MiRNA-disease association prediction method based on adaptive higher-order moment modeling and application

Abstract

The application discloses a miRNA-disease correlation prediction method based on adaptive high-order moment modeling and application, relates to the field of biology, and solves the problem of low efficiency of disease and miRNA correlation identification, and comprises the following steps: collecting multiple groups of correlated diseases and performing analysis to obtain a disease semantic similarity matrix corresponding to the multiple groups of correlated diseases; acquiring multiple groups of correlated RNAs, and calculating RNA function similarity matrix, RNA core similarity matrix and disease core similarity matrix in combination with the relationship between the correlated RNAs and the correlated diseases; constructing a comprehensive disease similarity matrix in combination with the disease semantic similarity matrix and the disease core similarity matrix; constructing a comprehensive RNA similarity matrix in combination with the RNA function similarity matrix and the RNA core similarity matrix; and establishing an adaptive high-order moment model in combination with the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix, so that the corresponding miRNA and disease can be efficiently identified.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics technology, specifically a miRNA-disease association prediction method and its application based on adaptive higher-order moment modeling. Background Technology

[0002] miRNAs are a class of endogenous, single-stranded non-coding RNA molecules that mediate posttranscriptional gene expression regulation by binding to messenger RNA. They play a crucial role in the development and progression of various complex diseases, including cancer, cardiovascular disease, metabolic syndrome, and immune-related diseases. Increasing research indicates that imbalances in miRNA expression profiles are closely related to disease occurrence, progression, and prognosis, making them important diagnostic biomarkers and potential therapeutic targets. Therefore, systematically and accurately identifying miRNA-disease associations not only helps in understanding disease mechanisms at the molecular level but also provides crucial clues for early disease screening, personalized treatment strategy development, and prognostic assessment.

[0003] However, at present, the association between miRNA and disease is usually determined by laboratory identification rather than by building predictive models through mathematical modeling and then using these models to determine the association between miRNA and disease. This makes the identification of the association between miRNA and disease costly and inefficient.

[0004] Therefore, this invention proposes a miRNA-disease association prediction method and its application based on adaptive higher-order moment modeling. Summary of the Invention

[0005] The purpose of this invention is to propose a miRNA-disease association prediction method and its application based on adaptive higher-order moment modeling, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] miRNA-disease association prediction methods and applications based on adaptive higher-order moment modeling, including:

[0008] Step S1: Collect and analyze multiple sets of related diseases to obtain a semantic similarity matrix of diseases corresponding to multiple sets of related diseases;

[0009] Step S2: Obtain multiple sets of associated RNAs, and calculate the RNA functional similarity matrix, RNA nuclear similarity matrix, and disease nuclear similarity matrix by combining the relationship between associated RNAs and associated diseases;

[0010] Step S3: Combine the disease semantic similarity matrix and the disease kernel similarity matrix to construct a comprehensive disease similarity matrix; combine the RNA functional similarity matrix and the RNA kernel similarity matrix to construct a comprehensive RNA similarity matrix;

[0011] Step S4: Construct an RNA-disease heterogeneity map between associated diseases and associated RNAs. Combine the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix to establish an adaptive higher-order moment model. Based on the adaptive higher-order moment model, determine the association probability between unobserved diseases and unobserved miRNAs.

[0012] Further, step S1 includes the following sub-steps:

[0013] Step S11: Denote the two related diseases for which similarity needs to be calculated as Disease A and Disease B, and read the names of Disease A and Disease B respectively;

[0014] Step S12: Query the NCBI database by the name of disease A, record the position of disease A in the NCBI database as node A, record the parent node of node A as node A1, record the parent node of node A1 as node A2, and so on, to obtain all ancestor nodes of node A.

[0015] Step S13: Query the NCBI database by the name of disease B, record the position of disease B in the NCBI database as node B, record the parent node of node B as node B1, record the parent node of node B1 as node B2, and so on, to obtain all ancestor nodes of node B.

[0016] Step S14: Calculate the node similarity between node A and node B by combining all ancestor nodes of node A and node B.

