Cancer subtype detection method based on multi-view clustering under diversity criterion

Abstract

The present invention discloses a method for detecting cancer subtypes based on multi-view clustering under a diversity criterion, which relates to the field of data processing technology and solves the problem that existing multi-view clustering methods ignore the differences between multiple views and the problems existing in the representation tensor. The present invention improves on the classic multi-view method, divides the representation tensor into a clean part and a noisy part, adds a weighted tensor nuclear norm to the clean representation tensor to take into account the prior knowledge of singular values, and applies l to the noisy part. 2,1 Norm constraints make the learned representation tensor clearer and enhance the robustness of the algorithm. A difference term is introduced into the objective function to describe the diversity of the multi-view representation matrix, and a manifold regularization term is introduced to preserve the local structure of the data. This improves the effectiveness and robustness of multi-view clustering, achieving excellent performance on various datasets.

Description

Technical Field

[0001] The present invention relates to the technical fields of signal processing and data analysis, and in particular to cancer subtype detection based on multi-view clustering under a diversity criterion. Background Art

[0002] Multi-view clustering is widely used in machine learning and is an important foundational computer science method. Because multi-view data is derived from different views of the same data, there is hidden high-dimensional information between the different views. Fully mining and utilizing this information can improve clustering accuracy.

[0003] In recent years, many multi-view clustering methods have been designed. Multi-view subspace clustering collects data from multiple views and can exploit the common features and differences between different views. Shi et al. [1] proposed Multiview Robust graph-based Clustering (MRGC) to reduce dimensionality and mitigate the impact of noise. Wu et al. [2] proposed a low-rank approximate clustering (LRAcluster) method, which uses a complex probabilistic model based on low-rank approximation to easily identify the shared main subspace from various data sources. Liu et al. [3] Survival supervised graph clustering (S2GC) is proposed to utilize the survival information of multiple groups of data in the clustering process.

[0004] Some models use tensors to capture high-order relationships between multi-view data, such as Zhang et al. [4] First, we consider constructing a representation tensor from the representation matrix, which captures the high-order connections of multi-view data. Although the current tensor-based multi-view clustering methods have achieved good performance, there are still some shortcomings: (1) Different views have different representation information, but many tensor-based clustering methods do not fully consider the diversity between multiple views; (2) The representation tensor lacks cleanliness and robustness, making the learned representation tensor unclear, affecting the robustness of the algorithm; (3) The local structure of the data is ignored. These problems all affect the final clustering results. In order to further enhance the application effect of multi-view clustering, the above problems need to be solved urgently. Summary of the Invention

[0005] The present invention provides a cancer subtype detection method, device and medium based on multi-view clustering under a diversity criterion, so as to solve the problem that the existing multi-view clustering method has a reduced clustering effect due to ignoring the above factors.

[0006] The present invention is achieved through the following technical solutions:

[0007] A first aspect of the present invention provides a method for detecting cancer subtypes based on multi-view clustering under a diversity criterion, comprising:

[0008] Acquire a data matrix of each view in the multi-view data, and represent the data matrix using a representation matrix of the data of each view and a self-representation error matrix;

[0009] Obtain a representation tensor of the multi-view data based on the representation matrix, split the representation tensor into a clean representation tensor and a representation error matrix; obtain a noise matrix of the data matrix based on the self-representation error matrix and the representation error matrix;

[0010] Applying norm constraints to the clean representation tensor and the noise matrix respectively, and introducing a regularization term and a difference term of the representation matrix to construct an optimization objective function for multi-view clustering;

[0011] Solving the optimization problem of the optimization objective function to obtain a solution result of the clean representation tensor;

[0012] Calculating an affinity matrix of the multi-view data according to a solution result of the clean representation tensor;

[0013] Perform spectral clustering on the affinity matrix to obtain a clustering result.

