Disease similarity prediction method and system based on multivariate data fusion

Abstract

The invention relates to a disease similarity prediction method and system based on multivariate data fusion, and the method comprises the steps: collecting gene network data, miRNA network data, a disease-gene incidence matrix and a disease-miRNA incidence matrix, and forming a unified heterogeneous biological network; extracting and generating a multi-modal feature set through the graph convolutional network and the improved graph attention network; the independent weight of each feature in the multi-modal feature set is calculated through a multi-layer perceptron, gene view angle embedding and miRNA view angle embedding of the disease are generated in combination with the incidence matrix and serve as model input training, and a lightweight bilinear tower model is obtained; model parameters are optimized through an alternate training strategy and a regularization mechanism inspired by ReconBoost, an optimized lightweight bilinear tower model is obtained, and a similarity prediction result of the target disease pair is obtained. The defects of multivariate data integration, sparse data processing and prediction stability in the prior art are effectively overcome.

Description

Technical Field

[0001] The present invention belongs to the technical field of biomedical data analysis, and in particular relates to a disease similarity prediction method and system based on multivariate data fusion. Background Art

[0002] Disease similarity prediction is an important technology in biomedical research. By analyzing the molecular and phenotypic associations between diseases, it reveals the underlying pathological mechanisms and provides key support for drug repositioning, disease classification optimization and personalized treatment. Its core value lies in identifying diseases with similar molecular mechanisms or clinical manifestations, thereby providing a basis for drug repositioning. Drug repositioning accelerates the drug development process by exploring new indications for approved drugs. Compared with traditional new drug development, the use of safety and pharmacokinetic data of existing drugs can significantly shorten the development cycle and reduce R&D costs and risks. This strategy has shown great potential in therapeutic research in fields such as cancer, neurodegenerative diseases, cardiovascular diseases and infectious diseases, bringing important commercial value and social benefits to the pharmaceutical industry and medical research. For example, by identifying the molecular similarities between diabetes and cardiovascular diseases, researchers can reapply approved antidiabetic drugs to the treatment of cardiovascular diseases, significantly improving drug development efficiency.

[0003] In recent years, the rapid development of high-throughput sequencing technology and bioinformatics has enabled researchers to utilize multidimensional biological data (e.g., gene expression, miRNA regulatory networks, and disease-molecule association data) to construct complex biological networks for assessing disease similarity. In particular, the rise of deep learning technologies, particularly graph neural networks, has provided new technical means for disease similarity prediction. Leveraging their powerful graph-structured data processing capabilities, graph neural networks efficiently model the complex relationships between diseases, genes, and miRNAs. For example, graph convolutional networks aggregate node neighborhood information through multi-layer convolution operations, enabling high-dimensional topological representation of gene networks. Improved graph attention networks dynamically capture local structural differences in miRNA networks by adaptively assigning node weights. These methods have demonstrated significant advantages in mining disease-molecule associations and have been widely applied in disease similarity prediction and drug repositioning research.

[0004] However, existing disease similarity prediction methods still face numerous challenges in practical application. Multivariate biological data, such as gene networks and miRNA networks, are high-dimensional, heterogeneous, and sparse. Existing methods struggle to effectively integrate their complementary information. For example, gene network data is massive but contains a lot of redundant information, while miRNA network data is sparse and has incomplete associations, resulting in inefficient feature extraction. Furthermore, existing methods often lack adaptive mechanisms to screen key biological features, making it difficult to highlight the most important genes or miRNAs for disease similarity prediction. Traditional methods may treat all genes equally, ignoring the significant contributions of key regulatory genes, thereby reducing prediction accuracy. Furthermore, complex models based on graph neural networks (GNNs) are computationally expensive and require long training times when processing multivariate data, making them difficult to apply in resource-constrained clinical or research settings. Furthermore, when integrating gene and miRNA data, competition or inconsistency between modalities can lead to fluctuations in prediction results. For example, the richness of gene data may dominate the prediction, while the sparsity of miRNA data is neglected, affecting the balance between modalities and the robustness of the model. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a disease similarity prediction method and system based on multivariate data fusion.

