Pneumovirus gene data multi-view clustering integration method and device and electronic equipment

Abstract

The invention discloses a pneumovirus gene data multi-view clustering integration method and device and electronic equipment. The method comprises the following steps: constructing a width self-encoder with a width learning network and a global subspace structure; training a width auto-encoder, determining model parameters of the width auto-encoder, inputting multi-view pneumonia gene data into the trained width auto-encoder, and outputting a compressed feature subspace; clustering the compressed feature subspaces to obtain a plurality of basic partitions as integration members of an integration pool; the basic partition is converted into a fuzzy partition matrix, an undirected bipartite graph is constructed based on the integration pool, and connection weights between nodes of the undirected bipartite graph correspond to values in the fuzzy partition matrix; and cutting the undirected bipartite graph into several disjoint sub-graphs, and obtaining a clustering integration result of multi-view pneumonia gene data according to the sample nodes of each sub-graph. According to the method, parameter quantity and computing resources are remarkably reduced, and multi-view clustering analysis of pneumonia gene data is efficiently completed.

Description

Technical Field

[0001] The present invention relates to the field of machine learning, and in particular to a method, device and electronic device for clustering and integrating multi-views of pneumonia virus gene data. Background Art

[0002] Ensemble clustering techniques can classify pneumonia genes into multiple distinct groups, reflecting different gene expression and biological characteristics. These characteristics are associated with diverse clinical manifestations and may affect treatment outcomes. Therefore, identifying the similarities and differences in pneumonia genes is crucial for research and treatment.

[0003] Multi-perspective cluster analysis combines information from different data sources, each of which may provide unique insights and contribute to a comprehensive understanding of pneumonia genes. However, this approach faces multiple challenges: balancing the information from each data source, assessing its value, and dealing with the complexity brought by high-dimensional data.

[0004] Current clustering techniques often have heavy computational loads and poor results when processing large-scale data, which limits the application prospects of multi-view clustering in clinical and research. Summary of the invention

[0005] The main purpose of the present invention is to overcome the problems in the prior art of complex high-dimensional data processing and high computational cost, which limit the application of multi-view clustering technology in actual bioinformatics research and clinical practice, and propose a multi-view clustering integration method, device and electronic device for pneumonia virus gene data, which breaks through the traditional algorithm's dependence on a large amount of computing resources and enables rapid implementation in a conventional computing environment.

[0006] The present invention adopts the following technical solution:

[0007] A multi-view clustering integration method for pneumonia virus gene data, comprising:

[0008] Constructing a wide autoencoder with a wide learning network and a global subspace structure; training the wide autoencoder, determining the model parameters of the wide autoencoder, inputting multi-view pneumonia gene data into the trained wide autoencoder, and outputting a compressed feature subspace;

[0009] Clustering the compressed feature subspace to obtain multiple basic partitions as integrated members of the integrated pool; transforming the basic partitions into a fuzzy partition matrix, constructing an undirected bipartite graph based on the integrated pool, wherein the connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix;

[0010] The undirected bipartite graph is cut into several non-intersecting subgraphs, and the clustering integration results of the pneumonia gene data of the multiple views are obtained according to the sample nodes of each subgraph.

[0011] The global subspace structure is trained using a training data set to reconstruct the global subspace structure and obtain a linear combination matrix, and the width autoencoder is trained based on the linear combination matrix to determine the model parameters of the width autoencoder.

[0012] The objective function for reconstructing the global subspace structure is as follows:

[0013] ;

[0014] in, A matrix representing different views of the multi-view pneumonia gene data in the training data set connected together, represents the trade-off parameter of the global subspace structure, represents the transpose of a matrix, The linear combination matrix representing the global subspace structure.

[0015] The objective function for training the width autoencoder is as follows:

[0016] ;

[0017] in, represents the hidden layer of the width autoencoder, Represents multi-view pneumonia gene data, represents the weight of the width autoencoder after training, represents the trade-off parameter for the regularization term; represents the trade-off parameter for the self-expression error term, represents the Frobenius norm, represents the rank of the matrix, represents the identity matrix, The linear combination matrix representing the global subspace structure.