[0014] The present invention represents multi-view data in a subspace based on the self-representation method, uses tensors to represent the multi-view representation matrix to capture high-order connections, and divides the representation tensor into a clean part and a noisy part, and imposes norm constraints on the clean part and the noisy part respectively, so that the learned representation tensor is clearer and the robustness of the algorithm is enhanced. By introducing a difference term in the objective function to describe the diversity of the multi-view representation matrix and introducing manifold regularization to maintain the local structure of the data, the problem of existing multi-view clustering ignoring view diversity and local features is solved. Finally, after solving the clean representation tensor, the affinity matrix is ​​calculated and spectral clustering is performed, thereby improving the effect and robustness of multi-view clustering. Experimental data show that the multi-view clustering method of the present invention has achieved excellent performance on a variety of data sets.

[0015] Furthermore, the data matrix is ​​represented by a representation matrix of each view data and a self-representation error matrix, specifically:

[0016]

[0017] Among them, X (v) is the data matrix of the v-th view, Z (v) is the representation matrix of the v-th view, is the self-representation error matrix of the v-th view.

[0018] Furthermore, the representation tensor of the multi-view data is:

[0019] Z=Φ(Z (1) ,Z (2) ,…Z (v) …,Z (v) )

[0020] Among them, Z (v) is the representation matrix of the v-th view, v∈[0,V], V is the number of views, and Φ(·) is a tensor operation.

[0021] Furthermore, the splitting of the representation tensor into a clean representation tensor and a representation error matrix is ​​specifically as follows:

[0022] Z=S+E2

[0023] Where Z is the representation tensor of multi-view data, S is the clean representation tensor, and E2 is the representation error matrix.

[0024] Furthermore, the regularization term of the representation matrix is ​​a graph-based Laplace regularization term, expressed as

[0025]

[0026] Among them, L (v) represents the Laplacian matrix of the v-th view, Z (v) represents the representation matrix of the v-th view, It's Z (v) is the transposed matrix of , and tr(·) represents the trace of the matrix.

[0027] Furthermore, the difference terms of the representation matrix are orthogonal difference terms, which are expressed as: tr(Z (v) ,Z (h) );

[0028] Among them, Z (v) is the representation matrix of the v-th view, Z (h) is the representation matrix of the h-th view, v≠h, tr(·) represents the trace of the matrix, and the larger the value of the difference term, the greater the diversity between the views.

[0029] Furthermore, the optimization objective function is:

[0030]

[0031] Z=Φ(Z (v) ,Z (2) ,…,Z (V) )

[0032] Z=S+E2

[0033]

[0034] Among them, Z (v) 、S (v) , E denote the representation matrix, clean representation tensor and noise matrix of the v-th view respectively;

[0035] The first term of the objective function is the weighted norm of the clean representation vector, which is used to fuse the prior information of the singular value and induce the low rank of the tensor; the second term is the l of the noise matrix 2,1 norm, used to reduce the impact of noise in the representation matrix and tensor; the third term is Laplace manifold regularization, which is used to maintain the local structure of the data, L (v) is the Laplacian matrix of the vth view, V is the number of views, tr(·) represents the trace of the matrix; the fourth term is the orthogonal difference term, which is used to represent the diversity of multiple views, Z (h) is the representation matrix of the h-th view, v≠h; α, β, γ are balance coefficients;

[0036] The first constraint is used to perform subspace clustering based on self-representation, X (v) is the data matrix of the v-th view, Z (v) is the representation matrix of the v-th view, is the self-representation error matrix of the v-th view; the second term takes the representation matrix of multiple views as the representation tensor; the third term divides the representation tensor into a clean part and an error part, S is the clean representation tensor, E2 is the representation error matrix corresponding to the v-th view; the fourth term is the noise matrix constraint.

[0037] Furthermore, the affinity matrix of the multi-view data is expressed as:

[0038]

[0039] Among them, C represents the affinity matrix, S is the solution result of the clean representation tensor, and S T is the transpose of S, v represents the vth view, and V is the number of views.

[0040] Furthermore, the cancer subtype detection method based on multi-view clustering under the diversity criterion is applied to the cancer sample data to obtain cancer subtype clustering results.

[0041] A second aspect of the present invention provides 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 cancer subtype detection method based on multi-view clustering under diversity criteria as described in the first aspect of the present invention when executing the computer program.