[0017] Step S15: Calculate the node similarity between any two diseases among all related diseases and construct a disease semantic similarity matrix based on the node similarity between all related diseases.

[0018] Furthermore, the calculation process for the node similarity is as follows:

[0019] Step S141: Using node A as the end node and the final ancestor node as the root node, connect all ancestor nodes with parent-child relationships through edges to obtain the directed acyclic graph corresponding to node A.

[0020] Step S142: Identify the shortest path between the root node and node A, and count the number of edges in the shortest path as the depth of node A;

[0021] Step S143: Calculate the depth of node B using the process of steps S141-S142.

[0022] Step S144: Traverse all ancestor nodes of node A and node B, obtain the lowest common ancestor node of node A and node B, and calculate the depth of the lowest common ancestor node following the process of step S143.

[0023] Step S145: Calculate the node similarity between node A and node B by combining the depth of node A, the depth of node B, and the depth of the lowest common ancestor node.

[0024] Further, step S2 includes the following sub-steps:

[0025] Step S201: For any two related RNAs whose functional similarity needs to be calculated, they are defined as Ri and Rj, respectively, where Ri represents one of the related RNAs and Rj represents the other related RNA.

[0026] Step S202: Read all the associated diseases related to Ri and summarize them into a set called the first disease set; summarize all the associated diseases related to Rj into a set called the second disease set.

[0027] Step S203: Calculate the functional similarity between the corresponding associated RNAs by combining the disease sets of the two associated RNAs;

[0028] Step S204: Calculate the functional similarity between any two associated RNAs, and obtain the RNA functional similarity matrix based on the functional similarity of all associated RNAs.

[0029] Furthermore, step S2 also includes the following sub-steps:

[0030] Step S205: Count the total number of all associated diseases n, and number the associated diseases q; count the total number of all associated RNAs m, and number the associated RNAs p; q = 1, 2, ..., n; p = 1, 2, ..., m;

[0031] Step S206: Construct an n×m blank matrix, fill in the blank matrix by the relationship between associated diseases and associated RNAs, and obtain the interaction spectrum matrix of associated diseases and associated RNAs;

[0032] Step S207: Divide the interaction spectrum matrix by column, denote the first column as the interaction spectrum vector of associated RNA 1, the second column as the interaction spectrum vector of associated RNA 2, and so on, to obtain the interaction spectrum vectors of all associated RNAs.

[0033] Step S208: Calculate the nuclear similarity between corresponding associated RNAs based on the interaction spectral vectors of any two associated RNAs;

[0034] Step S209: Calculate the nuclear similarity between all associated RNAs, and construct an RNA nuclear similarity matrix based on the nuclear similarity between all associated RNAs.

[0035] Furthermore, step S2 also includes:

[0036] Step S210: Divide the interaction spectrum matrix by row, denote the first row as the interaction spectrum vector corresponding to disease No. 1, the second row as the interaction spectrum vector corresponding to disease No. 2, and so on, to obtain the interaction spectrum vectors corresponding to all associated diseases.

[0037] Step S211: Calculate the kernel similarity between corresponding associated diseases based on the interaction spectral vectors of any two associated diseases;

[0038] Step S212: Calculate the kernel similarity between all associated diseases, and construct a disease kernel similarity matrix based on the kernel similarity between all associated diseases.

[0039] Further, step S4 includes the following sub-steps:

[0040] Step S41: Integrate all associated diseases and all associated RNAs to obtain a node set V; integrate the combinations of associated diseases and associated RNAs with a relationship to obtain an edge set E; the node set V and the edge set E form a heterogeneous graph G=(V,E); add bidirectional edges between associated diseases and associated RNAs with a relationship to obtain an adjacency matrix A;

[0041] Step S42: Obtain the comprehensive disease similarity matrix, and take each row of the comprehensive disease similarity matrix as the initial attribute vector of the disease node; obtain the comprehensive RNA similarity matrix, and take each row of the comprehensive RNA similarity matrix as the initial attribute vector of the RNA node.

