Multi-source data fusion method and apparatus, and electronic device and storage medium

Abstract

A multi-source data fusion method and apparatus, and an electronic device and a storage medium. The method comprises: taking multi-source data as different pieces of modal data; respectively constructing sub-hypergraphs; splicing the sub-hypergraphs into a hypergraph; then performing representation learning on a pre-constructed hypergraph neural network model according to the hypergraph, so as to obtain biological data node representations; and predicting an unknown or unobserved association relationship or interaction relationship according to the biological data node representations, so as to realize the construction of the relevance between biological nodes. The relevance makes data of different sources be fused together; the relevance is scored to obtain a prediction score, so as to determine the strength of the relevance; and finally, the hypergraph neural network model is repeatedly trained according to specific downstream tasks, and when a loss function converges, a satisfactory prediction result is obtained, such that fusion of multi-source data is realized.

Description

Multi-source data fusion method, device, electronic device and storage medium

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority to Chinese patent application number CN202310715583.5, entitled “Multi-source data fusion method, device, electronic device and storage medium,” filed with the China Patent Office on June 15, 2023, the entire contents of which are incorporated by reference into this disclosure. Technical Field

[0003] The present application relates to the field of data processing technology, and in particular to a multi-source data fusion method, device, electronic device and storage medium. Background Art

[0004] Biologically heterogeneous networks include protein-protein interaction networks, gene regulatory networks, drug-disease association networks, drug-target interaction networks, and gene-disease association networks. Each biologically heterogeneous network can only capture a portion of the biological complexity associated with the current biological entity. By integrating multiple types of related biological data, a more comprehensive understanding of the underlying biological processes can be achieved. Specifically, compared to using only a single type of biological network data, integrating data from multiple biological networks can improve the accuracy of outcome predictions. However, because biologically heterogeneous network data are heterogeneous and come in different types and formats, they are difficult to integrate.

[0005] Summary of the Invention

[0006] In view of this, the purpose of this application is to propose a multi-source data fusion method, device, electronic device and storage medium to solve the problem of difficulty in fusing biological heterogeneous network data.

[0007] Based on the above objectives, the first aspect of the present application provides a multi-source data fusion method, comprising:

[0008] Acquire biologically heterogeneous network data;

[0009] Constructing sub-hypergraphs according to the modal data in the biological heterogeneous network data, and splicing multiple sub-hypergraphs to obtain a hypergraph;

[0010] Performing representation learning on a pre-built hypergraph neural network model according to the hypergraph to obtain a biological data node representation;

[0011] Determining a prediction score based on the biological data node representation and a preset downstream task;

[0012] Fused data is determined according to the prediction scores and the downstream tasks.

[0013] A second aspect of the present application provides a multi-source data fusion device, comprising:

[0014] The data acquisition module is configured to: acquire biological heterogeneous network data;

[0015] A hypergraph generation module is configured to: construct sub-hypergraphs according to the modal data in the biological heterogeneous network data, and splice multiple sub-hypergraphs to obtain a hypergraph;

[0016] a representation learning module configured to: perform representation learning on a pre-built hypergraph neural network model according to the hypergraph to obtain a biological data node representation;

[0017] A prediction scoring module is configured to: determine a prediction score based on the biological data node representation and a preset downstream task;

[0018] The prediction output module is configured to determine fusion data according to the prediction score and the downstream task.

[0019] The third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the multi-source data fusion method provided in the first aspect of the present application is implemented.

[0020] A fourth aspect of the present application provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the multi-source data fusion method provided in the first aspect of the present application.

[0021] A fifth aspect of the present application provides a computer program, comprising computer-readable code, which, when executed on an electronic device, causes the electronic device to execute the multi-source data fusion method provided in the first aspect of the present application.

[0022] As can be seen from the above description, the multi-source data fusion method, device, electronic device and storage medium provided by the present application can convert different types of biological heterogeneous network data into sub-hypergraphs respectively, and splice multiple sub-hypergraphs to obtain a hypergraph, fuse different biological heterogeneous network data into the same hypergraph, and use the hypergraph to represent the complex association relationship between biological heterogeneous network data. Then, the pre-constructed hypergraph neural network model is represented and learned according to the hypergraph to obtain the biological data node representation, and the prediction score corresponding to the preset downstream task is determined according to the biological heterogeneous network data and the biological data node representation. Finally, the fused data is determined according to the prediction score and the downstream task. By introducing the prediction score, the present application can not only improve the accuracy of the data fusion process, but also screen out data that meets the requirements of the downstream task through the prediction score for data fusion, thereby improving the performance index of the fused data.

[0023] The above description is only an overview of the technical solution of the present disclosure. In order to more clearly understand the technical means of the present disclosure, 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 disclosure more obvious and easy to understand, the specific implementation methods of the present disclosure are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in this application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are merely embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0025] FIG1 is a schematic diagram of a biological heterogeneous network according to an embodiment of the present application;

[0026] FIG2 is a flow chart of a multi-source data fusion method according to an embodiment of the present application;

[0027] FIG3 is a flowchart of constructing a hypergraph according to an embodiment of the present application;

[0028] FIG4 is a schematic diagram of splicing sub-hypergraphs according to an embodiment of the present application;

[0029] FIG5 is a process for determining a biological data node representation according to an embodiment of the present application;

[0030] FIG6 is a schematic diagram of a biological heterogeneous network data node representation learning process diagram according to an embodiment of the present application;