[0042] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for detecting cancer subtypes based on multi-view clustering under a diversity criterion according to the first aspect is implemented.

[0043] Compared with the existing technology, the present invention has the following advantages and beneficial effects: it improves on the classic LT-MSC, represents multi-view data in a subspace based on the self-representation method, introduces a difference term in the objective function to describe the diversity of the multi-view representation matrix, and uses Laplace regularization to maintain the local structure, imposes robustness constraints on the representation tensor, and makes the learned representation tensor clearer. The clean representation tensor obtained by the solution is expanded and the affinity matrix is ​​calculated, and finally the affinity matrix is ​​spectrally clustered. The clustering effect and robustness of multi-view clustering are further improved, and both in conventional machine learning tasks and in multi-omics cancer classification, clearer classification results are achieved compared with existing classification methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the examples. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be considered as limiting the scope. A person of ordinary skill in the art can also derive other relevant drawings based on these drawings without inventive effort. In the drawings:

[0045] Figure 1 This is a flow chart of a method for detecting cancer subtypes based on multi-view clustering under a diversity criterion according to an embodiment of the present invention;

[0046] Figure 2 This is a performance comparison chart of the method proposed in the present invention and the comparative method on a cancer dataset;

[0047] Figure 3 The figure is a schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with examples and drawings. The exemplary embodiments of the present invention and their descriptions are only used to explain the present invention and are not intended to limit the present invention.

[0049] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to other steps or units inherent in the device.

[0050] The terms used in the various embodiments of the application are only used to describe the purpose of specific embodiments and are not intended to limit the various embodiments of the application. As used herein, the singular form is intended to also include the plural form, unless the context clearly indicates otherwise. Unless otherwise limited, all terms used here (including technical terms and scientific terms) have the same meaning as the meaning generally understood by those of ordinary skill in the art of the application. The terms (such as the terms defined in the dictionary generally used) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having idealized meaning or too formal meaning, unless clearly defined in the various embodiments of the application.

[0051] A classic method of tensor-based multi-view clustering (LT-MSC) is as follows:

[0052]

[0053] xT (v) =X (v) Z (v) +E (v) ,v=1,2,…V

[0054] Among them, X (v) represents the data matrix of the v-th view, Z (v) represents the representation matrix of the v-th view, E (v) represents the error matrix of the v-th view, Z represents the tensor, E represents the noise, represents the tensor nuclear norm, ‖·‖ 2,1 Representative 2,1 norm, and λ is the balance parameter.

[0055] Existing methods only represent the data and error components of the representation matrix, resulting in a lack of clean tensor representation. Furthermore, the learned representation tensor may contain errors, resulting in unclear representation tensors for multi-view data, which can lead to blurred clustering boundaries. Furthermore, existing methods only constrain the norm of the representation tensor and the noise component, failing to consider the diversity between multiple views and the local nature of the data, which also affects clustering results.

[0056] The present invention improves upon the classic LT-MSC, further enhancing the clustering effect and robustness of multi-view clustering, enabling it to achieve excellent performance in multi-view clustering and classification tasks. In particular, when applied to multi-omics cancer classification, it achieves clearer cancer subtype classification results compared to existing multi-omics cancer classification methods, providing an important basis for providing accurate diagnosis and treatment to patients.

[0057] Example 1

[0058] See also Figure 1 、 Figure 2 As shown, Figure 1 The flowchart of the cancer subtype detection method based on multi-view clustering under the diversity criterion according to an embodiment of the present invention is presented. Figure 2 The flowchart of the method is presented in FIG. 1 , which includes the following steps:

[0059] S1, obtaining a data matrix of each view in the multi-view data, and representing the data matrix by a representation matrix of each view data and a self-representation error matrix.

[0060] Using X (v) represents the data matrix of the v-th view, d v is the feature number of the data matrix, n is the number of samples of the data matrix, and the data matrix of each view is converted into a pure data part and a noise part. The pure data part is equal to the data matrix multiplied by the representation matrix, and the noise part is the self-representation error matrix, which is expressed as follows:

[0061]

[0062] Among them, X (v) is the data matrix of the v-th view, Z (v) is the representation matrix of the v-th view, is the self-representation error matrix of the v-th view.

