A cancer subtype classification method based on KAN network and multi-omics data

Abstract

The present invention provides a cancer subtype classification method based on a KAN network and multi-omics data, the method comprising obtaining multi-omics data, including genomic data, transcriptomics and proteomics data, and constructing a test data set; using a Kolmogorov-Arnold network model to extract and integrate features of the multi-omics data, and training to obtain a trained cancer typing model; in the model evaluation stage, using a Monte Carlo dropout method to perform multiple forward propagations to estimate the uncertainty of the model prediction results and improve the robustness of the model; cancer typing prediction, using a trained Kolmogorov-Arnold network model to perform high-precision cancer typing on multi-omics data. This method can effectively solve the problem of heterogeneity of multi-omics data, improve the accuracy and robustness of cancer typing, and help improve the effect of personalized treatment.

Description

Technical Field

[0001] The present invention relates to the fields of bioinformatics and data science, and in particular to a cancer subtype classification method based on a KAN network and multi-omics data. Background Art

[0002] With the development of high-throughput sequencing technology, a large amount of multi-omics data, such as genomics, transcriptomics, and proteomics, has been generated in the biomedical field. These data have important application value in the early diagnosis, personalized treatment, and prognosis assessment of diseases. However, how to effectively integrate and analyze these data has become a major challenge in current research. Traditional single-omics data analysis methods often find it difficult to fully capture the correlation information between different data types, resulting in insufficient information utilization and affecting the accuracy and reliability of cancer typing. Therefore, there is an urgent need for a new method to efficiently integrate and analyze multi-omics data to improve the accuracy and reliability of cancer typing.

[0003] In view of this, this application is filed. Summary of the invention

[0004] The present invention provides a cancer subtype classification method based on a KAN network and multi-omics data, which can at least partially improve the above problems.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A cancer subtype classification method based on a KAN network and multi-omics data, comprising:

[0007] Acquire multi-omics data, and construct a test data set according to the multi-omics data, wherein the multi-omics data includes genomics data, transcriptomics data, and proteomics data;

[0008] Using a preset Kolmogorov-Arnold network multi-omics data integration model to perform feature extraction, integration and training processing on the multi-omics data to obtain a trained cancer typing model;

[0009] The cancer classification model is evaluated by performing multiple forward propagations using the Monte Carlo dropout method to estimate the uncertainty of the prediction results of the cancer classification model to improve the robustness of the model and obtain the final Kolmogorov-Arnold network model;

[0010] The Kolmogorov-Arnold network model is used to perform high-precision cancer typing processing on the multi-omics data to obtain a classification result.

[0011] In summary, the cancer subtype classification method based on KAN network and multi-omics data includes obtaining multi-omics data, including genomic data, transcriptomics and proteomics data, and constructing a test data set; using the Kolmogorov-Arnold network model to extract and integrate features of the multi-omics data, and training to obtain a trained cancer classification model; in the model evaluation stage, the Monte Carlo dropout method is used for multiple forward propagations to estimate the uncertainty of the model prediction results and improve the robustness of the model; cancer classification prediction, using the trained Kolmogorov-Arnold network model to perform high-precision cancer classification on multi-omics data. This method can effectively solve the problem of heterogeneity of multi-omics data, improve the accuracy and robustness of cancer classification, and help improve the effect of personalized treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a schematic flow chart of a cancer subtype classification method based on a KAN network and multi-omics data provided by the first aspect of the present invention;

[0013] Figure 2 is a schematic flow chart of a cancer subtype classification method based on a KAN network and multi-omics data provided in the second aspect of the present invention;

[0014] Figure 3 is a flowchart for implementing training of a multi-omics cancer data typing model based on a Kolmogorov-Arnold network provided by an embodiment of the present invention;

[0015] Figure 4 It is the evaluation index result on the test data set provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] refer to Figure 1 , Figure 2 , Figure 3 As shown, the first embodiment of the present invention discloses a cancer subtype classification method based on a KAN network and multi-omics data, which can be performed by a multi-omics cancer data typing device based on a KAN network (hereinafter referred to as a typing device), and in particular, is performed by one or more processors in the typing device to implement the following method:

[0018] S1, acquiring multi-omics data, and constructing a test data set according to the multi-omics data, wherein the multi-omics data includes genomics data, transcriptomics data, and proteomics data;

[0019] Specifically, step S1 includes: obtaining the patient's genomic data, transcriptomic data, and proteomic data on a preset cancer genome atlas website;

[0020] The patient's genomic data, transcriptomics and proteomics data are preprocessed, wherein the data preprocessing includes data standardization and batch processing to obtain a test data set, wherein the test data set includes 80% training set, 10% validation set, and 10% test set.