[0031] FIG7 is a schematic diagram of a hypergraph neural network node representation learning process according to an embodiment of the present application;

[0032] FIG8 is a schematic diagram of another hypergraph neural network node representation learning process according to an embodiment of the present application;

[0033] FIG9 is a schematic structural diagram of a multi-source data fusion device according to an embodiment of the present application;

[0034] FIG10 is a schematic structural diagram of an electronic device according to an embodiment of the present application;

[0035] FIG11 schematically shows a block diagram of another electronic device for executing the method according to the present disclosure; and

[0036] FIG12 schematically shows a storage unit for holding or carrying program codes for implementing the method according to the present disclosure. Specific embodiments

[0037] In order to make the objectives, technical solutions and advantages of this application more clear, this application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0038] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should have the usual meanings understood by people with ordinary skills in the field to which this application belongs. The "first", "second" and similar words used in the embodiments of the present application do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0039] It should be understood herein that any number of elements in the drawings is for illustration only and not for limitation, and any naming is only for distinction and does not have any limiting meaning.

[0040] Based on the description of the above background technology, the following situations also exist in the related art:

[0041] Biological network analysis focuses on studying protein-protein networks, cell signaling and gene regulatory networks, and the interconnected structures and systemic functions of biological system components. However, traditional biological experiments cannot meet the analytical requirements of these biological networks. Therefore, computer-assisted methods can be used to analyze and study them at different levels and discover important information. For example, through text mining and machine learning, possible binary relationships can be discovered from massive amounts of data, thereby constructing biological networks. Network analysis methods such as graph clustering, node classification, and edge prediction can be proposed to study biological problems. However, each biological network can only capture a portion of the biological complexity relevant to the current biological entity. By integrating multiple types of related biological data, a more comprehensive understanding of the underlying biological processes can be achieved. Compared with using only a single type of biological network data, integrating data from multiple biological networks can improve the accuracy of prediction results because there are many known and unknown interactions and complementary information in the diverse biological data. However, biological network data in related technologies are mostly heterogeneous, with different types and formats, and are therefore difficult to fuse. Graph neural networks are usually used in related technologies to fuse different types of biological heterogeneous network data. However, for conventional graph neural networks, an edge in the graph can only connect two vertices, resulting in the inability of the fused data to represent the complexity of biological data, resulting in poor performance indicators of the fused data.

[0042] For example, for biological heterogeneous network data as shown in Figure 1, including gene regulatory networks, drug-disease association networks, drug-target interaction networks, and gene-disease association networks, targets, diseases, drugs, and genes can all serve as graph vertices. Due to the complexity of biological heterogeneous network data, there are cases where a vertex is connected to multiple vertices. For conventional graph neural networks, an edge in the graph can only connect two vertices and can only be used to describe the pairwise relationship between any two vertices. Therefore, it can be seen that the correlation between vertices in biological heterogeneous network data is more complex than the pairwise relationship, making it difficult to model using a graph structure. This makes traditional graph structures limited in expressing data correlations, restricting the application of graph convolutional neural networks. As a result, the data fusion effect is poor, and the performance indicators of the fused data are difficult to meet the application requirements.

[0043] The multi-source data fusion method, device, electronic device and storage medium provided in the embodiments of the present application can convert different types of biological heterogeneous network data into sub-hypergraphs respectively, and splice multiple sub-hypergraphs to obtain a hypergraph, fuse different biological heterogeneous network data into the same hypergraph, and use the hypergraph to represent the complex association relationship between biological heterogeneous network data. Then, the pre-constructed hypergraph neural network model is represented and learned according to the hypergraph to obtain the biological data node representation, and the prediction score corresponding to the preset downstream task is determined according to the biological heterogeneous network data and the biological data node representation. Finally, the fused data is determined according to the prediction score and the downstream task. By introducing the prediction score, the present application can not only improve the accuracy of the data fusion process, but also screen out the data that meets the requirements of the downstream task through the prediction score for data fusion, thereby improving the performance index of the fused data. This will be explained in conjunction with the accompanying drawings and embodiments.

[0044] In some embodiments of the present application, as shown in FIG2 , a multi-source data fusion method includes:

[0045] Step 201: Obtain biological heterogeneous network data.

[0046] In the embodiments of the present application, illustratively, taking the biological heterogeneous network data shown in Figure 1 as an example, the biological heterogeneous network data is a multi-source data, including a protein interaction network, a gene regulatory network, a drug-disease association network, a drug-target interaction network, and a gene-disease association network, wherein the protein interaction network is a network used to describe the interaction between different protein structures (including protein targets), and the gene regulatory network is a network used to describe the interaction between different genes; the drug-disease association network is a network used to represent the association relationship between drugs and diseases; the drug-target interaction network is a network used to represent the association relationship between drugs and targets; and the gene-disease association network is a network used to represent the association relationship between genes and diseases.

[0047] Biological entity nodes (including genes, proteins or targets, drugs, and diseases) in biological heterogeneous network data are connected by complex associations or interactions. For example, nodes with similar targets have interactions, sequence similarity, and structural similarity relationships; nodes with similar drugs also have interactions and structural similarity relationships; nodes with similar diseases also have similarity relationships; nodes with similar genes have interactions; targets and drugs have interactions; and there are associations between drugs and diseases, and between diseases and genes. Therefore, by aggregating and integrating different biological entity nodes, we can obtain biological heterogeneous network data.