[0006] The technical solution adopted in the present invention is:

[0007] In a first aspect, a disease similarity prediction method based on multivariate data fusion is provided, comprising:

[0008] Collect gene network data, miRNA network data, disease-gene association matrix, and disease-miRNA association matrix to form a unified heterogeneous biological network;

[0009] Gene embedding features and miRNA embedding features are extracted from heterogeneous biological networks through graph convolutional networks and improved graph attention networks, respectively, to generate a multimodal feature set.

[0010] A multi-layer perceptron is used to calculate the independent weights of each feature in the multimodal feature set, and combined with the disease-gene association matrix and the disease-miRNA association matrix to generate the gene-perspective embedding and miRNA-perspective embedding of the disease;

[0011] Gene-perspective embedding and miRNA-perspective embedding are used as model inputs to train a lightweight bilinear tower model.

[0012] The model parameters of the lightweight bilinear tower model are optimized through an alternating training strategy and a regularization mechanism inspired by ReconBoost, resulting in an optimized lightweight bilinear tower model.

[0013] The disease similarity prediction of the target disease pairs is performed by optimizing the lightweight bilinear tower model, and the similarity prediction results of the target disease pairs are obtained.

[0014] Furthermore, gene networks Contains a collection of gene nodes and edge sets based on gene co-expression or functional association ; The graph structure of the gene network is ;

[0015] miRNA network Contains miRNA node collection and edge sets based on miRNA similarity or regulatory relationships ; The graph structure of the miRNA network is .

[0016] Furthermore, gene network data, miRNA network data, disease-gene association matrix, and disease-miRNA association matrix are collected to form a unified heterogeneous biological network, including:

[0017] Collecting gene networks Gene network data and miRNA network miRNA network data;

[0018] Obtaining disease-gene association matrix and disease-miRNA association matrix , disease-gene association matrix and disease-miRNA association matrix Contains the association between known diseases and genes and miRNAs respectively;

[0019] Through the disease-gene association matrix and disease-miRNA association matrix , the gene network and miRNA networks The gene nodes and miRNA nodes in the dataset are mapped to disease nodes to form a unified heterogeneous biological network.

[0020] Furthermore, we extract gene embedding features and miRNA embedding features from heterogeneous biological networks through graph convolutional networks and improved graph attention networks, respectively, to generate a multimodal feature set, including:

[0021] Gene Networks for Heterogeneous Biological Networks Based on Graph Convolutional Networks (GCN) Perform feature extraction to obtain gene embedding features ;

[0022] miRNA network for heterogeneous biological networks based on improved graph attention network GATv2 Perform feature extraction to obtain miRNA embedded features ;

[0023] Collective gene embedding features and miRNA embedding features , generating a multimodal feature set.

[0024] Furthermore, we use a multi-layer perceptron to calculate the independent weights of each feature in the multimodal feature set, and combine the disease-gene association matrix and the disease-miRNA association matrix to generate the gene-perspective embedding and miRNA-perspective embedding of the disease, including:

[0025] Apply two layers of multi-layer perceptrons to each feature in the multimodal feature set and calculate independent weights;

[0026] Based on the independent weights and the disease-gene association matrix and the disease-miRNA association matrix, the disease representation calculation formula is substituted to calculate the gene perspective embedding of the disease. and miRNA perspective embedding , the disease characterization calculation formula is:

[0027] ;

[0028] in, Embedding for genetic perspective or miRNA perspective embedding , h is g or m, A represents the disease-gene association matrix or disease-miRNA association matrix, represents independent weights.