[0042] Furthermore, step S4 also includes the following sub-steps:

[0043] Step S43: Run Node2Vec on the heterogeneous graph to learn a 64-dimensional topological vector for each node; assign a one-hot vector to each disease node and RNA node; concatenate the initial attribute vector, the 64-dimensional topological vector and the one-hot vector and linearly map them to a unified hidden layer dimension H through a fully connected layer to obtain the initial node feature matrix of any disease node or RNA node.

[0044] Step S44: Stack two layers of adaptive higher-order moment convolutional layers, and set the input of each layer of adaptive higher-order moment convolutional layer as the node features and adjacency matrix of the node in the previous layer.

[0045] Furthermore, step S4 also includes the following sub-steps:

[0046] Step S45: Construct disease-RNA pairs from any disease node and RNA node, embed and concatenate the node features of disease nodes and RNA nodes, and input them into the fully connected layer to calculate the predicted probability that the corresponding disease-RNA pairs are associated, and obtain the adaptive higher-order moment model.

[0047] Step S46: Using known associations between associated diseases and associated RNAs as training data, train the adaptive higher-order moment model. After training, predict and output the association probability between any unobserved disease-RNA pair.

[0048] Application of a miRNA-disease association prediction method based on adaptive higher-order moment modeling in a bioinformatics context.

[0049] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0050] 1. This invention first collects and analyzes multiple sets of associated diseases to obtain disease semantic similarity matrices corresponding to multiple sets of associated diseases; then it obtains multiple sets of associated RNAs and calculates RNA functional similarity matrix, RNA nuclear similarity matrix and disease nuclear similarity matrix by combining the relationship between associated RNAs and associated diseases.

[0051] 2. This invention constructs a comprehensive disease similarity matrix by combining a disease semantic similarity matrix and a disease kernel similarity matrix; it also constructs a comprehensive RNA similarity matrix by combining an RNA functional similarity matrix and an RNA kernel similarity matrix; finally, it constructs an RNA-disease heterogeneity graph between associated diseases and associated RNAs, and establishes an adaptive higher-order moment model by combining the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix. Based on the adaptive higher-order moment model, it determines the association probability between unobserved diseases and unobserved miRNAs; by constructing an adaptive higher-order model for miRNAs, it completes the prediction of the association probability between unobserved diseases and miRNAs, and achieves efficient identification of miRNAs and diseases. Attached Figure Description

[0052] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0053] Figure 1 This is a flowchart illustrating the overall method of the present invention;

[0054] Figure 2 This is a framework diagram of the adaptive higher-order moment model in this invention;

[0055] Figure 3 This is a schematic diagram of the computer device in this invention. Detailed Implementation

[0056] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] miRNAs are a class of endogenous, single-stranded non-coding RNA molecules that mediate posttranscriptional gene expression regulation by binding to messenger RNA. They play a crucial role in the development and progression of various complex diseases, including tumors, cardiovascular diseases, metabolic syndromes, and immune-related diseases. Therefore, systematically and accurately identifying miRNA-disease associations is helpful for understanding disease mechanisms at the molecular level and for developing early screening and personalized treatment strategies. This invention focuses on hepatocellular carcinoma (HCC) and refers to the diseases involved in HCC as associated diseases and the miRNAs involved in HCC as associated RNAs.

[0058] Example 1: Please refer to Figure 1 and Figure 2 As shown, the technical solution provided by this invention is as follows: a miRNA-disease association prediction method based on adaptive higher-order moment modeling. This method analyzes multiple groups of associated diseases to obtain a disease semantic similarity matrix, and then analyzes multiple groups of joint RNAs to obtain an RNA functional similarity matrix, a disease kernel similarity matrix, and an RNA kernel similarity matrix. An RNA-disease heterogeneity map is constructed by combining the disease semantic similarity matrix, disease kernel similarity matrix, RNA functional similarity matrix, and RNA kernel similarity matrix corresponding to multiple groups of associated diseases and multiple groups of miRNAs.