[0063] S2, obtaining a representation tensor of the multi-view data based on the representation matrix, splitting the representation tensor into a clean representation tensor and a representation error matrix; and obtaining a noise matrix of the data matrix based on the self-representation error matrix and the representation error matrix.

[0064] Specifically, the representation tensor of multi-view data is obtained by the tensor representation of the representation matrix of each view, and the formula is:

[0065] Z=Φ(Z (1) ,Z (2) ,…Z (v) …,Z (V) )

[0066] Among them, Z (v) represents the representation matrix of the v-th view, v∈[0,V], V is the number of views, and Φ(·) is a tensor operation.

[0067] Split the representation tensor into the clean representation tensor and the representation error matrix, the formula is:

[0068] Z=S+E2

[0069] Where Z is the representation tensor of multi-view data, S is the clean representation tensor, and E2 is the representation error matrix.

[0070] The noise matrix of the data matrix consists of the self-representation error and the representation error matrix of the representation tensor, which can be expressed as:

[0071]

[0072] in, is the self-representation error matrix corresponding to the v-th view, is the representation error matrix corresponding to the v-th view.

[0073] S3 imposes norm constraints on the clean representation tensor and noise matrix respectively, and introduces the regularization term and difference term of the representation matrix to construct the optimization objective function of multi-view clustering.

[0074] Among them, the nuclear norm is applied to the clean representation tensor, which is expressed as Apply l to the noise matrix 2,1 norm, denoted as ‖E‖ 2,1 .

[0075] The regularization term uses manifold regularization to maintain the local structure of the data. In this embodiment, graph-based Laplace manifold regularization is used, but other manifold regularization terms can also be used instead. The graph-based Laplace operator effectively describes the similarity between data and is widely used to learn local manifold structures. For the vth view, Laplace regularization can be expressed as:

[0076]

[0077] in, and Z (v) The i-th and j-th columns of It's W (v) The element in row i and column j of is the square of the two-norm, L (v) represents the Laplacian matrix of the v-th view, Z (v) represents the representation matrix of the v-th view, It's Z (v) The transposed matrix of .

[0078] The difference term is used to represent the diversity between multiple views. The difference may be expressed by an orthogonal difference term, a HSIC difference term, or other difference terms. In this embodiment, an orthogonal difference term is used.

[0079] The diversity between two vectors x and y can be measured as follows:

[0080]

[0081] Where D(x,y) is used to represent the diversity (or difference) between vectors x and y. The larger D(x,y) is, the greater the difference between the surface vectors x and y is, that is, the greater the diversity is. ||.||0 is the l0 norm, and ⊙ is the Hadamard product.

[0082] Based on the above formula, the difference between two views can be expressed as:

[0083] D(Z (v) ,Z (h) )=tr(Z (v) ,Z (h) )

[0084] Among them, Z (v) is the representation matrix of the v-th view, Z (h) is the representation matrix of the h-th view, v≠h, and tr(·) is the trace of the matrix.

[0085] The optimization objective function is constructed to solve the optimization problem of the above four items, which can be expressed as:

[0086]

[0087] Z=Φ(Z (v) ,Z (2) ,…,Z (V) )

[0088] Z=S+E2

[0089]

[0090] Among them, Z (v) 、S (v) 、E (v) denote the representation matrix, clean representation vector and noise matrix of the v-th view respectively;

[0091] The first term of the objective function is the weighted norm of the clean representation vector, which is used to fuse the prior information of the singular value and induce the low rank of the tensor; the second term is the l of the noise matrix 2,1 norm, used to reduce the impact of noise in the representation matrix and tensor; the third term is Laplace manifold regularization, used to maintain the local structure of the data; the fourth term is the orthogonal difference term, used to represent the diversity of multiple views;

[0092] The first constraint is used to perform self-representation based subspace clustering; the second constraint uses the representation matrix of multiple views as the representation tensor; the third constraint divides the representation tensor into clean and error parts; and the fourth constraint is the noise matrix constraint.