[0029] Furthermore, the gene-perspective embedding and miRNA-perspective embedding are used as model inputs to train a lightweight bilinear tower model, including:

[0030] Embedding a genetic perspective and miRNA perspective embedding As model input;

[0031] Disease pairs for known disease i and known disease j , embedded from a genetic perspective and miRNA perspective embedding Extract the representation and construct the gene feature vector and miRNA feature vector;

[0032] The expression of gene feature vector is:

[0033] ;

[0034] The expression of miRNA feature vector is:

[0035] ;

[0036] After dimensionality reduction through linear projection, the gene tower score and miRNA tower score are calculated using the bilinear transformation formula. The bilinear transformation formula is:

[0037] ;

[0038] is a learnable bilinear weight matrix, Score for the gene tower or miRNA tower score ;

[0039] Find the gene tower score and miRNA tower score The average value of is used to obtain the final similarity score;

[0040] According to the model input and the final similarity score value, a lightweight bilinear tower model is trained.

[0041] Furthermore, the model parameters of the lightweight bilinear tower model are optimized through an alternating training strategy and a regularization mechanism inspired by ReconBoost, resulting in an optimized lightweight bilinear tower model, including:

[0042] Through the alternating training strategy, in each round of training, the encoder parameters and bilinear tower parameters of one modality in the gene modality and miRNA modality are fixed, and the encoder parameters and bilinear tower parameters of the other modality are updated. The dynamic loss function is defined as:

[0043] ;

[0044] in, is the cross entropy loss, is the current mode prediction, is the relative entropy KL divergence, is the trade-off coefficient;

[0045] Based on the regularization mechanism inspired by ReconBoost, all encoder parameters and bilinear tower parameters are jointly optimized every preset fixed rounds, and the global loss is defined as:

[0046] ;

[0047] is the weighted average forecast, is the learning rate;

[0048] An optimized lightweight bilinear tower model is obtained based on dynamic loss and global loss.

[0049] Secondly, a disease similarity prediction system based on multivariate data fusion is provided, including:

[0050] Multivariate data fusion module, which is used to collect gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix to form a unified heterogeneous biological network;

[0051] The feature extraction module is used to extract gene embedding features and miRNA embedding features from heterogeneous biological networks through graph convolutional networks and improved graph attention networks, respectively, to generate a multimodal feature set;

[0052] The dynamic weight embedding module is used to calculate the independent weight of each feature in the multimodal feature set using a multi-layer perceptron, and combine the disease-gene association matrix and the disease-miRNA association matrix to generate the gene-perspective embedding and miRNA-perspective embedding of the disease;

[0053] A lightweight bilinear tower prediction module is used to train a lightweight bilinear tower model using gene-perspective embedding and miRNA-perspective embedding as model inputs;

[0054] The alternating training and optimization module is used to optimize the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost to obtain an optimized lightweight bilinear tower model;

[0055] The disease similarity prediction module is used to predict the disease similarity of the target disease pair by optimizing the lightweight bilinear tower model to obtain the similarity prediction results of the target disease pair.

[0056] The beneficial effects achieved by the present invention are:

[0057] Gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix are collected to form a unified heterogeneous biological network; gene embedding features and miRNA embedding features are extracted from the heterogeneous biological network through graph convolutional network and improved graph attention network respectively to generate a multimodal feature set; multi-layer perceptron is used to calculate the independent weights of each feature in the multimodal feature set, and combined with the disease-gene association matrix and disease-miRNA association matrix to generate gene perspective embedding and miRNA perspective embedding of the disease; gene perspective embedding and miRNA perspective embedding are used as model input for training to obtain a lightweight bilinear tower model; the model parameters of the lightweight bilinear tower model are optimized through an alternating training strategy and a regularization mechanism inspired by ReconBoost to obtain an optimized lightweight bilinear tower model; disease similarity prediction of target disease pairs is performed through the optimized lightweight bilinear tower model to obtain the similarity prediction results of the target disease pairs. By deeply integrating gene networks, miRNA networks and disease association data, combined with graph neural networks, dynamic weight embedding, lightweight bilinear tower prediction and alternating training optimization strategies, high-precision, low-computational cost disease similarity prediction is achieved; it effectively makes up for the shortcomings of existing technologies in multivariate data integration, sparse data processing and prediction stability, and provides strong support for pathological mechanism research, disease classification optimization and drug repositioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 Flowchart of the disease similarity prediction method based on multivariate data fusion of the present invention;

