Cancer category detection method based on multi-omics data clustering remarking

Abstract

The invention discloses a cancer category detection method based on multi-omics data clustering remarking, and relates to the technical field of machine learning. The learning process of the category detection model is as follows: n multi-omics data samples are acquired, each sample comprises m different categories of omics data, sample labels are cancer categories corresponding to the samples, and K cancer categories are provided; preprocessing the sample, and generating sample features for the preprocessed sample by using the model; according to sample features, clustering the preprocessed samples into K clusters; performing impurity measurement on each cluster, and re-marking sample labels in the clusters; performing comparative learning based on a re-marking result to enable the model to generate accurate sample features; and according to accurate sample characteristics generated by the model, predicting cancer types. According to the invention, cancer category detection is carried out by using the category detection model, and the accuracy of cancer typing is improved through reconstruction of the sample labels and application of a contrast learning method.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and in particular to a method for detecting cancer categories based on multi-omics data clustering and re-labeling. Background Art

[0002] Cancer classification detection methods based on multi-omics data aim to analyze cancer samples using a variety of biological information (such as DNA methylation, mRNA expression, miRNA expression, etc.) to reveal the biological differences between different subtypes. This method can help us better understand the pathogenesis and prognosis of cancer and guide personalized treatment. The existing multi-omics fusion algorithm mainly includes the following steps: First, normalize each omics data. Then, fuse multiple omics data of the same sample, and then perform similarity measurement to obtain a similarity matrix for constructing multi-omics data. Different types of omics data information are integrated in the calculation process of this similarity matrix, so that the degree of association between samples can be more comprehensively reflected. Then, the similarity matrices of different data types are fused, such as calculating the weighted average of each similarity matrix with other matrices to obtain an updated fused similarity matrix. Finally, a clustering algorithm is used to perform cluster analysis on the fused similarity matrix to divide the samples into different clusters or subtypes.

[0003] Cancer classification detection methods based on multi-omics data have the following disadvantages:

[0004] In the actual research process, since the training of the model relies on a large amount of manually labeled data, and the labeling process is highly subjective and uncontrollable, the manually labeled labels often have high uncertainty. This uncertainty has two disadvantages:

[0005] 1. Cancer samples of the same category will be classified into different label categories. For example, cancer samples of the same subtype may have different labels if they are labeled by different labelers.

[0006] 2. There are cancer samples that cannot be classified into known label categories. Labelers can only label according to preset label categories, but in the actual labeling process, there are often some samples that cannot be classified into known label categories.

[0007] In summary, label uncertainty has been one of the main factors that seriously affects the performance degradation of cancer category detection models. Summary of the invention

[0008] In order to overcome the defects in the above-mentioned prior art, the present invention provides a cancer category detection method based on multi-omics data clustering re-labeling, which is beneficial to improve the accuracy of cancer classification through the reconstruction of sample labels and the application of comparative learning methods.

[0009] To achieve the above object, the present invention adopts the following technical solutions, including:

[0010] A cancer category detection method based on multi-omics data clustering and re-labeling, using a category detection model to perform cancer category detection, the learning process of the category detection model includes the following steps:

[0011] S1, obtain n multi-omics data samples, each sample contains m different categories of omics data, the sample label is the cancer category corresponding to the sample, and there are K cancer categories in total; preprocess the samples, and use the model to generate sample features for the preprocessed samples;

[0012] S2, clustering the preprocessed samples into K clusters according to sample characteristics;

[0013] S3, measure the impurity of each cluster and relabel the sample labels within the cluster;

[0014] S4, based on the re-labeling results, conduct comparative learning to enable the model to generate accurate sample features;

[0015] S5, predict cancer category based on the accurate sample features generated by the model.