[0021] In this embodiment, the genomic data, transcriptomic data and proteomic data of patients are obtained from the Cancer Genome Atlas website. The data preprocessing process includes data standardization and batch processing operations to ensure the consistency and stability of the data during the training process. 80% of the data is used as a training set, 10% as a validation set, and 10% as a test set. These data are converted into PyTorch tensors for subsequent model processing.

[0022] In this embodiment, multi-omics data provides rich information for revealing complex biological processes and disease mechanisms. It has now become an important data source for cancer analysis. Existing models are difficult to solve the data heterogeneity and computational efficiency problems faced in the process of multi-omics data integration. Data heterogeneity refers to the fact that different omics data have different data characteristics and noise levels due to different sources, measurement methods, and scales, which makes data integration complex and challenging.

[0023] S2, using a preset Kolmogorov-Arnold network multi-omics data integration model to perform feature extraction, integration and training processing on the multi-omics data to obtain a trained cancer typing model;

[0024] Specifically, step S2 includes: constructing a multi-omics data integration model based on the Kolmogorov-Arnold network, and performing data processing on the training set;

[0025] The multi-omics data integration model based on the Kolmogorov-Arnold network is trained according to the training set after data processing to obtain a trained cancer typing model.

[0026] Preferably, data processing is performed on the training set, specifically:

[0027] Initialize the training set, set the data loading path, data processing method and parameters;

[0028] Load different types of omics data from the training set and convert them into PyTorch tensors;

[0029] Each type of omics data was standardized separately, and the data were normalized according to omics characteristics to ensure that the numerical range of different omics data was consistent;

[0030] Perform batch processing on the standardized tensor data to end data processing.

[0031] Preferably, the multi-omics data integration model based on the Kolmogorov-Arnold network is trained according to the training set after data processing to obtain a trained cancer typing model, specifically:

[0032] The training set after data processing is input into the Kolmogorov-Arnold network model, and the input data is represented as ,in, represents the i-th omics data, N is the number of samples, is the characteristic dimension of the i-th omics data;

[0033] The training set after data processing passes through the first Kolmogorov-Arnold network layer, which performs preliminary feature extraction based on the learnable edge weight activation function of the B-spline basis function to generate an intermediate feature representation ,in, is the learnable edge weight activation function, and are the weight and bias parameters of the model;

[0034] The intermediate feature is represented as Input into the subsequent multiple Kolmogorov-Arnold network layers for hierarchical feature extraction and nonlinear transformation to update the feature representation , the expression is: ;

[0035] The updated final feature representation is linearly mapped through a fully connected layer and converted to the output layer to generate the result for classification. ,in, and are the weights and biases of the fully connected layer, and L is the final layer of the neural network;

[0036] Among them, during the model training process, the cross entropy loss function is used To optimize the classification task, the mathematical expression is: , and represent the true label and predicted probability respectively.

[0037] In this embodiment, a multi-omics data integration model based on the Kolmogorov-Arnold network is constructed, the data of the training data set is loaded and the data is processed. The constructed multi-omics data integration model is trained with the processed data to obtain a trained multi-omics data integration model.

[0038] Specifically, the data of the training dataset is loaded and processed, including: initializing the loading of the training dataset data, setting the data loading path, data processing method and parameters, such as batch size, data size and data type. Different types of omics data (such as genomics, transcriptomics and proteomics data) are loaded from the training dataset and converted into PyTorch tensors. Each type of omics data is standardized separately, and the data is normalized according to the omics features to ensure that the numerical range of different omics data is consistent. Batch processing is performed on the standardized tensor data to form training batches for subsequent training steps.