[0018] The obtained compressed feature subspace is as follows:

[0019] ;

[0020] represents the compressed feature subspace, represents the weight of the width autoencoder;

[0021] The K-Means clustering method is used to cluster the compressed feature subspace to obtain the basic partitions:

[0022] ;

[0023] in represents the integrated pool, Indicates a basic partition.

[0024] Transform the basic partition into a fuzzy partition matrix , as follows:

[0025] ;

[0026] ;

[0027] in, Indicates basic partition Middle The cluster of samples, Indicates basic partition Middle The cluster of samples, Indicates basic partition The cluster of the kth sample in express and The Euclidean distance, express and The Euclidean distance of .

[0028] An undirected bipartite graph is constructed based on the integrated pool. The nodes in the undirected bipartite graph use all samples and clusters in the integrated pool. The connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix. The formula is as follows:

[0029] ;

[0030] ;

[0031] in represents an undirected bipartite graph, represents the edge weight of the undirected bipartite graph node, represents a set of sample nodes, represents the set of cluster nodes, represents the fuzzy partitioning matrix, represents the index of the sample, represents the index of the cluster, represents a node consisting of samples and clusters, represents the connection weight based on the fuzzy partitioning matrix, Indicates View Samples To cluster The associated relationship.

[0032] The undirected bipartite graph is cut into several non-intersecting subgraphs using a graph cut algorithm, each of which contains the sample node set. and the clustering node set The subset of the undirected bipartite graph is maintained, and the sample nodes belonging to the same subgraph are divided into the same cluster, thus forming the final consensus partition. , where each sample node is uniquely assigned to a cluster, and the clustering integration result of the multi-view pneumonia gene data is obtained.

[0033] A multi-view clustering integration device for pneumonia virus gene data, comprising:

[0034] A width autoencoder construction module is used to construct a width autoencoder with a width learning network and a global subspace structure, and the model parameters of the width autoencoder are determined through training. The multi-view pneumonia gene data is input into the trained width autoencoder to output a compressed feature subspace;

[0035] An integration module, clustering the compressed feature subspace to obtain a plurality of basic partitions as integrated members of an integrated pool; transforming the basic partitions into a fuzzy partition matrix, constructing an undirected bipartite graph based on the integrated pool, wherein the connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix;

[0036] The clustering module cuts the undirected bipartite graph into several non-intersecting subgraphs, and obtains the clustering integration result of the pneumonia gene data of the multiple views according to the sample nodes of each subgraph.

[0037] An electronic device, comprising:

[0038] one or more processors;

[0039] a storage device for storing one or more programs,

[0040] When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-view clustering integration method of pneumonia virus gene data as described above.

[0041] It can be seen from the above description of the present invention that, compared with the prior art, the present invention has the following beneficial effects:

[0042] The wide learning autoencoder for multi-view tasks constructed by the present invention obtains a lightweight model by optimizing the structure and training method, uses the trained wide autoencoder to extract features of the input pneumonia virus gene data to obtain a compressed feature subspace, and performs clustering integration based on the compressed feature subspace, which significantly reduces the number of parameters and computing resources, and efficiently completes the multi-view clustering analysis of pneumonia gene data.

[0043] The present invention also includes clustering the compressed feature subspace to obtain multiple basic partitions as integrated members of the integrated pool, and then transforming the basic partitions into a fuzzy partition matrix to capture the relationship between samples and clusters; running a graph cut algorithm on the constructed bipartite graph to obtain high-quality consensus partitions, and then completing multi-view clustering integration, which significantly optimizes the accuracy and adaptability of the model and enhances its application value in a variety of practical scenarios.

[0044] This method breaks through the limitation of traditional multi-view clustering algorithms that only learn shared representations and fully retains view-specific information. In the feature generation stage, this method innovatively combines the global subspace structure and local feature mapping through MVBLSAE, thereby effectively extracting the compressed feature subspace and avoiding noise interference. In the integration stage, the sample-cluster association relationship is accurately captured by constructing a fuzzy partition matrix, and high-quality consensus partitioning is achieved based on the bipartite graph structure, thereby enhancing the clustering effect of multi-view data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is the main flow chart of the method of the present invention;

[0046] Figure 2 and Figure 3 A comparison chart showing the effect of using ablation experiments to verify the model.