[0059] Figure 2 This is a structural diagram of the disease similarity prediction system based on multivariate data fusion of the present invention. DETAILED DESCRIPTION

[0060] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0061] like Figure 1 As shown, an embodiment of the present invention provides a disease similarity prediction method based on multivariate data fusion, comprising:

[0062] 101. Collect gene network data, miRNA network data, disease-gene association matrix, and disease-miRNA association matrix to form a unified heterogeneous biological network;

[0063] Gene Network Contains a collection of gene nodes and edge sets based on gene co-expression or functional association ; The graph structure of the gene network is ;

[0064] miRNA network Contains miRNA node collection and edge sets based on miRNA similarity or regulatory relationships ; The graph structure of the miRNA network is ;

[0065] Collecting gene networks Gene network data and miRNA network miRNA network data;

[0066] Obtaining disease-gene association matrix and disease-miRNA association matrix , the elements in the matrix are 1 when the association exists, otherwise they are 0;

[0067] Disease-gene association matrix and disease-miRNA association matrix Contains the association between known diseases and genes and miRNAs respectively;

[0068] For example, the gene network data contains 17,247 gene nodes and 371,501 edges, which are derived from the HumanNet database; the miRNA network data contains 4,798 miRNA nodes and 126,264 edges, which are derived from the RNADisease database; and the disease association data contains 1,030,082 disease-gene associations and 355,196 disease-miRNA associations, which are derived from the DisGeNET database and the RNADisease database; the training set is pre-set to contain 6,182 disease pairs (of which 3,091 pairs are positive samples), and the test set contains 1,330 disease pairs (of which 665 pairs are positive samples); the gene network and miRNA networks It is stored in the form of a sparse adjacency matrix, and the edge weights are normalized based on the biological association strength (such as co-expression coefficient or regulatory strength). The association matrix is ​​binarized to reduce computational complexity.

[0069] Through the disease-gene association matrix and disease-miRNA association matrix , the gene network and miRNA networks The gene nodes and miRNA nodes in the dataset are mapped to disease nodes to form a unified heterogeneous biological network.

[0070] 102, using graph convolutional networks and improved graph attention networks to extract gene embedding features and miRNA embedding features from heterogeneous biological networks, generating a multimodal feature set;

[0071] Genetic Networks of Heterogeneous Biological Networks Based on Graph Convolutional Network (GCN) Perform feature extraction to obtain gene embedding features ; The gene network node feature update formula is:

[0072] ;

[0073] in, is the feature matrix of the gene nodes in the first layer, is the normalized adjacency matrix, is the adjacency matrix, is the degree matrix, is the learnable weight matrix, As the activation function, the first layer of GCN maps the initial features (based on gene expression levels) to a 128-dimensional hidden layer, and the second layer further aggregates neighborhood information to generate gene embedding features. (dimension 128), dropout rate 0.4, weight matrix randomly initialized, optimized by backpropagation during training;

[0074] miRNA network for heterogeneous biological networks based on improved graph attention network GATv2 Perform feature extraction to obtain miRNA embedded features ; Node features are updated through a multi-head attention mechanism, and the attention coefficient calculation formula is:

[0075] ;

[0076] in, and is the node feature, is the weight matrix, is the learnable attention vector, Represents the splicing operation. GATv2 uses a two-head attention mechanism to generate miRNA embedding features. The number of attention heads is 2, the dropout rate is 0.4, the weight matrix and attention vector are randomly initialized, and the training is optimized by backpropagation.

[0077] Collective gene embedding features and miRNA embedding features , generate a multimodal feature set;

[0078] In this step, gene node features are based on expression levels (extracted from RNA sequencing data), and miRNA node features are based on regulatory relationships (extracted from RNA disease data). The initial dimension is determined by the dataset (512 dimensions by default). Both GCN and GATv2 apply dropout (default 0.4) and L2 regularization (coefficient 0.0001) to enhance model generalization. Sparse matrix operations are implemented using PyTorch Geometric to reduce memory usage and accelerate feature extraction.