[0039] Specifically, the constructed multi-omics data integration model is trained with the processed data to obtain a trained multi-omics data integration model, which specifically includes: data input and preliminary processing, and the multi-omics data (e.g., genomics, transcriptomics, and proteomics data) loaded and processed from the training data set are input into the Kolmogorov-Arnold network model. The input data is represented as ,in represents the i-th omics data, N is the number of samples, is the feature dimension of the i-th omics data. Feature extraction and nonlinear transformation, the data passes through the first Kolmogorov-Arnold network layer, which applies a learnable edge weight activation function based on B-spline basis function to perform preliminary feature extraction and generate intermediate feature representation. Let the output of this Kolmogorov-Arnold network layer be: ,in, is the learnable edge weight activation function, and are the weight and bias parameters of the model.

[0040] Perform multi-layer feature cascading and optimization to represent the intermediate features Input to the subsequent multiple Kolmogorov-Arnold network layers for hierarchical feature extraction and nonlinear transformation. In each Kolmogorov-Arnold network layer, the feature representation Updated via: The output features of these layers are processed layer by layer, and residual connections are used to ensure smooth information transmission and prevent gradient disappearance. Feature concatenation and full connection mapping, the final feature representation is linearly mapped through a fully connected layer and converted to the output layer. The output layer produces the result for classification: ,in, and are the weight and bias of the fully connected layer respectively.

[0041] Finally, perform model training and loss calculation. During model training, the cross entropy loss function is used. To optimize the classification task, it is defined as follows: ,in, and Denote the true label and predicted probability respectively. The AdamW optimizer is used to update the model parameters, the learning rate is set to 0.01, and the learning rate decay mechanism is used to gradually reduce the learning rate.

[0042] S3, evaluating the cancer classification model, performing multiple forward propagations using a Monte Carlo dropout method, estimating the uncertainty of the prediction results of the cancer classification model, so as to improve the robustness of the model, and obtaining the final Kolmogorov-Arnold network model;

[0043] Specifically, step S3 includes: for each input sample of the cancer typing model, performing T forward propagations to obtain T prediction results ;

[0044] Calculate the mean of the input samples and variance As the prediction result and uncertainty estimate, the calculation formula is: , ,in, is the tth input sample;

[0045] According to the loss function, it is back-propagated into the cancer classification model, the model parameters are adjusted, and the model with the smallest loss is saved as the final Kolmogorov-Arnold network model.

[0046] In this embodiment, the Monte Carlo dropout method is used to perform multiple forward propagations to estimate the uncertainty of the model prediction results. For each input sample, T forward propagations are performed to obtain T prediction results. , and calculate the mean and variance As forecast results and uncertainty estimates: , , back-propagates into the model according to the loss function, adjusts the model parameters, and saves the model with the smallest loss as the final trained model.

[0047] S4, using the Kolmogorov-Arnold network model to perform high-precision cancer typing processing on the multi-omics data to obtain a classification result.

[0048] In this embodiment, the classification result is generated by the output layer of the model, and the effectiveness of the method of the present invention is verified by comparing the classification performance of different methods.

[0049] Specifically, in the present embodiment, real multi-omics cancer data is used for testing, and data is summarized from the cancer genome atlas, a total of 612 samples. For all experiments, training and testing are carried out under the same hardware environment. The cancer sample data collected on the cancer genome atlas website used by the training data set, this data set contains three kinds of group information of 612 breast cancer patients, including messenger RNA (messenger RNA, mRNA), micro RNA (microRNA, miRNA) and methylation (methylation, methy). It is divided into five categories, and the category and the corresponding number of samples are respectively Normal-like: 80, Basal-like: 92, HER2-enriched: 32, Luminal A:305, Luminal B: 103. The present embodiment uses Adam optimizer for training, and the initial learning rate is 0.01, and the batch size is 32. This example uses four common evaluation indicators: Accuracy (ACC), Mean Average Precision (MAP), Mean Average Recall (MAR), and Mean Average F1-score (MAF) to measure the performance of the model.

[0050] Figure 4 The figure shows the evaluation index of the method in this embodiment on the test data set, with the horizontal axis representing different model algorithms and the bottom row representing the method. Figure 4 It can be seen that all indicators of this method have reached relatively high values, proving that this method can better perform cancer classification.