[0047] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. DETAILED DESCRIPTION

[0048] The present invention is further described below through specific implementation modes.

[0049] A multi-view clustering integration method for pneumonia virus gene data, comprising the following:

[0050] S1 constructs a wide autoencoder with a wide learning network and a global subspace structure; trains the wide autoencoder, determines the model parameters of the wide autoencoder, inputs multi-view pneumonia gene data into the trained wide autoencoder, and outputs a compressed feature subspace.

[0051] Among them, the width learning network is used to connect the matrices from different views into a whole, and the global subspace structure is used to learn the global subspace structure of the input data according to the least squares regression problem. The global subspace structure is trained using the training data set to reconstruct the global subspace structure and obtain the linear combination matrix. The width autoencoder is trained based on the linear combination matrix to determine the model parameters of the width autoencoder.

[0052] The training dataset is a dataset constructed from multi-view pneumonia gene data, and the objective function of the reconstructed global subspace structure is as follows:

[0053] ;

[0054] in, Represents the matrix formed by connecting different views of the multi-view pneumonia gene data in the training dataset, represents the trade-off parameter of the global subspace structure, represents the transpose of a matrix, The linear combination matrix representing the global subspace structure.

[0055] The objective function for training the width autoencoder is as follows:

[0056] ;

[0057] in, represents the hidden layer of the width autoencoder, Represents multi-view pneumonia gene data, represents the weight of the width autoencoder after training, represents the trade-off parameter for the regularization term; represents the trade-off parameter for the self-expression error term, represents the Frobenius norm, represents the rank of the matrix, represents the identity matrix, The linear combination matrix representing the global subspace structure.

[0058] After the multi-view pneumonia gene data is input into the trained width autoencoder, the compressed feature subspace is output. The compressed feature subspace is the multi-view pneumonia gene data after dimensionality reduction, as follows:

[0059] ;

[0060] represents the compressed feature subspace, Represents the weights of the width autoencoder.

[0061] S2 clusters the compressed feature subspace to obtain multiple basic partitions as integrated members of the integrated pool; transforms the basic partitions into a fuzzy partition matrix, and constructs an undirected bipartite graph based on the integrated pool. The connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix.

[0062] The K-Means clustering method is used to cluster the compressed feature subspace to obtain the basic partition:

[0063] ;

[0064] in represents the integrated pool, Indicates a basic partition.

[0065] The basic partition is transformed into a fuzzy partition matrix as follows:

[0066] ;

[0067] ;

[0068] in, Indicates basic partition Middle The cluster of samples, Indicates basic partition Middle The cluster of samples, Indicates basic partition The cluster of the kth sample in express and The Euclidean distance of express and The Euclidean distance of .

[0069] An undirected bipartite graph is constructed based on the integrated pool. The nodes in the undirected bipartite graph use all samples and clusters in the integrated pool. The connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix. The formula is as follows:

[0070] ;

[0071] ;

[0072] in represents an undirected bipartite graph, represents the edge weight of the bipartite graph node, represents the set of sample nodes, represents the set of cluster nodes, represents the fuzzy partitioning matrix, represents the index of the sample, represents the index of the cluster, represents a node consisting of samples and clusters, represents the connection weight based on the fuzzy partitioning matrix, Indicates View Samples To cluster The associated relationship.

[0073] S3 cuts the undirected bipartite graph into several non-overlapping subgraphs and obtains the clustering integration results of multi-view pneumonia gene data based on the sample nodes of each subgraph.

[0074] The undirected bipartite graph is cut into disjoint subgraphs, each of which still contains a set of sample nodes and cluster node set The subset of the bipartite graph is maintained, and the sample nodes belonging to the same subgraph are divided into the same cluster, thus forming the final consensus partition. , where each sample node is uniquely assigned to a cluster, and the clustering integration result of the multi-view pneumonia gene data is obtained.