[0048] Although it is possible to connect two biological entity nodes based on the associations or interactions between known targets, proteins, diseases, drugs, genes, and other nodes, the number of associations or interactions between known nodes is far less than the number of associations or interactions between unknown or unobserved nodes. Therefore, further fusion processing of multi-source data in biological heterogeneous network data is necessary.

[0049] Step 202: construct sub-hypergraphs according to the modal data in the biological heterogeneous network data, and splice multiple sub-hypergraphs to obtain a hypergraph.

[0050] In an embodiment of the present application, each data source of biological heterogeneous network data corresponds to a modal data, that is, the protein interaction network, the gene regulatory network, the drug-disease association network, the drug-target interaction network, and the gene-disease association network each correspond to a modal data.

[0051] For example, taking the construction of a target sub-hypergraph based on drug-target interaction relationships as an example, the targets and drugs in the drug-target interaction network are represented by vertices. The interaction between a specific drug and a certain number of targets can be modeled through hyperedges to obtain a hyperedge, and all targets interacting with the same drug are connected by this hyperedge. Therefore, all target vertices connected by a hyperedge can be regarded as a set, and all vertices and hyperedges constitute a sub-hypergraph corresponding to the drug-target interaction network. Unlike graph edges in heterogeneous biological graphs that represent second-order pairwise correlations (correlations determined using graph neural networks), hyperedges in heterogeneous biological hypergraphs replace the modeling and analysis of high-order multi-edge (i.e., many-to-many) correlations between targets, which can clearly represent the association or interaction relationship between different vertices. The matrix representation of each sub-hypergraph is: the number of rows in the matrix is ​​the number of targets, and the number of columns in the matrix is ​​the number of hyperedges. Similarly, multiple sub-hypergraphs based on modal data can be obtained, and then these sub-hypergraphs are spliced ​​into a large hypergraph.

[0052] Step 203: Perform representation learning on the pre-built hypergraph neural network model based on the hypergraph to obtain biological data node representation.

[0053] In an embodiment of the present application, a matrix representation corresponding to a hypergraph (an adjacency matrix and initial node features in biological heterogeneous network data) is input into a pre-built hypergraph neural network model, where the initial node features are the associations or interactions between known vertices across different data sources. The hypergraph neural network layer can perform node-edge-node transformations, and this transformation representation learning method can better refine the features, allowing the hypergraph neural network to effectively extract high-order correlations on the hypergraph and output the biological data node representations obtained after each transformation.

[0054] Step 204: Determine a prediction score based on the biological data node representation and a preset downstream task.

[0055] In an embodiment of the present application, after obtaining the biological data node representation representing the high-order features of the node, it is necessary to reconstruct the association relationship or interaction relationship between the biological entity nodes (mainly the association relationship or interaction relationship between unknown or unobserved nodes) according to the specific downstream tasks to achieve specific downstream tasks. The downstream tasks include: prediction of interactions between proteins (targets), prediction of interactions between targets and drugs, prediction of association relationships between diseases and genes, prediction of interaction relationships between drugs, prediction of interaction relationships between genes, etc. The specific downstream tasks need to be determined according to the actual situation, and are not specifically limited here. In an embodiment of the present application, a prediction score is used to represent the reconstruction of the association relationship or interaction relationship between biological entity nodes, wherein the prediction score is the correlation prediction score between biological entity nodes. The higher the prediction score, the stronger the association relationship or interaction relationship between the reconstructed biological entity nodes, the more accurate the prediction, and the higher the degree of completion of the downstream task.

[0056] Step 205: Determine fusion data based on the prediction scores and downstream tasks.

[0057] In an embodiment of the present application, it is necessary to construct a loss function based on the prediction score, and train the network structure parameters of the hypergraph neural network for downstream tasks. Each round of training will obtain a new biological data node representation and a new prediction score, and a new loss function value will be obtained until the loss function converges and the training ends. The hypergraph neural network outputs the prediction result for the downstream task, which includes the biological data node representation obtained in the last round of training and the prediction score between the biological entity nodes. In the process of obtaining the prediction result, the present application repeatedly reconstructs the association relationship or interaction relationship between the biological entity nodes. A large number of hyperedges have been constructed between the nodes in different data sources, completing the data fusion and obtaining fused data.

[0058] In summary, the multi-source data fusion method provided by the embodiment of the present application divides multi-source biological heterogeneous network data into different modal data, constructs sub-hypergraphs respectively, and splices them into a hypergraph. Each vertex in the hypergraph structure generated after splicing represents an entity in biology (gene, RNA, protein and disease, etc.). The original biological data nodes in the hypergraph structure are represented as high-dimensional entity attribute information (gene expression information, methylation data and protein expression spectrum data, etc.). Subsequently, the pre-constructed hypergraph neural network model is represented and learned according to the hypergraph to obtain a biological data node representation. According to the biological data node representation, unknown or unobserved associations or interaction relationships can be predicted to realize the construction of correlation between biological nodes. The correlation enables data from different sources to be fused together, and the correlation is scored to obtain a prediction score to judge the strength of the correlation. Finally, the hypergraph neural network model is repeatedly trained according to the specific downstream tasks. When the loss function converges, a satisfactory prediction result is obtained to realize the fusion of multi-source data. By introducing prediction scores, this application can not only improve the accuracy of the data fusion process, but also screen out data that meets the requirements of downstream tasks through prediction scores for data fusion, thereby improving the performance indicators of the fused data.