[0016] Preferably, the specific process of step S3 is as follows:

[0017] The cluster label R(c) of cluster c is:

[0018]

[0019] in, represents the sample probability of sample label y in cluster c; y represents the sample label; Y c represents the sample label set of all samples in cluster c, and the argmax function is used to obtain The sample label y corresponding to the maximum value;

[0020] Impurity level Imp of cluster c c for:

[0021]

[0022] Where C represents the total number of clusters;

[0023] Set the impurity lower limit θ respectively low and the impurity upper limit θ high :

[0024] If Imp c <θ low, then the samples in cluster c whose sample labels are different from the cluster label R(c) are considered to be samples with initial incorrect labels. The samples with initial incorrect labels are relabeled, and the sample labels are relabeled to R(c).

[0025] If θ low <Imp c <θ high , then the samples in cluster c whose sample labels are different from the cluster label R(c) are considered to be noise samples, and the noise samples are relabeled, and the sample labels are relabeled as noise sample labels.

[0026] If Imp c >θ high , then all samples in cluster c are relabeled, and the sample label is relabeled as a new unknown sample label.

[0027] Preferably, the comparative learning in step S4 is specifically as follows:

[0028] After re-labeling, samples whose sample labels are the same as the cluster labels are called accurate samples;

[0029] Randomly find two accurate samples and perform comparative learning on the two accurate samples. When the two accurate samples have the same sample label, that is, the same cluster label, the distance between the two accurate samples is minimized; when the two accurate samples have different sample labels, that is, different cluster labels, the distance between the two accurate samples is maximized;

[0030] Randomly find an accurate sample, compare the accurate sample with the noise sample, and maximize the distance between the accurate sample and the noise sample;

[0031] The distance refers to the distance between sample features.

[0032] Preferably, the formula for contrastive learning is:

[0033]

[0034] Where L represents the loss function of contrastive learning, and the purpose of contrastive learning is to minimize the loss function L; τ and q both represent coefficients; d + Represents the distance between the positive sample pairs corresponding to the current accurate sample, d i Represents the distance between any pair of samples corresponding to the current accurate sample;

[0035] The current accurate sample and the accurate sample with the same sample label constitute the positive sample pair; the current accurate sample and all other accurate samples or noise samples constitute the arbitrary sample.

[0036] Preferably, the preprocessing method of step S1 is specifically as follows:

[0037] n multi-omics data samples are denoted as in, is the nth sample, l is the category index of omics data, l = 1, 2, ..., m;

[0038] Normalize multi-omics data:

[0039]

[0040] in, is the normalized omics data of the lth category in the ith sample, is the l-th type of omics data in the i-th sample, and Respectively The expectation and variance of

[0041] The normalized omics data of different categories of the same sample are spliced ​​and fused to obtain the normalized sample, that is, the preprocessed sample:

[0042]

[0043] in, It represents the i-th sample after normalization or preprocessing.

[0044] Preferably, in step S2, clustering is performed using a K-means clustering algorithm, as shown below:

[0045] S21, select K value: set the number of clusters to K, corresponding to K cancer categories;

[0046] S22, initialization: samples with the same sample label are initialized into a cluster, the initialized cluster label is the same as the sample label, and the mean of the samples with the same sample label is calculated as the center of the initialized cluster; the mean is calculated using sample features;

[0047] S23, allocation: respectively calculating the distance between each sample and each cluster center, and assigning each sample to the cluster corresponding to the nearest cluster center; the distance is calculated using sample features and cluster center features;

[0048] S24, update: recalculate the cluster center of each cluster according to the current cluster assignment;

[0049] S25, repeating step S23 and step S24 until the cluster center no longer changes or the specified maximum number of iterations is reached.

[0050] Preferably, in the category detection model, the preprocessed samples Sample feature X extracted using CNN networki ; Use the embedding layer to embed the preprocessed samples The sample feature pairs (X i ,X j ) is mapped into a feature space and the feature vector pair (E i ,E j ); Use Multi-Layer Perceptron MLP to predict cancer category.

[0051] The present invention also provides a readable storage medium having a computer program stored thereon, and when the computer program is executed, the method for detecting cancer categories based on multi-omics data clustering and re-labeling is implemented.

[0052] The present invention also provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned cancer category detection method based on multi-omics data clustering and re-labeling is implemented.

[0053] The present invention also provides a computer program product, which includes a computer program / instruction, and when the computer program / instruction is executed by a processor, the above-mentioned cancer category detection method based on multi-omics data clustering and re-labeling is implemented.