[0079] 103. Use a multi-layer perceptron to calculate the independent weights of each feature in the multimodal feature set, combine the disease-gene association matrix and the disease-miRNA association matrix, and generate the gene-perspective embedding and miRNA-perspective embedding of the disease;

[0080] For each feature in the multimodal feature set Apply two layers of multilayer perceptron (MLP) respectively to calculate the independent weights:

[0081] ;

[0082] MLP consists of two layers of linear transformation. The first layer reduces the dimension to 64 dimensions, and the second layer outputs a single weight value. The activation function is Sigmoid. The dimension of the first layer of MLP is 128→64, and the dimension of the second layer is 64→1. The activation functions are ReLU (first layer) and Sigmoid (second layer). The weights are randomly initialized.

[0083] Based on independent weights and disease-gene association matrix and disease-miRNA association matrix , substitute the disease representation calculation formula to calculate the genetic perspective embedding of the disease and miRNA perspective embedding , the disease characterization calculation formula is:

[0084] ;

[0085] in, Embedding for genetic perspective or miRNA perspective embedding , h is g or m, A represents the disease-gene association matrix or disease-miRNA association matrix, Represents independent weights; aggregation is achieved through matrix multiplication to ensure computational efficiency. Weight normalization avoids numerical instability.

[0086] 104, gene-perspective embedding and miRNA-perspective embedding are used as model inputs to train a lightweight bilinear tower model;

[0087] Embedding a genetic perspective and miRNA perspective embedding As model input;

[0088] Disease pairs for known disease i and known disease j , embedded from a genetic perspective and miRNA perspective embedding Extract the representation and construct the gene feature vector and miRNA feature vector;

[0089] The expression of gene feature vector is:

[0090] ;

[0091] The expression of miRNA feature vector is:

[0092] ;

[0093] The feature vector includes concatenation, difference, and element-wise product, with a dimension of 128 × 4 = 512, ensuring that multiple interaction patterns of disease pairs are captured;

[0094] The dimensionality is reduced to a hidden dimension of 64 by linear projection, and the gene tower score and miRNA tower score are calculated using the bilinear transformation formula. The bilinear transformation formula is:

[0095] ;

[0096] is a learnable bilinear weight matrix, Score for the gene tower or miRNA tower score ; Linear projection layer dimension 512→64, bilinear matrix Dimension 64×64, randomly initialized, and optimized by backpropagation during training;

[0097] Find the gene tower score and miRNA tower score The average value of is used to obtain the final similarity score; ;Average fusion ensures the balance between gene perspective and miRNA perspective;

[0098] Based on the model input and the final similarity score, a lightweight bilinear tower model is trained. The linear projection and bilinear transformation in this step are implemented through matrix operations, with a single prediction time of approximately 0.05 seconds, making it suitable for real-time applications. The lightweight design (with approximately 100,000 parameters) reduces the computational complexity by 50% compared to traditional fully connected networks, making it suitable for resource-constrained scenarios. Overfitting is avoided through dropout (rate 0.3) and weight regularization (coefficient 0.0001).

[0099] 105, optimize the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost, and obtain the optimized lightweight bilinear tower model;

[0100] The process of alternating training strategy is:

[0101] In each round of training, the encoder parameters and bilinear tower parameters of one modality in the gene modality and miRNA modality are fixed, and the encoder parameters and bilinear tower parameters of the other modality are updated. The dynamic loss function is defined as:

[0102] ;

[0103] in, is the cross entropy loss, is the current mode prediction, is the relative entropy KL divergence, is the weight coefficient (default is 0.1); in each iteration, the gene modality (GCN+MLP+gene tower) and the miRNA modality (GATv2+MLP+miRNA tower) are updated alternately, and each modality is updated 10 times;

[0104] The process of global correction is:

[0105] Based on the regularization mechanism inspired by ReconBoost, all encoder parameters and bilinear tower parameters are jointly optimized every preset fixed rounds (specifically every 3 rounds). The global loss is defined as:

[0106] ;

[0107] is the weighted average forecast, is the learning rate; global correction is performed by the SGD optimizer (learning rate 0.005) to ensure modal consistency;

[0108] An optimized lightweight bilinear tower model is obtained based on dynamic loss and global loss;

[0109] In this step, RMSprop (learning rate 0.001) was used for the gene modality and RMSprop (learning rate 0.002) was used for the miRNA modality. SGD (learning rate 0.005) was used for global optimization. A total of 60 iterations, a batch size of 128, and training time of approximately 15 seconds were used (implemented on an NVIDIA RTX 4090 GPU). ReconBoost used the KL divergence constraint to constrain the prediction distribution between modalities ( \lambda = 0.1 ) to balance the loss and regularization term to prevent overfitting. Alternating training was used to reduce modality competition.

[0110] 106. Disease similarity prediction of target disease pairs is performed by optimizing the lightweight bilinear tower model to obtain the similarity prediction results of the target disease pairs.

[0111] After the optimized lightweight bilinear tower model is obtained through the above training, disease similarity prediction can be performed on the target disease pairs to obtain the similarity prediction results of the target disease pairs.

[0112] The beneficial effects achieved by the embodiments of the present invention are:

[0113] Gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix are collected to form a unified heterogeneous biological network; gene embedding features and miRNA embedding features are extracted from the heterogeneous biological network through graph convolutional network and improved graph attention network respectively to generate a multimodal feature set; multi-layer perceptron is used to calculate the independent weights of each feature in the multimodal feature set, and combined with the disease-gene association matrix and disease-miRNA association matrix to generate gene perspective embedding and miRNA perspective embedding of the disease; gene perspective embedding and miRNA perspective embedding are used as model input for training to obtain a lightweight bilinear tower model; the model parameters of the lightweight bilinear tower model are optimized through an alternating training strategy and a regularization mechanism inspired by ReconBoost to obtain an optimized lightweight bilinear tower model; disease similarity prediction of target disease pairs is performed through the optimized lightweight bilinear tower model to obtain the similarity prediction results of the target disease pairs. By deeply integrating gene networks, miRNA networks and disease association data, combined with graph neural networks, dynamic weight embedding, lightweight bilinear tower prediction and alternating training optimization strategies, high-precision, low-computational cost disease similarity prediction is achieved; it effectively makes up for the shortcomings of existing technologies in multivariate data integration, sparse data processing and prediction stability, and provides strong support for pathological mechanism research, disease classification optimization and drug repositioning.

[0114] In combination with the disease similarity prediction method based on multivariate data fusion described in the above embodiment, a disease similarity prediction system based on multivariate data fusion is described below through an embodiment.

[0115] like Figure 2 As shown, an embodiment of the present invention provides a disease similarity prediction system based on multivariate data fusion, comprising:

[0116] The multivariate data fusion module 201 is used to collect gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix to form a unified heterogeneous biological network;

[0117] A feature extraction module 202 is configured to extract gene embedding features and miRNA embedding features from heterogeneous biological networks using a graph convolutional network and an improved graph attention network, respectively, to generate a multimodal feature set;

[0118] Dynamic weight embedding module 203, for calculating the independent weight of each feature in the multimodal feature set using a multi-layer perceptron, combining the disease-gene association matrix and the disease-miRNA association matrix to generate a gene-perspective embedding and a miRNA-perspective embedding of the disease;

[0119] A lightweight bilinear tower prediction module 204 is used to train a lightweight bilinear tower model using gene perspective embedding and miRNA perspective embedding as model inputs;

[0120] an alternating training and optimization module 205 , configured to optimize the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost, thereby obtaining an optimized lightweight bilinear tower model;

[0121] The disease similarity prediction module 206 is used to perform disease similarity prediction on the target disease pair by optimizing the lightweight bilinear tower model to obtain a similarity prediction result of the target disease pair.