[0059] In some embodiments of the present application, as shown in FIG3 , sub-hypergraphs are constructed according to different modalities in biological heterogeneous network data, including:

[0060] Step 301: Group the nodes in the biological heterogeneous network data according to the modality type to obtain multiple modality data.

[0061] In an embodiment of the present application, biological heterogeneous network data includes multiple sources such as protein interaction networks, gene regulatory networks, drug-disease association networks, drug-target interaction networks, and gene-disease association networks. The network data from each source corresponds to a type of modal data. Therefore, when constructing a sub-hypergraph, it is necessary to construct homologous data. Therefore, it is necessary to group the nodes in the biological heterogeneous network data according to the modality type to obtain multiple modal data. Each modal data includes all information about the corresponding network data, including: the number of nodes, known inter-node associations, etc. For example, if you want to fuse the target-drug interaction relationship and the interaction relationship between targets, then the protein interaction network and the drug-target interaction network each correspond to a modal data.

[0062] Step 302: Use the middle nodes of the modal data as vertices of the sub-hypergraph to obtain a vertex set.

[0063] In an embodiment of the present application, all nodes in the same modal data are placed in the same set to obtain a vertex set corresponding to the modal data, so the number of vertex sets is the same as the number of types of modal data. It should be noted that in order to ensure the efficiency of sub-hypergraph splicing, the number of nodes in each modal data should be consistent, so that the number of vertices in different vertex sets is consistent. Exemplary, taking the modal data corresponding to the protein interaction network and the drug-target interaction network as an example, a first vertex set corresponding to the protein interaction network and a second vertex set corresponding to the drug-target interaction network can be obtained.

[0064] Step 303: Determine hyperedges between vertices in the vertex set according to the modal data to obtain a hyperedge set.

[0065] In the embodiments of the present application, illustratively, taking the modal data corresponding to the protein interaction network and the drug-target interaction network as an example, the known inter-node association relationships in the first vertex set can be determined based on the modal data corresponding to the protein interaction network, and then hyperedges are constructed to obtain a first hyperedge set; the known inter-node association relationships in the second vertex set can be determined based on the modal data corresponding to the drug-target interaction network, and then hyperedges are constructed to obtain a second hyperedge set.

[0066] Step 304: construct sub-hypergraphs according to the vertex set and hyperedge set of each modal data to obtain multiple sub-hypergraphs.

[0067] In an embodiment of the present application, as shown in FIG4 , a sub-hypergraph 1 can be constructed based on the first vertex set and the first hyperedge set, and a sub-hypergraph N (in this case, N=2) can be constructed based on the second vertex set and the second hyperedge set. In the hypergraph 1, n1 to n8 represent vertices in the first vertex set, and e1 to e3 represent hyperedges in the first hyperedge set. In the hypergraph N, n1 to n8 represent vertices in the second vertex set, and e1 to e3 represent hyperedges in the first hyperedge set. m-1 to e m Respectively represent the hyperedges in the second hyperedge set. Among them, the matrix representation of sub-hypergraph 1 is to represent hypergraph H1, and the matrix representation of sub-hypergraph N is to represent hypergraph H N .

[0068] Step 305: Concatenate multiple sub-hypergraphs to obtain a hypergraph.

[0069] In the embodiment of the present application, as shown in FIG4 , the sub-representation hypergraph corresponding to each sub-hypergraph is determined. Since the number of vertices of sub-hypergraph 1 and sub-hypergraph N is the same, the two sub-representation hypergraphs H1 and H2 can be directly represented. NPerform row alignment and splicing to form an adjacency matrix H, which represents the hypergraph corresponding to the hypergraph. The corresponding connected hypergraph is determined based on the representation hypergraph to obtain the spliced ​​hypergraph. If the number of vertices in the sub-hypergraphs is inconsistent, the maximum number of vertices in the sub-hypergraphs with the largest number of vertices is used as the vertex number of the hypergraph. Weakly connected vertices can also be removed to make the number of rows in different sub-hypergraphs equal. The greater the number of zeros in the row containing the vertex in the representation hypergraph, the weaker the connection.

[0070] In some embodiments of the present application, step 303 includes:

[0071] Step 3031: Take each vertex in the vertex set as a target vertex, and determine the neighboring vertices of the target vertex according to the modal data.

[0072] In an embodiment of the present application, taking the sub-hypergraph 1 in Figure 4 as an example, when vertex n1 is used as the target vertex, it can be determined based on the modal data that there is an association relationship between the target vertex n1 and vertex n6 and vertex n7, so vertex n6 and vertex n7 are neighbor vertices of the target vertex n1; similarly, the neighbor nodes corresponding to each target vertex can be obtained.

[0073] Step 3032: Determine the target hyperedge corresponding to the target vertex based on the target vertex and neighboring vertices.

[0074] In the embodiments of the present application, since there is a known association between the target vertex and its neighboring vertices, a target hyperedge is constructed that connects both the target vertex and the neighboring vertices. Taking sub-hypergraph 1 in Figure 4 as an example, with vertex n1 as the target vertex, and vertices n6 and n7 as the neighboring vertices of target vertex n1, a target hyperedge e2 is constructed.

[0075] Step 3033: De-duplicate multiple target hyperedges to obtain a hyperedge set.