[0059] The miRNA-disease association prediction method based on adaptive higher-order moment modeling is as follows:

[0060] Step S1: Collect and analyze multiple sets of associated diseases to obtain a semantic similarity matrix of diseases corresponding to multiple sets of associated diseases; among them, multiple sets of associated diseases are diseases known to be associated with miRNAs, which are public data resources and can be directly downloaded from the human miRNA-disease association database;

[0061] In this invention, step S1 includes the following sub-steps:

[0062] Step S11: Denote the two related diseases for which similarity needs to be calculated as Disease A and Disease B, and read the names of Disease A and Disease B respectively;

[0063] Step S12: Query the NCBI database by the name of disease A, record the position of disease A in the NCBI database as node A, record the parent node of node A as node A1, record the parent node of node A1 as node A2, and so on, to obtain all ancestor nodes of node A.

[0064] Step S13: Query the NCBI database by the name of disease B, record the position of disease B in the NCBI database as node B, record the parent node of node B as node B1, record the parent node of node B1 as node B2, and so on, to obtain all ancestor nodes of node B.

[0065] Step S14: Calculate the node similarity between node A and node B by combining all ancestor nodes of node A and node B.

[0066] The specific process for calculating node similarity is as follows:

[0067] Step S141: Using node A as the end node and the final ancestor node as the root node, connect all ancestor nodes with parent-child relationships through edges to obtain the directed acyclic graph corresponding to node A; it should be noted that the final ancestor node has no parent node.

[0068] Step S142: Identify the shortest path between the root node and node A, and count the number of edges in the shortest path, denoted as the depth SDA of node A;

[0069] Step S143: Calculate the depth SDB of node B using the process of steps S141-S142.

[0070] Step S144: Traverse all ancestor nodes of node A and node B, obtain the lowest common ancestor node of node A and node B, and calculate the depth SDG of the lowest common ancestor node following the process of step S143.

[0071] Step S145: Calculate the node similarity XS(A, B) between node A and node B by combining the depths of node A, node B, and the depth of the lowest common ancestor node. The specific formula is as follows:

[0072] XS(A,B)=2×SDG / (SDA+SDB);

[0073] Step S15: Calculate the node similarity between any two diseases among all related diseases and construct the disease semantic similarity matrix S1 based on the node similarity among all related diseases. Specifically:

[0074] ; where XS(1,2) represents the node similarity between disease 1 and disease 2.

[0075] Step S2: Obtain multiple sets of associated RNAs, and calculate the RNA functional similarity matrix, RNA nuclear similarity matrix, and disease nuclear similarity matrix by combining the relationship between associated RNAs and associated diseases;

[0076] In this invention, step S2 is specifically as follows:

[0077] Step S201: For any two related RNAs whose functional similarity needs to be calculated, they are defined as Ri and Rj, respectively, where Ri represents one of the related RNAs and Rj represents the other related RNA.

[0078] Step S202: Read all the associated diseases related to Ri and summarize them into a set called the first disease set Ji; summarize all the associated diseases related to Rj into a set called the second disease set Jj.

[0079] Step S203: Calculate the functional similarity GXS(Ri, Rj) between the corresponding associated RNAs by combining the disease sets of the two associated RNAs; specifically, calculate the similarity between the two disease sets using the best matching average, and use it as the functional similarity between the corresponding associated RNAs.

[0080] Step S204: Similarly, calculate the functional similarity between any two associated RNAs, and obtain the RNA functional similarity matrix S2 based on the functional similarity of all associated RNAs.

[0081]

[0082] Step S205: Count the total number of all associated diseases n, and number the associated diseases q; count the total number of all associated RNAs m, and number the associated RNAs p; q = 1, 2, ..., n; p = 1, 2, ..., m;

[0083] Step S206: Construct an n×m blank matrix, fill in the blank matrix by the relationship between associated diseases and associated RNAs, and obtain the interaction spectrum matrix of associated diseases and associated RNAs;

[0084] Specifically, if there is an association between the disease associated with q and the RNA associated with p, then the position (p, q) in the blank matrix will be filled with 1; if there is no association between the disease associated with q and the RNA associated with p, then the position (p, q) in the blank matrix will be filled with 0.