[0075] The method of the present invention is verified as follows: the experimental dataset used mainly consists of a dataset of 100 plant leaves (100 leaves), a facial image dataset (ORL), 474 images (Caltech 7), and 7 different categories of images (MSRC-v1). For the evaluation of clustering performance, two widely used metrics are used, namely normalized mutual information (NMI) and adjusted Rand index (ARI). The standard mutual information is normalized, where 0 means the two partitions are completely unrelated and 1 means the two partitions are consistent. It is a cluster evaluation metric obtained by adjusting the original Rand Index based on the consistency of sample pair distribution.

[0076] Table 1 Comparative analysis of NMI measurements

[0077]

[0078] Table 2 Comparative analysis of ARI measurements

[0079]

[0080] The best results are highlighted in bold. According to the NMI metric, our method MVEC-BLSAE ranks first on three datasets and second on the MSRC-v1 dataset. In terms of ARI, it ranks first on two datasets.

[0081] At the same time, ablation experiments are used to prove the effect of the model constructed by the present invention. Figure 2 As shown in the figure, the effectiveness of the model of the present invention is proved by comparing the three models. This phenomenon can be attributed to two main reasons:

[0082] 1. MVEC-BLSAE can synergistically integrate the global subspace structure and local view features to extract a compact feature subspace to avoid noise interference.

[0083] 2. The graph-based consistency method absorbs rich sample-cluster relationships by constructing a fuzzy partitioning matrix, and achieves high-quality clustering partitioning through a graph cutting algorithm.

[0084] In summary, the method of the present invention innovatively applies the widely learned stacked autoencoder (BLSAE) to multi-view data. The method first designs a multi-view BLSAE (MVBLSAE) to simultaneously consider the global subspace structure and local view-specific features, extracts a compressed feature subspace for each view and generates an initial partition. The initial partition is then converted into a fuzzy membership matrix to capture the association between samples and clusters. Finally, it runs a bipartite graph constructed on a graph cut algorithm to obtain high-quality consensus partitions. A large number of experiments verify the superior performance of the method, prove the effectiveness of the MVBLSAE and graph cut consistency steps, and demonstrate the good prospects of BLSAE innovatively applied to multi-view clustering tasks.

[0085] Based on this, the present invention also proposes a multi-view clustering integration device for pneumonia virus gene data, which adopts the above-mentioned multi-view clustering integration method for pneumonia virus gene data, including:

[0086] A wide autoencoder construction module is used to construct a wide autoencoder with a wide learning network and a global subspace structure, and the model parameters of the wide autoencoder are determined through training. The multi-view pneumonia gene data is input into the trained wide autoencoder to output a compressed feature subspace. The wide autoencoder construction module is used to perform step S1 of the above method.

[0087] The integration module clusters the compressed feature subspace to obtain multiple basic partitions as integrated members of the integration pool; transforms the basic partitions into a fuzzy partition matrix, constructs an undirected bipartite graph based on the integration pool, and the connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix. The integration module is used to execute step S2 of the above method.

[0088] The clustering module cuts the undirected bipartite graph into several non-intersecting subgraphs, and obtains the clustering integration results of the pneumonia gene data of multiple views according to the sample nodes of each subgraph. The clustering module is used to execute step S3 of the above method.

[0089] An electronic device, comprising:

[0090] one or more processors;

[0091] a storage device for storing one or more programs,

[0092] When one or more programs are executed by one or more processors, the one or more processors implement a multi-view clustering integration method for pneumonia virus gene data.

[0093] The present invention further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device.

[0094] The computer-readable medium carries one or more programs. When the one or more programs are executed by an electronic device, the electronic device implements the method in the above embodiment.

[0095] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0096] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0097] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure.

[0098] The above is only a specific implementation of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A multi-view clustering integration method for pneumonia virus gene data, characterized in that: include: Construct a wide autoencoder with a wide learning network and a global subspace structure; Training the width autoencoder to determine the model parameters of the width autoencoder, inputting the multi-view pneumonia gene data into the trained width autoencoder and outputting a compressed feature subspace; Clustering the compressed feature subspace to obtain multiple basic partitions as integrated members of the integrated pool; transforming the basic partitions into a fuzzy partition matrix, constructing an undirected bipartite graph based on the integrated pool, wherein the connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix; The undirected bipartite graph is cut into several non-intersecting subgraphs, and the clustering integration results of the pneumonia gene data of the multiple views are obtained according to the sample nodes of each subgraph.