[0093] S4, solving the optimization problem of the optimization objective function to obtain a solution result of the clean representation tensor; and calculating the affinity matrix of the multi-view data according to the solution result of the clean representation tensor.

[0094] Solve the optimization problem of the above optimization objective function, that is, solve the parameter value corresponding to the minimization of the objective function. After obtaining the solution of the clean representation tensor, substitute it into the following formula to obtain the affinity matrix of the multi-view data:

[0095]

[0096] Among them, C represents the affinity matrix, S is the solution result of the clean representation tensor, and S T is the transpose of S, v represents the vth view, and V is the number of views.

[0097] S5, finally, spectral clustering is applied to the C matrix to obtain the final clustering result.

[0098] In summary, the present invention has the following improvements based on the existing technology:

[0099] 1. The orthogonal difference terms are used to enhance the difference of multi-view representation matrices.

[0100] 2. Divide the representation tensor into a clean part and a noisy part, and apply the tensor nuclear norm to the clean part and l to the noisy part. 2,1 Norm constraints make the learned representation tensor clearer and enhance the robustness of the algorithm.

[0101] 3. The introduction of Laplace manifold learning protects the local geometric structure of the data and improves clustering performance.

[0102] Example 2

[0103] In order to verify the effectiveness and efficiency of the cancer subtype detection method based on multi-view clustering under the diversity criterion of the present invention, a large number of experiments were carried out on 8 commonly used machine learning data sets and 8 cancer data sets. Six common judgment indicators were used: accuracy (ACC), normalized mutual information (NMI), adjusted Rand coefficient (AR), F score (F-score), precision (Precision) and recall (Recall) to quantitatively evaluate the clustering results. For these six indicators, the higher the value, the better the clustering effect. In addition, the p-value indicator generated by survival analysis was used on the cancer data set. The smaller the p-value indicator, the better the clustering effect.

[0104] (1) Performance of the multi-view clustering method of the present invention on cancer datasets.

[0105] Comparison method: Three commonly used cancer subtype classification methods were selected and compared with the performance of the method of the present invention on the machine learning dataset. The three comparison methods include: SNF [5] , MALS [6] , MRGC [1] .

[0106] The cancer datasets used in this example include: acute myeloid leukemia (AML), colon cancer (COAD), clear cell renal cell carcinoma (KIRC), hepatocellular carcinoma (LIHC), lung squamous cell carcinoma (LUSC), cutaneous melanoma (SKCM), ovarian serous cystadenocarcinoma (OV), and sarcoma (SARC). These datasets include three omics: gene expression, DNA methylation, and miRNA expression. Cancer databases do not provide the true number of clusters, so we use the rotation cost method to estimate the number of clusters.

[0107] The cancer subtype detection method based on multi-view clustering under the diversity criterion described in Example 1 of the present invention was applied to cancer sample data, resulting in cancer subtype clustering results. Table 1 shows a comparison of the proposed method and the method of the present invention on eight cancer datasets. The mean and standard deviation of 10 replicates are reported, with the results in bold representing the best results. Compared to three leading cancer subtype classification methods, the proposed method achieved superior clustering performance on the cancer datasets.

[0108] Table 1 Clustering results (p-value) of the proposed method and three other methods on 8 cancer datasets

[0109]

[0110]

[0111] The present invention conducted a clustering result visualization experiment on the UCI dataset. In the figure, the greater the distance between clusters, or the smaller the distance between points in each cluster, the better the clustering effect. Compared with three advanced cancer subtype classification methods, Figure 3 It can be seen that the clustering results obtained by this method are significantly superior to those of the other methods. This method uses Laplace manifold learning to preserve the local geometric structure of the original data, employs orthogonal dissimilarity constraints to reflect the differences between different views, and separates the representation tensor into a clean and noisy part, enhancing the robustness of the algorithm. Therefore, considering these characteristics, this method achieves significantly better clustering results than the other three compared methods.