[0122] The beneficial effects achieved by the embodiments of the present invention are:

[0123] The multivariate data fusion module 201 is used to collect gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix to form a unified heterogeneous biological network; the feature extraction module 202 is used to extract gene embedding features and miRNA embedding features from the heterogeneous biological network through graph convolution network and improved graph attention network respectively to generate a multimodal feature set; the dynamic weight embedding module 203 is used to calculate the independent weight of each feature in the multimodal feature set by using a multilayer perceptron, and combine the disease-gene association matrix and the disease-miRNA association matrix to generate a disease feature set. Gene perspective embedding and miRNA perspective embedding; lightweight bilinear tower prediction module 204, used to train gene perspective embedding and miRNA perspective embedding as model input to obtain a lightweight bilinear tower model; alternating training and optimization module 205, used to optimize the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost, to obtain an optimized lightweight bilinear tower model; disease similarity prediction module 206, used to predict disease similarity for target disease pairs by optimizing the lightweight bilinear tower model, to obtain similarity prediction results for the target disease pairs. By deeply integrating gene networks, miRNA networks, and disease association data, combined with graph neural networks, dynamic weight embedding, lightweight bilinear tower prediction, and alternating training optimization strategies, high-precision, low-computational-cost disease similarity prediction is achieved; effectively making up for the shortcomings of existing technologies in multivariate data integration, sparse data processing, and prediction stability, providing strong support for pathological mechanism research, disease classification optimization, and drug repositioning.

[0124] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0126] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0128] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are included in the scope of the claims of the present invention to be approved.

Claims

1. A disease similarity prediction method based on multivariate data fusion, characterized in that: include: Collect gene network data, miRNA network data, disease-gene association matrix, and disease-miRNA association matrix to form a unified heterogeneous biological network; Gene embedding features and miRNA embedding features are extracted from the heterogeneous biological network through a graph convolutional network and an improved graph attention network, respectively, to generate a multimodal feature set; Calculating the independent weight of each feature in the multimodal feature set using a multilayer perceptron, and combining the disease-gene association matrix and the disease-miRNA association matrix to generate a gene perspective embedding and a miRNA perspective embedding of the disease; The gene perspective embedding and the miRNA perspective embedding are used as model inputs for training to obtain a lightweight bilinear tower model; Optimizing the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost to obtain an optimized lightweight bilinear tower model; The disease similarity prediction of the target disease pair is performed using the optimized lightweight bilinear tower model to obtain a similarity prediction result of the target disease pair.

2. The disease similarity prediction method based on multivariate data fusion according to claim 1, characterized in that: Gene Network Contains a collection of gene nodes and edge sets based on gene co-expression or functional association ; The graph structure of the gene network is ; miRNA network Contains miRNA node collection and edge sets based on miRNA similarity or regulatory relationships ; The graph structure of the miRNA network is .

3. The disease similarity prediction method based on multivariate data fusion according to claim 2, characterized in that: The collection of gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix to form a unified heterogeneous biological network includes: Collect the gene network Gene network data and miRNA network miRNA network data; Obtaining disease-gene association matrix and disease-miRNA association matrix , the disease-gene association matrix and the disease-miRNA association matrix Contains the association between known diseases and genes and miRNAs respectively; Through the disease-gene association matrix and the disease-miRNA association matrix , the gene network and the miRNA network The gene nodes and miRNA nodes in the dataset are mapped to disease nodes to form a unified heterogeneous biological network.

4. The disease similarity prediction method based on multivariate data fusion according to claim 3, characterized in that: The gene embedding features and miRNA embedding features are extracted from the heterogeneous biological network by using a graph convolutional network and an improved graph attention network, respectively, to generate a multimodal feature set, including: The gene network of the heterogeneous biological network based on the graph convolutional network GCN Perform feature extraction to obtain gene embedding features ; The miRNA network of the heterogeneous biological network based on the improved graph attention network GATv2 Perform feature extraction to obtain miRNA embedded features ; Collect the gene embedding features and the miRNA embedding characteristics , generating a multimodal feature set.