[0076] In the embodiment of the present application, taking sub-hypergraph 1 in FIG4 as an example, when the target vertices are vertex n1, vertex n6, and vertex n7, the target hyperedges constructed are all e2, so it is necessary to perform deduplication processing on the target hyperedges to obtain a hyperedge set. Optionally, after determining the neighbor vertices of the target vertex, the neighbor vertices can be disqualified as target vertices, thereby avoiding duplicate hyperedges in the hyperedge set.

[0077] In some embodiments of the present application, step 3031 includes:

[0078] Step 30311: Determine whether there is an association relationship between the target vertex and any vertex based on the modal data.

[0079] Step 30312: Determine all vertices with associated relationships as neighbor vertices of the target vertex.

[0080] In an embodiment of the present application, taking the sub-hypergraph 1 in Figure 4 as an example, when vertex n1 is used as the target vertex, each vertex is judged in turn based on the modal data to determine whether there is an association relationship with the target vertex, until all nodes in the vertex set are traversed, and it is determined that there is an association relationship between the target vertex n1 and vertex n6 and vertex n7, so vertex n6 and vertex n7 are determined as neighbor vertices of the target vertex n1.

[0081] In some embodiments of the present application, step 305 includes:

[0082] Step 3051: Determine the sub-representation hypergraph of each sub-hypergraph, where the sub-representation hypergraph is a two-dimensional matrix, the number of rows of the two-dimensional matrix is ​​equal to the number of vertices in the vertex set, and the number of columns of the two-dimensional matrix is ​​equal to the number of hyperedges in the hyperedge set.

[0083] In an embodiment of the present application, taking the sub-hypergraph 1 in Figure 4 as an example, the sub-representation hypergraph corresponding to the sub-hypergraph 1 is a two-dimensional matrix H1, each row of the sub-representation hypergraph H1 corresponds to a vertex, and each column of the sub-representation hypergraph H1 corresponds to a hyperedge, so the number of rows of the two-dimensional matrix is ​​equal to the number of vertices in the vertex set, and the number of columns of the two-dimensional matrix is ​​equal to the number of hyperedges in the hyperedge set.

[0084] Step 3052: perform row alignment and splicing on the sub-representation hypergraphs of different sub-hypergraphs to obtain an adjacency matrix.

[0085] In the embodiment of the present application, taking the sub-hypergraph 1 and sub-hypergraph N in FIG4 as an example, since the number of vertices in the sub-hypergraph 1 and the sub-hypergraph N is the same, the sub-hypergraph H1 and the sub-hypergraph H N By performing row alignment and splicing, we can get the adjacency matrix H.

[0086] Step 3053: Connect the vertices according to the adjacency matrix to obtain the hypergraph.

[0087] In the embodiment of the present application, it can be seen that the adjacency matrix H has the same number of rows as the two-dimensional matrix corresponding to the sub-representation hypergraph, but the connection relationship between the vertices is aggregated, so it is necessary to connect the vertices according to the adjacency matrix to obtain the hypergraph. Taking the sub-hypergraph 1 and the sub-representation hypergraph H1 as examples, it can be seen that the first row and the first column of the sub-representation hypergraph H1 are 0, indicating that the vertex n1 is not on the hyperedge e1; the first row and the second column of the sub-representation hypergraph H1 are 1, indicating that the vertex n1 is on the hyperedge e2, so in the sub-representation hypergraph H1, (n, e) = 1 indicates that the vertex is on the hyperedge, and (n, e) = 0 indicates that the vertex is not on the hyperedge.

[0088] Furthermore, we define a hypergraph as G = (V, E, W), where V is the vertex set of the hypergraph, E is the hyperedge set of the hypergraph, and each hyperedge in the hypergraph is assigned a weight by W, which is a diagonal matrix of hyperedge weights. The hypergraph G can be represented by the |V| × |E| adjacency matrix H, and is defined as:

[0089] For vertices Its degree is defined as: d(v)=Σ e∈ε ω(e)h(v, e)

[0090] For a hyperedge e∈ε, its degree is defined as:

[0091] In some embodiments of the present application, as shown in FIG5 and FIG6 , a representation learning is performed on a pre-built hypergraph neural network model according to a hypergraph to obtain a biological data node representation, including:

[0092] Step 501: The adjacency matrix in the hypergraph and the initial node features in the biological heterogeneous network data are input into the hypergraph neural network model as input data.

[0093] In the embodiments of the present application, the initial node feature data in the biological heterogeneous network data is the known association relationship or interaction relationship between nodes in different source data. The initial node feature can be used as the initial representation data for classification prediction by the hypergraph neural network model:

[0094] Among them, X represents the initial node feature, X (0) It represents the biological data node representation obtained by the hypergraph at the 0th layer of the hypergraph convolution layer, that is, the initial representation data.

[0095] Step 502: Perform node feature aggregation and edge feature aggregation on the input data through the hypergraph convolution layer in the hypergraph neural network model, extract high-order correlation features in the hypergraph, and obtain biological data node representation.

[0096] In an embodiment of the present application, the hypergraph convolution layer is updated according to the adjacency matrix to obtain the hypergraph convolution layer f(X, W, Θ):

[0097] Among them, X (l+1) represents the biological data node representation obtained by the hypergraph at the l+1 layer hypergraph convolution layer; X l represents the biological data node representation obtained by the hypergraph at the lth hypergraph convolution layer; D e The diagonal matrix representing the degree of the hyperedges in the hypergraph, D vThe diagonal matrix representing the degrees of vertices in the hypergraph, Dv and De, serves as normalization. σ is a nonlinear activation function. Θ(l) represents the filter matrix of the lth hypergraph convolutional layer. This filter matrix is ​​learnable and is updated at each layer. The biological data node representation output by the final hypergraph convolutional layer is input into the fully connected classification network of the hypergraph neural network model to obtain the prediction results.