[0085] Step S207: Divide the interaction spectrum matrix by column, denote the first column as the interaction spectrum vector of associated RNA 1, the second column as the interaction spectrum vector of associated RNA 2, and so on, to obtain the interaction spectrum vectors of all associated RNAs.

[0086] Step S208: Calculate the nuclear similarity RHX(p1, p2) between corresponding associated RNAs based on the interaction spectral vectors of any two associated RNAs. The specific formula is as follows:

[0087] RHS(p1,p2)=exp(-γ1×||P(p1)-P(p2)|| 2 In the formula, exp represents an exponential function with base e, γ1 is the bandwidth parameter, p1 is the number of one associated RNA, and p2 is the number of the other associated RNA; P(p1) is the interaction spectral vector corresponding to associated RNA p1, and P(p2) is the interaction spectral vector corresponding to associated RNA p2; ||P(p1)-P(p2)|| 2 It is the squared Euclidean distance between two interacting spectral vectors;

[0088] Step S209: Calculate the nuclear similarity between all associated RNAs, and construct the RNA nuclear similarity matrix S3 based on the nuclear similarity between all associated RNAs;

[0089] ;

[0090] Step S210: Divide the interaction spectrum matrix by row, denote the first row as the interaction spectrum vector corresponding to disease No. 1, the second row as the interaction spectrum vector corresponding to disease No. 2, and so on, to obtain the interaction spectrum vectors corresponding to all associated diseases.

[0091] Step S211: Calculate the kernel similarity BHS(q1, q2) between any two associated diseases based on their interaction spectral vectors. The specific formula is as follows:

[0092] BHS(q1,q2)=exp(-γ2×||P(q1)-P(q2)|| 2 In the formula, γ2 is the bandwidth parameter, q1 is the number of one of the associated diseases, q2 is the number of another associated disease; P(q1) is the interaction spectrum vector corresponding to the associated disease q1, and P(q2) is the interaction spectrum vector corresponding to the associated disease q2.

[0093] Step S212: Calculate the kernel similarity among all associated diseases, and construct the disease kernel similarity matrix S4 based on the kernel similarity among all associated diseases. Specifically:

[0094] .

[0095] Step S3: Construct a comprehensive disease similarity matrix by combining the disease semantic similarity matrix and the disease kernel similarity matrix; construct a comprehensive RNA similarity matrix by combining the RNA functional similarity matrix and the RNA kernel similarity matrix; specifically, obtain the comprehensive disease similarity matrix by multiplying the disease semantic similarity matrix by the first coefficient and the disease kernel similarity matrix by the second coefficient; obtain the comprehensive RNA similarity matrix by multiplying the RNA functional similarity matrix by the third coefficient and the RNA kernel similarity matrix by the fourth coefficient.

[0096] Step S4: Construct an RNA-disease heterogeneity map between associated diseases and associated RNAs, and establish an adaptive higher-order moment model by combining the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix. Based on the adaptive higher-order moment model, determine the association probability between unobserved diseases and unobserved miRNAs.

[0097] In this invention, step S4 includes the following sub-steps:

[0098] Step S41: Integrate all associated diseases and all associated RNAs to obtain a node set V; integrate the combinations of associated diseases and associated RNAs with a relationship to obtain an edge set E; the node set V and the edge set E form a heterogeneous graph G=(V,E); add bidirectional edges between associated diseases and associated RNAs with a relationship to obtain an adjacency matrix A;

[0099] Specifically, if there are 3 nodes, associated with disease B1, associated with RNA R1, and associated with RNA R2; and given that B1 is associated with R1, then after adding bidirectional edges, the resulting adjacency matrix A is:

[0100] ;

[0101] Step S42: Obtain the comprehensive disease similarity matrix, and take each row of the comprehensive disease similarity matrix as the initial attribute vector of the disease node; obtain the comprehensive RNA similarity matrix, and take each row of the comprehensive RNA similarity matrix as the initial attribute vector of the RNA node;

[0102] Step S43: Run Node2Vec on the heterogeneous graph to learn a 64-dimensional topological vector for each node; assign a one-hot vector to each disease node and RNA node; concatenate the initial attribute vector, the 64-dimensional topological vector and the one-hot vector and linearly map them to a unified hidden layer dimension H through a fully connected layer to obtain the initial node feature matrix of any disease node or RNA node.