[0054] The advantages of the present invention are:

[0055] (1) Through the reconstruction of sample labels and the application of contrastive learning methods, the model can learn strong discriminative features, which is conducive to improving the accuracy of cancer classification.

[0056] (2) During the clustering process, in addition to the manually labeled cancer sample labels, an additional cluster label (cluster label) is generated, and the sample labels of the samples are dynamically adjusted according to the clustering results. Based on the corrected sample labels, the cancer category prediction process of the model is further guided.

[0057] (3) The re-labeling method is introduced, which enables the model to better explore the intrinsic characteristics of the data itself during the clustering process and achieve more complex sample label reconstruction.

[0058] (4) Clustering, cluster impurity measurement, and relabeling can be used to subdivide the same cancer category and discover new subtypes. During the training process, some noisy samples may be assigned a sample label, but the samples will be re-identified as noise samples during the clustering iteration. In this case, cluster impurity measurement and relabeling can be used to discover the existence of a new cancer subtype. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1This is a flow chart of a cancer category detection method based on multi-omics data clustering and re-labeling of the present invention.

[0060] Figure 2 A schematic diagram of a cancer category detection method based on multi-omics data clustering and re-labeling according to the present invention.

[0061] Figure 3 It is a comparative learning schematic diagram of the present invention. DETAILED DESCRIPTION

[0062] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] Example 1

[0064] Depend on Figure 1 and Figure 2 As shown, a cancer category detection method based on multi-omics data clustering and re-labeling of the present invention comprises the following steps:

[0065] S1, multi-omics data preprocessing: normalize the collected multi-omics data, and perform multi-omics data fusion and feature extraction.

[0066] Given n omics samples, each sample contains m different types of omics data (such as DNA methylation, mRNA expression, miRNA expression, etc.); n samples are denoted as {x i |i=1,2,...,n}, Among them, x i is the i-th sample, and l is the category index of the omics data.

[0067] First, the multi-omics data is normalized and formalized as follows:

[0068]

[0069] in, is the normalized omics data of the lth category in the ith sample, is the l-th type of omics data in the i-th sample, and Respectively The expectation and variance of .

[0070] Then, the normalized omics data of different categories of the same sample are spliced ​​and fused to obtain the normalized sample, which is formalized as follows:

[0071]

[0072] in, represents the normalized i-th sample.

[0073] The normalized samples Extract features through CNN network and get samples Features of X i ; Then the normalized sample pair The feature representation X i and X j , through an embedding layer, mapped to a feature space, and the corresponding feature vector pair (E i ,E j ).

[0074] S2, sample clustering: Use the K-means clustering algorithm to cluster the normalized samples and generate corresponding clustering scores.

[0075] Assume that there are K cancer categories in total, each sample has a sample label, i.e., the cancer category, and the initial cluster label is the same as the sample label. The K-means algorithm is used to generate a clustering score for each sample in the cluster, which reflects the distance between the sample and the cluster center.

[0076] Although each sample has a unique sample label, i.e., cancer category, which is marked by a human, there are widespread noise labels with low quality or wrong labels due to human uncontrollability. This application uses a clustering method to help exclude or correct these low-quality labels, find the true sample label, i.e., the true cancer category, corresponding to the sample, thereby improving the label quality and further improving the accuracy of cancer category prediction.

[0077] The purpose of clustering in this application is to give samples an additional cluster label (i.e., cluster label) in addition to the sample label, and through a series of steps and algorithms, find samples with lower confidence based on the clustering results, and give the algorithm some ability to correct sample labels.

[0078] The K-means clustering algorithm clusters data points within a cluster by minimizing the sum of the squares of the distances between each sample in the cluster and the cluster center. The cluster center is the average value of all points (samples) in the cluster, indicating the center position of the cluster. The K-means clustering algorithm gradually converges by continuously updating cluster assignments and cluster centers.

[0079] The process of the K-means clustering algorithm includes two main steps: allocation and update, as shown below:

[0080] S21, select K value: set the number of clusters.