[0098] As shown in Figures 7 and 8, the hypergraph convolutional layer of the hypergraph neural network transforms the initial node features into point feature aggregation, edge feature aggregation, and point feature aggregation, extracting high-order correlations within the hypergraph to obtain and output biological data node representations. In Figure 8, N represents the number of nodes, E represents the number of hyperedges, and C1 and C2 represent different feature dimensions.

[0099] In some embodiments of the present application, step 204 includes:

[0100] Step 2041: Determine the prediction target based on the downstream task.

[0101] In an embodiment of the present application, it is necessary to reconstruct the association relationship or interaction relationship between biological entity nodes (mainly the association relationship or interaction relationship between unknown or unobserved nodes) according to the specific downstream tasks to achieve specific downstream tasks. Among them, downstream tasks include: prediction of interactions between proteins (targets), prediction of interactions between targets and drugs, prediction of association relationships between diseases and genes, prediction of interaction relationships between drugs, prediction of interaction relationships between genes, etc. It is necessary to determine specific downstream tasks according to actual conditions, which are not specifically limited here. For example, if the downstream task is prediction of interactions between targets and drugs, the prediction target is to predict the association relationship or interaction relationship between the target and the drug.

[0102] Step 2043: Determine a prediction probability score matrix based on the prediction target and the biological data node representation.

[0103] In the embodiment of the present application, in order to reconstruct the association or interaction relationship between biological entities, the decoder f(H D1 ,H D2 )Build the probability score matrix:

[0104] Where A′∈R u×v Represents the prediction probability score matrix. If the prediction goal is to predict the association or interaction between the target and the drug, then H D1 represents the target feature set in the biological data node representation, H D2Denote the set of drug features in the biological data node representation. There are u target features in the target feature set and v drug features in the drug set. Then, the prediction score between target i and drug j is determined by the A' ij entry in the prediction probability score matrix.

[0105] Step 2044: Determine the prediction scores between nodes according to the prediction probability score matrix. Among them, the higher the prediction score, the higher the probability that there is an association relationship between nodes.

[0106] In an embodiment of the present application, the prediction score between biological entity node i and biological entity node j is determined according to the A' ij entry in the prediction probability score matrix. The prediction score is a score for measuring the relevance between biological entity nodes. The higher the prediction score, the higher the probability that there is an association relationship or an interaction relationship between the reconstructed biological entity nodes, the more accurate the prediction, and the higher the completion degree of the downstream task.

[0107] In some embodiments of the present application, Step 205 includes:

[0108] Step 2051: Construct a loss function according to the prediction scores.

[0109] In an embodiment of the present application, constructing a loss function according to the prediction scores includes:

[0110] Determine the known association set and the unknown association set according to the biological heterogeneous network data;

[0111] Determine the balance factor according to the known association set and the unknown association set;

[0112] Construct a loss function according to the balance factor and the prediction scores.

[0113] According to the prediction scores, learn the parameters between two biological entity nodes by minimizing the weighted binary cross-entropy loss, and obtain the loss function as shown below:

[0114] where, (i, j) represents the biological entity nodes d i and d j pair corresponding to the prediction task, S + represents the set of all known association pairs or interactions, and S - represents the set of all unknown or unobserved association pairs or interactions. The balance factor is used to reduce the impact of data imbalance, where |S + | and |S - | are the logarithms of S + and S - respectively.

[0115] Step 2052: Train the hypergraph neural network model based on the prediction score and downstream tasks until the loss function converges to obtain the prediction result.

[0116] In an embodiment of the present application, a hypergraph neural network model is trained according to the prediction score and downstream tasks until the loss function converges to obtain a prediction result, including:

[0117] Update the network structure parameters of the hypergraph neural network model according to the prediction score and downstream tasks to obtain a new prediction score;

[0118] Calculate the loss function value based on the new prediction score, and determine whether the loss function converges based on the loss function value;

[0119] In response to the loss function converging, the prediction result is output.

[0120] Among them, the network structure parameters of the hypergraph neural network model are updated according to the prediction score and downstream tasks, and the hypergraph neural network model is continuously optimized so that the prediction results obtained after each conversion are more accurate and the accuracy of data fusion is improved. According to the biological data node representation, unknown or unobserved associations or interaction relationships can be predicted to realize the construction of the correlation between biological nodes. This correlation enables the data from different sources to be fused together, and the correlation is scored to obtain a prediction score to judge the strength of the correlation. Finally, the hypergraph neural network model is repeatedly trained according to the specific downstream task. When the loss function converges or the training rounds reach the preset maximum training rounds, the training is terminated to obtain a satisfactory prediction result to realize the fusion of multi-source data. By introducing the prediction score, the present application can not only improve the accuracy of the data fusion process, but also filter out the data that meets the requirements of the downstream task through the prediction score for data fusion, thereby improving the performance index of the fused data. The prediction result mainly includes the probability (or probability score) of the existence of an association relationship between unknown or unobserved nodes, which can be represented in the form of a prediction probability score matrix, as well as the biological data node representation of each node. The output biological data node representation can be used as input data for the next round of training.

[0121] Step 2053: Integrate the prediction results and the biological heterogeneous network data to obtain fused data.

