first association matrix and a second association matrix, wherein the first association matrix represents the association relationship between each element in any two omics data, and the second association matrix represents the association relationship between each element in each omics data and each element in the disease data; a relationship graph construction module, configured to construct a characteristic relationship graph according to the first association matrix and the second association matrix, wherein the characteristic relationship graph includes the interactive relationship between each omics data, wherein the interactive relationship between any two omics data is determined according to the correlation characteristics between each element in each omics data of the two omics data and each element in the at least one disease data; An association prediction module is used to predict the association relationship between the multiple omics data and the at least one disease data based on the feature relationship graph.

11. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to any one of claims 1 to 9.

Images