[0051] In summary, compared with the prior art, the cancer subtype classification method based on KAN network and multi-omics data is used to integrate and analyze multi-omics data, and is particularly suitable for cancer subtype classification. This method achieves efficient integration of multi-omics data by using learnable edge weight activation functions and flexible network structures, thereby improving the accuracy and stability of the model. This method also introduces a sample weight random sampler and a Monte Carlo dropout method to deal with the imbalance of the data and estimate the confidence of the model. Experimental results show that the proposed method is superior to traditional methods in classification accuracy and stability, and can better capture the association information between different omics data. This method achieves efficient integration of multi-omics data and significantly improves the accuracy of cancer typing. Through experimental verification, this method shows excellent performance and performs better than traditional methods in classification tasks.

[0052] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A cancer subtype classification method based on KAN network and multi-omics data, characterized in that: include: Acquire multi-omics data, and construct a test data set according to the multi-omics data, wherein the multi-omics data includes genomics data, transcriptomics data, and proteomics data; Using a preset Kolmogorov-Arnold network multi-omics data integration model to perform feature extraction, integration and training processing on the multi-omics data to obtain a trained cancer typing model; The cancer classification model is evaluated by performing multiple forward propagations using the Monte Carlo dropout method to estimate the uncertainty of the prediction results of the cancer classification model to improve the robustness of the model and obtain the final Kolmogorov-Arnold network model; Using the Kolmogorov-Arnold network model to perform high-precision cancer typing processing on the multi-omics data to obtain a classification result; The multi-omics data is subjected to feature extraction, integration and training processing using a preset Kolmogorov-Arnold network multi-omics data integration model to obtain a trained cancer typing model, specifically: Construct a multi-omics data integration model based on the Kolmogorov-Arnold network and perform data processing on the training set; Training the multi-omics data integration model based on the Kolmogorov-Arnold network according to the training set after data processing to obtain a trained cancer typing model; The training set is subjected to data processing, specifically: Initialize the training set, set the data loading path, data processing method and parameters; Load different types of omics data from the training set and convert them into PyTorch tensors; Each type of omics data was standardized separately, and the data were normalized according to omics characteristics to ensure that the numerical range of different omics data was consistent; Perform batch processing on the standardized tensor data to end data processing; The multi-omics data integration model based on the Kolmogorov-Arnold network is trained according to the training set after data processing to obtain a trained cancer typing model, specifically: The training set after data processing is input into the Kolmogorov-Arnold network model, and the input data is represented as ,in, represents the i-th omics data, N is the number of samples, is the characteristic dimension of the i-th omics data; The training set after data processing passes through the first Kolmogorov-Arnold network layer, which performs preliminary feature extraction based on the learnable edge weight activation function of the B-spline basis function to generate an intermediate feature representation ,in, is the learnable edge weight activation function, and are the weight and bias parameters of the model; The intermediate feature is represented as Input into the subsequent multiple Kolmogorov-Arnold network layers for hierarchical feature extraction and nonlinear transformation to update the feature representation , the expression is: ; The updated final feature representation is linearly mapped through a fully connected layer and converted to the output layer to generate the result for classification. ,in, and are the weights and biases of the fully connected layer, and L is the final layer of the neural network; Among them, during the model training process, the cross entropy loss function is used To optimize the classification task, the mathematical expression is: , and Represent the true label and predicted probability respectively; The cancer classification model was evaluated by performing multiple forward propagations using the Monte Carlo dropout method to estimate the uncertainty of the prediction results of the cancer classification model, specifically: For each input sample of the cancer classification model, perform T forward propagations to obtain T prediction results ; Calculate the mean of the input samples and variance As the prediction result and uncertainty estimate, the calculation formula is: , ,in, is the tth input sample; According to the loss function, it is back-propagated into the cancer classification model, the model parameters are adjusted, and the model with the smallest loss is saved as the final Kolmogorov-Arnold network model.

2. The cancer subtype classification method based on KAN network and multi-omics data according to claim 1, characterized in that: Acquire multi-omics data, and construct a test data set based on the multi-omics data, specifically: Obtain patient genomic, transcriptomic, and proteomic data on the pre-defined Cancer Genome Atlas website; The patient's genomic data, transcriptomics and proteomics data are preprocessed, wherein the data preprocessing includes data standardization and batch processing to obtain a test data set, wherein the test data set includes 80% training set, 10% validation set, and 10% test set.

Images