[0081] S22, initialization: samples with the same sample label are initialized into a cluster, the initialized cluster label is the same as the sample label, and the mean of samples with the same sample label is calculated as the initialized cluster center, with a total of K cluster centers; the mean is calculated using sample features.

[0082] S23, allocation: for each sample in the data set, respectively calculate the distance between each sample and each cluster center, and allocate each sample to the cluster corresponding to the nearest cluster center; the distance is calculated using the features of the sample and the features of the cluster center.

[0083] S24, update: according to the current cluster assignment, recalculate the cluster center of each cluster, that is, calculate the mean of all samples in the cluster as the new cluster center.

[0084] S25, repeating step S23 and step S24, and continuously repeating the steps of allocating and updating until the cluster center no longer changes (ie, converges) or the specified maximum number of iterations is reached.

[0085] S3, cluster impurity measurement and re-labeling: cluster analysis is performed on omics data using clustering algorithms, samples with close distances are divided into the same cluster, and corresponding sample labels and cluster labels are generated. Automatic re-labeling is performed according to the corresponding cluster labels, and the sample labels corresponding to the samples are modified.

[0086] Furthermore, we can check whether there are subclusters in cluster c. A subcluster means that there are two subcluster centers that are far apart in the same cluster, that is, the distance between the two subcluster centers is greater than the set threshold, and the samples in the subcluster satisfy the Gaussian distribution. For example, we find two subclusters c1 and c2 in cluster c through the K-means clustering algorithm. If the cluster center distance between the two subclusters c1 and c2 is greater than the set threshold, then the two subclusters c1 and c2 exist in cluster c. Otherwise, the two subclusters c1 and c2 do not exist in cluster c.

[0087] When the distance between the samples in cluster c (or sub-clusters c1, c2) and the cluster center exceeds the preset threshold δ, they are judged as outlier samples (noise samples). belongs to cluster c, and the sample The distance μ from the cluster center c When it is greater than the threshold δ, the sample is judged is a noise sample.

[0088] The impurity level of a cluster is an important metric to determine the relabeling strategy for samples within the cluster. If the impurity level is low, it means that the cluster is mainly composed of samples of one cancer category, then the cluster is identified as a group representing the cancer category, and the cluster label of the cluster is the cancer category. However, if the impurity level is high, it means that the cluster is composed of samples of different cancer categories in similar proportions, then the cluster cannot represent the cancer category. Noise samples are not considered in the cluster impurity measurement and relabeling stage. For noise samples within the cluster, a new sample label, namely the noise sample label, is assigned.

[0089] The cluster label R(c) of cluster c is expressed as:

[0090]

[0091] in, represents the sample probability of sample label y in cluster c; y represents the sample label; Y c represents the sample label set of all samples in cluster c, and the argmax function is used to obtain The sample label y corresponding to the maximum value.

[0092] Impurity level Imp of cluster c c The metrics are expressed as:

[0093]

[0094] Where C represents the total number of clusters.

[0095] Set the impurity lower limit θ respectively low and the impurity upper limit θ high , the impurity level of the cluster is divided into three levels, and different re-labeling strategies are adopted for clusters of each level:

[0096] If Imp c <θ low , then the impurity level of cluster c is low impurity, and the samples in cluster c whose sample labels are different from the cluster label R(c) are considered to be samples with initial incorrect labels, and the samples with initial incorrect labels are relabeled, and the sample labels are relabeled to R(c).

[0097] If θ low <Imp c <θ high , then the impurity level of cluster c is medium impurity. In order to minimize the error re-labeling, the samples in cluster c whose sample labels are different from the cluster label R(c) are re-labeled, and the sample labels are re-labeled as noise sample labels.

[0098] If Imp c >θ high, then the impurity level of cluster c is high impurity, all samples in cluster c are relabeled, and the sample label is relabeled as a new unknown sample label.

[0099] S4, contrastive learning: Based on the results of re-labeling, the sample features of accurately labeled sample labels and cluster labels are screened for contrastive learning to promote the model to generate more accurate sample features.