[0103] One-hot vector is an encoding method that converts categorical variables into numerical form. It uses binary vectors to represent discrete categories, with each category corresponding to a unique position in the vector.

[0104] Step S44: Stack two layers of adaptive higher-order moment convolutional layers, and set the input of each layer of adaptive higher-order moment convolutional layer as the node features and adjacency matrix of the node in the previous layer; where the node features are the learned low-dimensional, dense numerical vectors, which are the output vectors of the previous layer of adaptive higher-order moment convolutional layer for each node.

[0105] In this invention, the calculation process of the adaptive higher-order convolutional layer is as follows:

[0106] Step S441: For any node U, calculate the first to maximum order raw moments of its neighborhood features; the specific calculation formula is as follows:

[0107] ;k=1,……,Max;where L represents the neighboring nodes of node U, XUY are the coefficients of any neighboring nodes, TU is the node characteristic of node U, and ⊙k represents the element-wise power of k.

[0108] Step S442: Calculate the soft weights of the adaptive moments based on the Softmax algorithm, and then determine the adaptive moments;

[0109] Step S443: Map the adaptive moments to the latent space and output the node features of the corresponding nodes;

[0110] Step S45: Construct disease-RNA pairs from any disease node and RNA node, embed and concatenate the node features of disease nodes and RNA nodes, and input them into the fully connected layer to calculate the predicted probability that the corresponding disease-RNA pairs are associated, and obtain the adaptive higher-order moment model.

[0111] Step S46: Using known associations between associated diseases and associated RNAs as training data, train the adaptive higher-order moment model. After training, predict and output the association probability between any unobserved disease-RNA pair.

[0112] Example 2: The miRNA-disease association prediction method based on adaptive higher-order moment modeling is applied to disease association prediction, that is, disease association prediction is performed in a bioinformatics context.

[0113] Example 3: As Figure 3As shown, this embodiment provides an electronic device, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor can call logical instructions in the memory to execute a miRNA-disease association prediction method based on adaptive higher-order moment modeling. This method includes: collecting and analyzing multiple sets of associated diseases to obtain disease semantic similarity matrices corresponding to the multiple sets of associated diseases; acquiring multiple sets of associated RNAs and calculating RNA functional similarity matrices, RNA kernel similarity matrices, and disease kernel similarity matrices based on the relationship between associated RNAs and associated diseases; constructing a comprehensive disease similarity matrix by combining the disease semantic similarity matrix and the disease kernel similarity matrix; constructing a comprehensive RNA similarity matrix by combining the RNA functional similarity matrix and the RNA kernel similarity matrix; constructing an RNA-disease heterogeneity graph between associated diseases and associated RNAs; establishing an adaptive higher-order moment model by combining the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix; and determining the association probability between unobserved diseases and unobserved miRNAs based on the adaptive higher-order moment model.

[0114] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] Example 4: This application also provides a computer program product, which includes a computer program stored on a computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the miRNA-disease association prediction method based on adaptive higher-order moment modeling provided by the above methods. The method includes: collecting and analyzing multiple sets of associated diseases to obtain disease semantic similarity matrices corresponding to multiple sets of associated diseases; obtaining multiple sets of associated RNAs and calculating RNA functional similarity matrices, RNA nuclear similarity matrices, and disease nuclear similarity matrices based on the relationship between associated RNAs and associated diseases; constructing a comprehensive disease similarity matrix by combining the disease semantic similarity matrix and the disease nuclear similarity matrix; constructing a comprehensive RNA similarity matrix by combining the RNA functional similarity matrix and the RNA nuclear similarity matrix; constructing an RNA-disease heterogeneity graph between associated diseases and associated RNAs; establishing an adaptive higher-order moment model by combining the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix; and determining the association probability between unobserved diseases and unobserved miRNAs based on the adaptive higher-order moment model.