[0122] In an embodiment of the present application, each node is connected according to the predicted probability score matrix in the prediction result, wherein different prediction score ranges can be represented by different connecting lines, or their probabilities or predicted scores can be labeled on the connection path. Due to the reconstruction of the inter-node association relationship in the prediction result, the fusion of different biological data has been completed. However, the reconstruction of the association relationship may cause the loss of the original known inter-node association relationship or interaction relationship. Therefore, the prediction results are proofread using biological heterogeneous network data to improve the data integrity of the fused data.

[0123] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied in a distributed scenario and performed by multiple devices working together. In such a distributed scenario, one of the multiple devices may only perform one or more steps of the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the method.

[0124] It should be noted that the above description is limited to some embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in an order different from that described in the above embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0125] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a multi-source data fusion device.

[0126] Referring to FIG9 , the multi-source data fusion device includes:

[0127] The data acquisition module 10 is configured to: acquire biological heterogeneous network data;

[0128] The hypergraph generation module 20 is configured to: construct sub-hypergraphs according to the modal data in the biological heterogeneous network data, and splice multiple sub-hypergraphs to obtain a hypergraph;

[0129] The representation learning module 30 is configured to: perform representation learning on a pre-built hypergraph neural network model according to the hypergraph to obtain a biological data node representation;

[0130] The prediction scoring module 40 is configured to: determine a prediction score based on the biological data node representation and a preset downstream task;

[0131] The prediction output module 50 is configured to determine fusion data according to the prediction score and downstream tasks.

[0132] For the convenience of description, the above devices are described as being divided into various modules according to their functions. Of course, when implementing this application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0133] The apparatus of the above embodiment is used to implement the corresponding multi-source data fusion method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0134] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor, wherein when the processor executes the program, the multi-source data fusion method described in any of the above-mentioned embodiments is implemented.

[0135] FIG10 shows a more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment. The device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are communicatively connected to each other within the device via the bus 1050.

[0136] The processor 1010 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0137] The memory 1020 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 1020 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1020 and is called and executed by the processor 1010.

[0138] The input / output interface 1030 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components within the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0139] The communication interface 1040 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0140] The bus 1050 comprises a path for transmitting information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).

[0141] It should be noted that although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040, and the bus 1050, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of this specification, and does not necessarily include all the components shown in the figure.

[0142] The electronic device of the above embodiment is used to implement the corresponding multi-source data fusion method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0143] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the multi-source data fusion method described in any of the above-mentioned embodiments.

[0144] The computer-readable media of this embodiment include permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media 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 memory technology, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0145] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the multi-source data fusion method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0146] The various component embodiments of the present disclosure can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It will be appreciated by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the electronic device according to the embodiments of the present disclosure. The present disclosure can also be implemented as a device or apparatus program (e.g., computer program and computer program product) for executing a part or all of the methods described herein. Such a program implementing the present disclosure can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0147] For example, FIG11 illustrates an electronic device that can implement the methods disclosed herein. The electronic device typically includes a processor 1010 and a computer program product or computer-readable medium in the form of a memory 1020. Memory 1020 can be an electronic memory such as flash memory, EEPROM (Electrically Erasable Programmable Read-Only Memory), EPROM, a hard disk, or ROM. Memory 1020 has storage space 1032 for program code 1031 for executing any of the method steps described above. For example, storage space 1032 for program code can include individual program codes 1031 for implementing various steps in the above method. These program codes can be read from or written to one or more computer program products. These computer program products include program code carriers such as hard disks, compact disks (CDs), memory cards, or floppy disks. Such computer program products are typically portable or fixed storage units, as described with reference to FIG12. The storage unit can have storage segments, storage space, and the like arranged similarly to memory 1020 in the electronic device of FIG11. The program code can, for example, be compressed in a suitable form. Typically, the storage unit includes computer-readable codes 1031 ′, ie, codes that can be read by a processor such as 1010 . When these codes are executed by an electronic device, they cause the electronic device to perform the steps of the method described above.

[0148] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a computer program, including computer-readable code, which, when executed on an electronic device, causes the electronic device to execute the multi-source data fusion method described in any of the above-mentioned embodiments.

[0149] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. Within the scope of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0150] In addition, for simplicity of description and discussion, and in order not to make the embodiment of the application difficult to understand, the known power supply / ground connection with integrated circuit (IC) chip and other components may or may not be shown in the accompanying drawings provided. In addition, the device can be shown in the form of a block diagram to avoid making the embodiment of the application difficult to understand, and this also takes into account the following fact, that is, the details of the embodiment of these block diagram devices are highly dependent on the platform to be implemented in the embodiment of the application (that is, these details should be fully within the scope of understanding of those skilled in the art). When specific details (for example, circuit) are set forth to describe exemplary embodiments of the application, it will be apparent to those skilled in the art that the embodiment of the application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0151] Although the present invention has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of these embodiments will be apparent to those skilled in the art based on the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may utilize the embodiments discussed.

[0152] The embodiments of the present application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of this application.

Claims

1. A multi-source data fusion method, characterized in that: include: Acquiring biologically heterogeneous network data; Constructing sub-hypergraphs respectively according to the modal data in the biological heterogeneous network data, and splicing a plurality of the sub-hypergraphs to obtain a hypergraph; Performing representation learning on a pre-constructed hypergraph neural network model according to the hypergraph to obtain a biological data node representation; Determining a prediction score based on the biological data node representation and a preset downstream task; as well as The fused data is determined according to the prediction score and the preset downstream task.