2. A multi-view clustering integration method for pneumonia virus gene data as claimed in claim 1, characterized in that: The global subspace structure is trained using a training data set to reconstruct the global subspace structure and obtain a linear combination matrix, and the width autoencoder is trained based on the linear combination matrix to determine the model parameters of the width autoencoder.

3. A multi-view clustering integration method for pneumonia virus gene data as described in claim 2, characterized in that: The objective function for reconstructing the global subspace structure is as follows: ； in, A matrix representing different views of the multi-view pneumonia gene data in the training data set connected together, represents the trade-off parameter of the global subspace structure, represents the transpose of a matrix, The linear combination matrix representing the global subspace structure; The objective function for training the width autoencoder is as follows: ； in, represents the hidden layer of the width autoencoder, Represents multi-view pneumonia gene data, represents the weight of the width autoencoder after training, represents the trade-off parameter for the regularization term; represents the trade-off parameter for the self-expression error term, represents the Frobenius norm, represents the rank of the matrix, represents the identity matrix, The linear combination matrix representing the global subspace structure.

4. A multi-view clustering integration method for pneumonia virus gene data as claimed in claim 1, characterized in that: The obtained compressed feature subspace is as follows: ； represents the compressed feature subspace, represents the weight of the width autoencoder; The K-Means clustering method is used to cluster the compressed feature subspace to obtain the basic partitions: ； in represents the integrated pool, Indicates a basic partition.

5. A multi-view clustering integration method for pneumonia virus gene data as described in claim 4, characterized in that: Transform the basic partition into a fuzzy partition matrix , as follows: ； ； in, Indicates basic partition Middle The cluster of samples, Indicates basic partition Middle The cluster of samples, Indicates basic partition The cluster of the kth sample in express and The Euclidean distance, express and The Euclidean distance of .

6. A multi-view clustering integration method for pneumonia virus gene data as claimed in claim 5, characterized in that: An undirected bipartite graph is constructed based on the integrated pool. The nodes in the undirected bipartite graph use all samples and clusters in the integrated pool. The connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix. The formula is as follows: ； ； in represents an undirected bipartite graph, represents the edge weight of the undirected bipartite graph node, represents a set of sample nodes, represents the set of cluster nodes, represents the fuzzy partitioning matrix, represents the index of the sample, represents the index of the cluster, represents a node consisting of samples and clusters, represents the connection weight based on the fuzzy partitioning matrix, Indicates View Samples To cluster The associated relationship.

7. A multi-view clustering integration method for pneumonia virus gene data as claimed in claim 6, characterized in that: The undirected bipartite graph is cut into several non-intersecting subgraphs using a graph cut algorithm, each of which contains the sample node set. and the clustering node set The subset of the undirected bipartite graph is maintained, and the sample nodes belonging to the same subgraph are divided into the same cluster, thus forming the final consensus partition. , where each sample node is uniquely assigned to a cluster, and the clustering integration result of the multi-view pneumonia gene data is obtained.

8. A multi-view clustering integration device for pneumonia virus gene data, characterized in that: include: A width autoencoder construction module is used to construct a width autoencoder with a width learning network and a global subspace structure, and the model parameters of the width autoencoder are determined through training. The multi-view pneumonia gene data is input into the trained width autoencoder to output a compressed feature subspace; An integration module, clustering the compressed feature subspace to obtain a plurality of basic partitions as integrated members of an integrated pool; transforming the basic partitions into a fuzzy partition matrix, constructing an undirected bipartite graph based on the integrated pool, wherein the connection weights between the nodes of the undirected bipartite graph correspond to the values ​​in the fuzzy partition matrix; The clustering module cuts the undirected bipartite graph into several non-intersecting subgraphs, and obtains the clustering integration result of the pneumonia gene data of the multiple views according to the sample nodes of each subgraph.

9. An electronic device, comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement A multi-view clustering integration method for pneumonia virus gene data is provided as described in any one of claims 1-7.

Images