[0116] Example 5: This application also provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the miRNA-disease association prediction methods based on adaptive higher-order moment modeling provided above. The method includes: collecting and analyzing multiple sets of associated diseases to obtain disease semantic similarity matrices corresponding to multiple sets of associated diseases; acquiring multiple sets of associated RNAs and calculating RNA functional similarity matrices, RNA nuclear similarity matrices, and disease nuclear similarity matrices based on the relationship between associated RNAs and associated diseases; constructing a comprehensive disease similarity matrix by combining the disease semantic similarity matrix and the disease nuclear similarity matrix; constructing a comprehensive RNA similarity matrix by combining the RNA functional similarity matrix and the RNA nuclear similarity matrix; constructing an RNA-disease heterogeneity graph between associated diseases and associated RNAs; establishing an adaptive higher-order moment model by combining the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix; and determining the association probability between unobserved diseases and unobserved miRNAs based on the adaptive higher-order moment model.

[0117] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0118] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A miRNA-disease association prediction method based on adaptive higher-order moment modeling, characterized in that, include: Step S1: Collect and analyze multiple sets of related diseases to obtain a semantic similarity matrix of diseases corresponding to multiple sets of related diseases; Step S2: Obtain multiple sets of associated RNAs, and calculate the RNA functional similarity matrix, RNA nuclear similarity matrix, and disease nuclear similarity matrix by combining the relationship between associated RNAs and associated diseases; Step S3: Combine the disease semantic similarity matrix and the disease kernel similarity matrix to construct a comprehensive disease similarity matrix; combine the RNA functional similarity matrix and the RNA kernel similarity matrix to construct a comprehensive RNA similarity matrix; Step S4: Construct an RNA-disease heterogeneity map between associated diseases and associated RNAs. Combine the comprehensive RNA similarity matrix and the comprehensive disease similarity matrix to establish an adaptive higher-order moment model. Based on the adaptive higher-order moment model, determine the association probability between unobserved diseases and unobserved miRNAs.

2. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Denote the two related diseases for which similarity needs to be calculated as Disease A and Disease B, and read the names of Disease A and Disease B respectively; Step S12: Query the NCBI database by the name of disease A, record the position of disease A in the NCBI database as node A, record the parent node of node A as node A1, record the parent node of node A1 as node A2, and so on, to obtain all ancestor nodes of node A. Step S13: Query the NCBI database by the name of disease B, record the position of disease B in the NCBI database as node B, record the parent node of node B as node B1, record the parent node of node B1 as node B2, and so on, to obtain all ancestor nodes of node B. Step S14: Calculate the node similarity between node A and node B by combining all ancestor nodes of node A and node B. Step S15: Calculate the node similarity between any two diseases among all related diseases and construct a disease semantic similarity matrix based on the node similarity between all related diseases.

3. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 2, characterized in that, The calculation process for node similarity is as follows: Step S141: Using node A as the end node and the final ancestor node as the root node, connect all ancestor nodes with parent-child relationships through edges to obtain the directed acyclic graph corresponding to node A. Step S142: Identify the shortest path between the root node and node A, and count the number of edges in the shortest path as the depth of node A; Step S143: Calculate the depth of node B using the process of steps S141-S142. Step S144: Traverse all ancestor nodes of node A and node B, obtain the lowest common ancestor node of node A and node B, and calculate the depth of the lowest common ancestor node following the process of step S143. Step S145: Calculate the node similarity between node A and node B by combining the depth of node A, the depth of node B, and the depth of the lowest common ancestor node.

4. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 1, characterized in that, Step S2 includes the following sub-steps: Step S201: For any two related RNAs whose functional similarity needs to be calculated, they are defined as Ri and Rj, respectively, where Ri represents one of the related RNAs and Rj represents the other related RNA. Step S202: Read all the associated diseases related to Ri and summarize them into a set called the first disease set; summarize all the associated diseases related to Rj into a set called the second disease set. Step S203: Calculate the functional similarity between the corresponding associated RNAs by combining the disease sets of the two associated RNAs; Step S204: Calculate the functional similarity between any two associated RNAs, and obtain the RNA functional similarity matrix based on the functional similarity of all associated RNAs.

5. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 4, characterized in that, Step S2 further includes the following sub-steps: Step S205: Count the total number of all associated diseases n, and number the associated diseases q; count the total number of all associated RNAs m, and number the associated RNAs p; q = 1, 2, ..., n; p = 1, 2, ..., m; Step S206: Construct an n×m blank matrix, fill in the blank matrix by the relationship between associated diseases and associated RNAs, and obtain the interaction spectrum matrix of associated diseases and associated RNAs; Step S207: Divide the interaction spectrum matrix by column, denote the first column as the interaction spectrum vector of associated RNA 1, the second column as the interaction spectrum vector of associated RNA 2, and so on, to obtain the interaction spectrum vectors of all associated RNAs. Step S208: Calculate the nuclear similarity between corresponding associated RNAs based on the interaction spectral vectors of any two associated RNAs; Step S209: Calculate the nuclear similarity between all associated RNAs, and construct an RNA nuclear similarity matrix based on the nuclear similarity between all associated RNAs.

6. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 5, characterized in that, Step S2 further includes: Step S210: Divide the interaction spectrum matrix by row, denote the first row as the interaction spectrum vector corresponding to disease No. 1, the second row as the interaction spectrum vector corresponding to disease No. 2, and so on, to obtain the interaction spectrum vectors corresponding to all associated diseases. Step S211: Calculate the kernel similarity between corresponding associated diseases based on the interaction spectral vectors of any two associated diseases; Step S212: Calculate the kernel similarity between all associated diseases, and construct a disease kernel similarity matrix based on the kernel similarity between all associated diseases.

7. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 1, characterized in that, Step S4 includes the following sub-steps: Step S41: Integrate all associated diseases and all associated RNAs to obtain a node set V; integrate the combinations of associated diseases and associated RNAs with a relationship to obtain an edge set E; the node set V and the edge set E form a heterogeneous graph G=(V,E); add bidirectional edges between associated diseases and associated RNAs with a relationship to obtain an adjacency matrix A; Step S42: Obtain the comprehensive disease similarity matrix, and take each row of the comprehensive disease similarity matrix as the initial attribute vector of the disease node; obtain the comprehensive RNA similarity matrix, and take each row of the comprehensive RNA similarity matrix as the initial attribute vector of the RNA node.

8. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 7, characterized in that, Step S4 further includes the following sub-steps: Step S43: Run Node2Vec on the heterogeneous graph to learn a 64-dimensional topological vector for each node; assign a one-hot vector to each disease node and RNA node; concatenate the initial attribute vector, the 64-dimensional topological vector and the one-hot vector and linearly map them to a unified hidden layer dimension H through a fully connected layer to obtain the initial node feature matrix of any disease node or RNA node. Step S44: Stack two layers of adaptive higher-order moment convolutional layers, and set the input of each layer of adaptive higher-order moment convolutional layer as the node features and adjacency matrix of the node in the previous layer.

9. The miRNA-disease association prediction method based on adaptive higher-order moment modeling according to claim 8, characterized in that, Step S4 further includes the following sub-steps: Step S45: Construct disease-RNA pairs from any disease node and RNA node, embed and concatenate the node features of disease nodes and RNA nodes, and input them into the fully connected layer to calculate the predicted probability that the corresponding disease-RNA pairs are associated, and obtain the adaptive higher-order moment model. Step S46: Using known associations between associated diseases and associated RNAs as training data, train the adaptive higher-order moment model. After training, predict and output the association probability between any unobserved disease-RNA pair.

10. An application of the miRNA-disease association prediction method based on adaptive higher-order moment modeling as described in any one of claims 1-9 in a bioinformatics context.

Images