[0100] Depend on Figure 3 As shown, samples with the same sample label and cluster label are called accurate samples; two accurate samples are randomly found and compared and learned. When the two accurate samples have the same sample label, that is, the same cluster label (the two accurate samples are called positive sample pairs), the distance between the two accurate samples is minimized; when the two accurate samples have different sample labels, that is, different cluster labels (the two accurate samples are called negative sample pairs), the distance between the two accurate samples is maximized.

[0101] Randomly find an accurate sample, compare the accurate sample with the noise sample, and maximize the distance between the accurate sample and the noise sample.

[0102] The distance is calculated by using Euclidean distance or cosine distance to calculate the sample features.

[0103] The formula for contrastive learning is as follows:

[0104]

[0105] Where L represents the loss function of contrastive learning. The purpose of contrastive learning is to minimize the loss function L; τ is the temperature coefficient of contrastive learning, which is set to 1 by default; q is the weight coefficient, which is set to 1 by default; d + represents the distance between the positive sample pair corresponding to the current sample, and the current accurate sample and the accurate sample with the same sample label constitute the positive sample pair; d i Represents the distance between any pair of samples corresponding to the current sample. The current accurate sample and all other accurate samples or noise samples constitute the arbitrary sample; K represents the number of labels.

[0106] S5, Cancer Category Prediction: Predict the final category of cancer based on the more accurate sample features generated by the model.

[0107] The more accurate sample features generated by the model are input into a classifier (such as a multi-layer perceptron MLP) to predict the category of cancer. The prediction process can be formalized as follows:

[0108] S = σ(MLP(X;θ MLP ))

[0109] Among them, σ is a nonlinear activation function, such as the softmax function, which maps the output of the classifier to the interval [0,1]. MLP represents the learnable weight parameters of the classifier, and X is a more accurate sample feature generated by the model.

[0110] Example 2

[0111] An electronic device comprises a processor, a memory and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, a cancer category detection method based on multi-omics data clustering and re-labeling according to the above-mentioned embodiment 1 is implemented.

[0112] The electronic device of the embodiment of the present application can be the mobile device itself, or a stand-alone device independent of the mobile device, which can communicate with the mobile device to receive the collected input signals from them and send the selected target decision behavior to them.

[0113] The electronic device includes one or more processors and memory. The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may run the program instructions to implement the decision behavior decision method and / or other desired functions of the various embodiments of the present application described above.

[0114] Electronic devices may also include input devices and output devices.

[0115] Example 3

[0116] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the decision-making behavior decision method according to various embodiments of the present application described in the above-mentioned embodiment 1 of this specification.

[0117] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0118] Example 4

[0119] An embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the decision-making behavior decision method according to various embodiments of the present application described in the above-mentioned embodiment 1.

[0120] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0121] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A cancer category detection method based on multi-omics data clustering and re-labeling, characterized in that: Cancer category detection is performed using a category detection model. The learning process of the category detection model includes the following steps: S1, obtain n multi-omics data samples, each sample contains m different categories of omics data, the sample label is the cancer category corresponding to the sample, and there are K cancer categories in total; preprocess the samples, and use the model to generate sample features for the preprocessed samples; S2, clustering the preprocessed samples into K clusters according to sample characteristics; S3, measure the impurity of each cluster and relabel the sample labels within the cluster; S4, based on the re-labeling results, conduct comparative learning to enable the model to generate accurate sample features; S5, predict cancer category based on the accurate sample features generated by the model.

2. A cancer category detection method based on multi-omics data clustering and re-labeling according to claim 1, characterized in that: The specific process of step S3 is as follows: The cluster label R(c) of cluster c is: in, represents the sample probability of sample label y in cluster c; y represents the sample label; Y c represents the sample label set of all samples in cluster c, and the argmax function is used to obtain The sample label y corresponding to the maximum value; Impurity level Imp of cluster c c for: Where C represents the total number of clusters; Set the impurity lower limit θ respectively low and the impurity upper limit θ high : If Imp c <θ low , then the samples in cluster c whose sample labels are different from the cluster label R(c) are considered as samples with initial incorrect labels, and the samples with initial incorrect labels are relabeled, and the sample labels are relabeled to R(c); If θ low <Imp c <θ high , then the samples in cluster c whose sample labels are different from the cluster label R(c) are considered to be noise samples, and the noise samples are relabeled, and the sample labels are relabeled as noise sample labels; If Imp c >θ high , then all samples in cluster c are relabeled, and the sample label is relabeled as a new unknown sample label.