[0112] (2) The performance of the multi-view clustering method of the present invention on commonly used machine learning datasets.

[0113] The machine learning datasets used include: YaleA face image dataset, ORL face dataset, UCI handwritten digit dataset, NGs news dataset, BBC news dataset, 3Sources text dataset, COIL20 dataset, and 100leaves plant sample dataset.

[0114] The comparison methods, SNF, MALS, and MRGC, were used to compare the performance of the present method with that of the present invention on machine learning datasets. Table 2 shows the comparison of the present method and the comparison methods on eight machine learning datasets. The average and standard deviation of 10 replicates are recorded, with the results in bold representing the best results. As can be seen, the present method achieved superior clustering performance on all eight machine learning datasets compared to the other three.

[0115] Table 2 Clustering results of the proposed method and three other methods on eight machine learning datasets

[0116]

[0117]

[0118]

[0119] In summary, the method proposed in the present invention achieves excellent results compared with the three comparison methods both on cancer datasets and conventional machine learning datasets.

[0120] Furthermore, based on the cancer subtype detection method based on multi-view clustering under the diversity criterion provided in Example 1, cancer sample data is detected to obtain cancer subtype clustering results.

[0121] A second aspect of the present invention provides an electronic device, such as Figure 3 As shown, Figure 3 The electronic device of the present invention is a structural diagram of an electronic device, which includes a processor 40, a memory 41, an input device 42, an output device 43 and a communication device 44; the number of processors 40 in the computer device can be one or more, Figure 2 Take a processor 40 in the electronic device as an example; the processor 40, memory 41, input device 42 and output device 43 in the electronic device can be connected by a bus or other means. Figure 3 The bus connection is taken as an example.

[0122] Memory 41, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules. Processor 40 executes the software programs, instructions, and modules stored in memory 41 to perform various electronic device functions and data processing, thereby implementing the cancer subtype detection method based on multi-view clustering under a diversity criterion according to any of the above-described embodiments of the present invention.

[0123] The memory 41 may primarily include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function; the data storage area may store data created based on the use of the terminal, etc. Furthermore, the memory 41 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state memory device. In some instances, the memory 41 may further include a memory remotely located relative to the processor 40, and these remote memories may be connected to the electronic device via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0124] The input device 23 can be used to receive the ID and password input by the user, etc. The output device 24 is used to output the network configuration page.

[0125] In a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the method of any embodiment of the present invention. The storage medium may be ROM / RAM, a magnetic disk, an optical disk, or the like.

[0126] According to a fifth aspect of the present invention, a computer program product is provided. When the computer program product is run on a computer, the computer is enabled to execute the method of any one of the above embodiments of the present invention.

[0127] The background technology references involved are as follows:

[0128] [1]Shi

[0129] [2]Wu D,Wang D,Zhang M Q,et al.Fast dimension reduction andintegrative clustering of multi-omics data using low-rank approximation:application to cancer molecular classification[J].BMC Genomics,2015,16:1-10。

[0130] [3]Liu C,Cao W,Wu S,et al.Supervised graph clustering for cancersubtyping based on survival analysis and integration of multi-omic tumor data[J].IEEE / ACM Transactions on Computational Biology and Bioinformatics,2020,19(2):1193-1202。

[0131] [4]Zhang,C.,Fu,H.,Liu,S.,Liu,G.,&Cao,X.(2015).Low-Rank TensorConstrained Multiview Subspace Clustering.2015IEEE International Conferenceon Computer Vision(ICCV),1582-1590。

[0132] [5]Wang B,Mezlini A M,Demir F,et al.Similarity network fusion foraggregating data types on a genomic scale[J].Nature methods,2014,11(3):333-337。

[0133] [6] Liu H, Shang M, Zhang H, et al. Cancer subtype identification based on multi-view subspace clustering with adaptive local structure learning [C] / / 2021IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE, 2021: 484-490.