5. The disease similarity prediction method based on multivariate data fusion according to claim 1, characterized in that: The method of calculating the independent weight of each feature in the multimodal feature set using a multilayer perceptron and combining the disease-gene association matrix and the disease-miRNA association matrix to generate a gene perspective embedding and a miRNA perspective embedding of the disease includes: Applying two layers of multi-layer perceptrons to each feature in the multimodal feature set to calculate independent weights; Based on the independent weights and the disease-gene association matrix and the disease-miRNA association matrix, the disease representation calculation formula is substituted to calculate the gene perspective embedding of the disease. and miRNA perspective embedding , the disease characterization calculation formula is: ； Among them, the Embedding for genetic perspective or miRNA perspective embedding , the h is g or m, the A represents the disease-gene association matrix or the disease-miRNA association matrix, the represents independent weights.

6. The disease similarity prediction method based on multivariate data fusion according to claim 5, characterized in that: The method of using the gene perspective embedding and the miRNA perspective embedding as model inputs to train a lightweight bilinear tower model includes: Embedding the genetic perspective and the miRNA perspective embedded As model input; Disease pairs for known disease i and known disease j , embedded from the perspective of the gene and the miRNA perspective embedded Extract the representation and construct the gene feature vector and miRNA feature vector; The expression of the gene feature vector is: ； The expression of the miRNA feature vector is: ； After dimensionality reduction through linear projection, the gene tower score and miRNA tower score are calculated using the bilinear transformation formula. The bilinear transformation formula is: ； described is a learnable bilinear weight matrix, Score for the gene tower or miRNA tower score ; Calculate the gene tower score and the miRNA tower score The average value of is used to obtain the final similarity score; A lightweight bilinear tower model is trained based on the model input and the final similarity score value.

7. The disease similarity prediction method based on multivariate data fusion according to claim 1, characterized in that: The method optimizes the model parameters of the lightweight bilinear tower model by an alternating training strategy and a regularization mechanism inspired by ReconBoost to obtain an optimized lightweight bilinear tower model, including: Through the alternating training strategy, in each round of training, the encoder parameters and bilinear tower parameters of one modality in the gene modality and miRNA modality are fixed, and the encoder parameters and bilinear tower parameters of the other modality are updated. The dynamic loss function is defined as: ； Among them, the is the cross entropy loss, is the current modal prediction, is the relative entropy KL divergence, is the trade-off coefficient; Based on the regularization mechanism inspired by ReconBoost, all encoder parameters and bilinear tower parameters are jointly optimized every preset fixed rounds, and the global loss is defined as: ； described is the weighted average prediction, is the learning rate; An optimized lightweight bilinear tower model is obtained according to the dynamic loss and the global loss.

8. A disease similarity prediction system based on multivariate data fusion, characterized in that: include: Multivariate data fusion module, which is used to collect gene network data, miRNA network data, disease-gene association matrix and disease-miRNA association matrix to form a unified heterogeneous biological network; A feature extraction module is used to extract gene embedding features and miRNA embedding features from the heterogeneous biological network through a graph convolutional network and an improved graph attention network, respectively, to generate a multimodal feature set; A dynamic weight embedding module is used to calculate the independent weight of each feature in the multimodal feature set using a multilayer perceptron, and generate a gene perspective embedding and a miRNA perspective embedding of the disease by combining the disease-gene association matrix and the disease-miRNA association matrix; A lightweight bilinear tower prediction module, configured to train the gene perspective embedding and the miRNA perspective embedding as model inputs to obtain a lightweight bilinear tower model; an alternating training and optimization module, configured to optimize the model parameters of the lightweight bilinear tower model through an alternating training strategy and a regularization mechanism inspired by ReconBoost, thereby obtaining an optimized lightweight bilinear tower model; The disease similarity prediction module is used to perform disease similarity prediction on the target disease pair through the optimized lightweight bilinear tower model to obtain the similarity prediction result of the target disease pair.

Images