2. The method according to claim 1, wherein: The step of respectively constructing the sub-hypergraphs according to the modal data in the biological heterogeneous network data comprises: Grouping the nodes in the biological heterogeneous network data according to the modality types to obtain a plurality of modality data; Taking the middle nodes of the plurality of modal data as vertices of the sub-hypergraph to obtain a vertex set; Determine the hyperedges between the vertices in the vertex set according to the multiple modal data to obtain a hyperedge set; Constructing the sub-hypergraphs according to the vertex set and the hyperedge set of each modality data, respectively, to obtain a plurality of the sub-hypergraphs; and The multiple sub-hypergraphs are concatenated to obtain the hypergraph.

3. The method according to claim 2, wherein: The step of determining the hyperedges between the vertices in the vertex set according to the plurality of modal data to obtain the hyperedge set comprises: Taking each vertex in the vertex set as a target vertex, and determining neighbor vertices of the target vertex according to the multiple modal data; Determine a target hyperedge corresponding to the target vertex according to the target vertex and the neighbor vertices; and Deduplication processing is performed on the multiple target hyperedges to obtain the hyperedge set.

4. The method according to claim 3, wherein: The step of determining the neighboring vertices of the target vertex comprises: Determining whether there is an association relationship between the target vertex and any vertex according to the multiple modal data; and All vertices that have an associated relationship are determined as the neighbor vertices of the target vertex.

5. The method according to claim 2, wherein: The step of concatenating a plurality of said sub-hypergraphs to obtain said hypergraph comprises: Determine a sub-representation hypergraph of each of the sub-hypergraphs, wherein the sub-representation hypergraph is a two-dimensional matrix, the number of rows of the two-dimensional matrix is ​​equal to the number of vertices in the vertex set, and the number of columns of the two-dimensional matrix is ​​equal to the number of hyperedges in the hyperedge set; Performing row alignment and splicing on the sub-representation hypergraphs of different sub-hypergraphs to obtain an adjacency matrix; and Vertices are connected according to the adjacency matrix to obtain the hypergraph.

6. The method according to claim 1, wherein: The step of performing representation learning on the pre-built hypergraph neural network model according to the hypergraph to obtain the biological data node representation comprises: Inputting the adjacency matrix of the hypergraph and the initial node features in the biological heterogeneous network data as input data into the pre-built hypergraph neural network model; and The input data is subjected to node feature aggregation and edge feature aggregation through the hypergraph convolution layer in the pre-constructed hypergraph neural network model, and high-order correlation features in the hypergraph are extracted to obtain the biological data node representation.

7. The method according to claim 6, wherein: The initial node feature is the association relationship or interaction relationship between known vertices between different data sources.

8. The method according to claim 1, wherein: The step of determining the prediction score according to the biological data node representation and the preset downstream task comprises: Determining a prediction target according to the preset downstream task; Determining a prediction probability score matrix according to the prediction target and the biological data node representation; and The predicted scores between nodes are determined according to the predicted probability score matrix, wherein the higher the predicted scores are, the higher the probability that there is an association relationship between the nodes is.

9. The method according to claim 1, wherein: The step of determining the fusion data according to the prediction score and the preset downstream task includes: Constructing the loss function according to the prediction score; Training the pre-built hypergraph neural network model according to the prediction score and the preset downstream task until the loss function converges to obtain a prediction result; and The prediction result and the biological heterogeneous network data are integrated to obtain the fused data.

10. The method according to claim 9, wherein: The step of constructing the loss function according to the prediction score comprises: Determining a known association set and an unknown association set based on the biological heterogeneous network data; Determining a balance factor according to the known association set and the unknown association set; and The loss function is constructed according to the balance factor and the prediction score.

11. The method according to claim 9, wherein: The step of training the pre-built hypergraph neural network model according to the prediction score and the preset downstream task until the loss function converges to obtain the prediction result comprises: Update the network structure parameters of the pre-built hypergraph neural network model according to the prediction score and the preset downstream task to obtain a new prediction score; Calculating the loss function value according to the new prediction score, and determining whether the loss function converges according to the loss function value; and In response to the loss function converging, the prediction result is output.

12. The method according to claim 1, wherein: The preset downstream tasks include: one or more of the following: prediction of interaction between targets, prediction of interaction between targets and drugs, prediction of association between diseases and genes, prediction of interaction between drugs, and prediction of interaction between genes.

13. A multi-source data fusion device, characterized in that: include: The data acquisition module is configured to: acquire biological heterogeneous network data; A hypergraph generation module is configured to: construct sub-hypergraphs according to the modal data in the biological heterogeneous network data, and splice a plurality of the sub-hypergraphs to obtain a hypergraph; A representation learning module is configured to: perform representation learning on a pre-built hypergraph neural network model according to the hypergraph to obtain a biological data node representation; A prediction scoring module is configured to: determine a prediction score based on the biological data node representation and a preset downstream task; as well as The prediction output module is configured to determine fusion data according to the prediction score and the preset downstream task.

14. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the multi-source data fusion method according to any one of claims 1 to 12 when executing the program. 15 . A non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the multi-source data fusion method according to any one of claims 1 to 12.

16. A computer program, comprising a computer readable code, which, when executed on an electronic device, causes the electronic device to execute the multi-source data fusion method according to any one of claims 1 to 12.

Images