[0134] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cancer subtype detection method based on multi-view clustering under diversity criteria, characterized by: include: Acquire a data matrix of each view in the multi-view data, and represent the data matrix using a representation matrix of the data of each view and a self-representation error matrix; Based on the representation matrix, a representation tensor of the multi-view data is obtained, and the representation tensor is split into a clean representation tensor and a representation error matrix; based on the self-representation error matrix and the representation error matrix, a noise matrix of the data matrix is ​​obtained, and the noise matrix is ​​expressed as , It is The self-representation error matrix of the views, It is The representation error matrix of the representation tensor of each view; Norm constraints are imposed on the clean representation tensor and the noise matrix respectively, and regularization terms and difference terms of the representation matrix are introduced to construct an optimization objective function for multi-view clustering; the difference term of the representation matrix is ​​an orthogonal difference term, and the regularization term of the representation matrix is ​​a graph-based Laplace regularization term; The optimization objective function is expressed as: in, 、 、 Respectively represent The representation matrix, clean representation matrix and noise matrix of each view; The first term of the objective function is the weighted norm of the clean representation vector, which is used to fuse the prior information of the singular value and induce the low rank of the tensor; the second term is the noise matrix norm, used to reduce the impact of noise in the representation matrix and tensor; the third term is Laplace manifold regularization, which is used to maintain the local structure of the data. It is The Laplacian matrix of the views, is the number of views, represents the trace of the matrix; the fourth term is the orthogonal difference term, which is used to represent the diversity of multiple views. It is The representation matrix of each view, ; 、 、 is the balance coefficient; The first constraint is used to perform subspace clustering based on self-representation. It is The data matrix of the views, It is The representation matrix of each view, It is The self-representation error matrix of each view; the second term uses the representation matrix of multiple views as the representation tensor; the third term divides the representation tensor into a clean part and an error part, is the clean representation tensor, It is The representation error matrix of the representation tensor of each view; the fourth term is the noise matrix constraint; Solving the optimization problem of the optimization objective function to obtain a solution result of the clean representation tensor; Calculating an affinity matrix of the multi-view data according to a solution result of the clean representation tensor; Perform spectral clustering on the affinity matrix to obtain a clustering result.

2. The method for detecting cancer subtypes based on multi-view clustering under diversity criteria according to claim 1, characterized in that: The data matrix is ​​represented by the representation matrix of each view data and the self-representation error matrix, specifically: in, It is The data matrix of the views, It is The representation matrix of each view, It is The self-representation error matrix of the views.

3. The method for cancer subtype detection based on multi-view clustering under diversity criteria according to claim 2, characterized in that: The representation tensor of the multi-view data is: in, It is The representation matrix of each view, , is the number of views, For tensor operations.

4. The method for detecting cancer subtypes based on multi-view clustering under diversity criteria according to claim 3, characterized in that: The splitting of the representation tensor into a clean representation tensor and a representation error matrix is ​​specifically: in, is the representation tensor of multi-view data, is the clean representation tensor, is the error matrix.

5. The method for detecting cancer subtypes based on multi-view clustering under diversity criteria according to claim 1, characterized in that: The graph-based Laplace regularization term of the representation matrix is ​​expressed as: ; in, Indicates the The Laplacian matrix of the views, Indicates the The representation matrix of each view, yes The transposed matrix of represents the trace of the matrix.

6. The method for cancer subtype detection based on multi-view clustering under diversity criteria according to claim 1, characterized in that: The orthogonal difference terms of the representation matrix are expressed as: ; in, It is The representation matrix of each view, It is The representation matrix of each view, , It represents the trace of the matrix, and the larger the value of the difference term is, the greater the diversity between views is.

7. The method for cancer subtype detection based on multi-view clustering under diversity criteria according to claim 1, characterized in that: The affinity matrix of the multi-view data is expressed as: in, represents the affinity matrix, is the solution of the clean representation tensor, yes The transpose of Indicates the Views, is the number of views.

8. 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 computer program, the method for detecting cancer subtypes based on multi-view clustering under a diversity criterion according to any one of claims 1 to 7 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting cancer subtypes based on multi-view clustering under a diversity criterion according to any one of claims 1 to 7 is implemented.

Images