3. A method for detecting cancer categories based on multi-omics data clustering and re-labeling according to claim 2, characterized in that: The comparative learning of step S4 is specifically as follows: After re-labeling, samples whose sample labels are the same as the cluster labels are called accurate samples; Randomly find two accurate samples and perform comparative learning on the two accurate samples. When the two accurate samples have the same sample label, that is, the same cluster label, the distance between the two accurate samples is minimized; when the two accurate samples have different sample labels, that is, different cluster labels, the distance between the two accurate samples is maximized; Randomly find an accurate sample, compare the accurate sample with the noise sample, and maximize the distance between the accurate sample and the noise sample; The distance refers to the distance between sample features.

4. A method for detecting cancer categories based on multi-omics data clustering and re-labeling according to claim 3, characterized in that: The formula for contrastive learning is: Where L represents the loss function of contrastive learning, and the purpose of contrastive learning is to minimize the loss function L; τ and q both represent coefficients; d + Represents the distance between the positive sample pairs corresponding to the current accurate sample, d i Represents the distance between any pair of samples corresponding to the current accurate sample; The current accurate sample and the accurate sample with the same sample label constitute the positive sample pair; the current accurate sample and all other accurate samples or noise samples constitute the arbitrary sample.

5. A method for detecting cancer categories based on multi-omics data clustering and re-labeling according to any one of claims 1 to 3, characterized in that: The preprocessing method of step S1 is specifically as follows: n multi-omics data samples are denoted as in, is the nth sample, l is the category index of omics data, l = 1, 2, ..., m; Normalize multi-omics data: in, is the normalized omics data of the lth category in the ith sample, is the l-th type of omics data in the i-th sample, and Respectively The expectation and variance of The normalized omics data of different categories of the same sample are spliced ​​and fused to obtain the normalized sample, that is, the preprocessed sample: in, It represents the i-th sample after normalization or preprocessing.

6. A method for detecting cancer categories based on multi-omics data clustering and re-labeling according to any one of claims 1 to 3, characterized in that: In step S2, clustering is performed using the K-means clustering algorithm, as shown below: S21, select K value: set the number of clusters to K, corresponding to K cancer categories; S22, initialization: samples with the same sample label are initialized into a cluster, the initialized cluster label is the same as the sample label, and the mean of the samples with the same sample label is calculated as the center of the initialized cluster; the mean is calculated using sample features; S23, allocation: calculate the distance between each sample and each cluster center respectively, and allocate each sample to the cluster corresponding to the nearest cluster center; The distance is calculated using sample features and cluster center features; S24, update: recalculate the cluster center of each cluster according to the current cluster assignment; S25, repeating step S23 and step S24 until the cluster center no longer changes or the specified maximum number of iterations is reached.

7. A method for detecting cancer categories based on multi-omics data clustering and re-labeling according to any one of claims 1 to 3, characterized in that: In the category detection model, the preprocessed samples Sample feature X extracted using CNN network i ; Use the embedding layer to embed the preprocessed samples The sample feature pairs (X i ,X j ) is mapped into a feature space and the feature vector pair (E i ,E j ); Use Multi-Layer Perceptron MLP to predict cancer category.

8. A readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed, the method for detecting cancer categories based on multi-omics data clustering and re-labeling according to any one of claims 1 to 7 is implemented.

9. An electronic device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a cancer category detection method based on multi-omics data clustering and re-labeling as described in any one of claims 1 to 7.

10. A computer program product, characterized in that It includes a computer program / instruction, which, when executed by a processor, implements a cancer category detection method based on multi-omics data clustering and re-labeling as described in any one of claims 1 to 7.

Images