A method for predicting the prognosis of lung cancer based on multi-omics data fusion

Abstract

This invention relates to a method for predicting the prognosis of lung cancer based on multi-omics data fusion, comprising: acquiring multi-omics data and survival data of lung cancer patients, and preprocessing the multi-omics data of each lung cancer patient; concatenating the multi-omics data of each lung cancer patient to obtain a multi-omics ensemble feature matrix for each training sample; mapping the multi-omics ensemble feature matrix of the training samples to a low-dimensional space using an autoencoder to obtain a low-dimensional multi-omics feature matrix of the training samples; performing feature selection on the low-dimensional multi-omics feature matrix of the training samples using a VSOEnetbag algorithm combining elastic networks and bagging ideas based on the survival data of lung cancer patients to obtain a salient expression feature matrix of the training samples; performing risk subtype clustering on the training samples using K-means clustering based on the salient feature expression matrix of the training samples to obtain subtype clustering results; using the subtype clustering results of the training samples as labels and the RNA-Seq expression profile data of lung cancer patients as independent variables to construct a multivariate Cox prognostic model; inputting the RNA-Seq expression profile data of the lung cancer patients to be tested into the multivariate Cox prognostic model to obtain the prognostic results of the lung cancer patients.

Description

Technical Field

[0001] This invention belongs to the fields of bioinformatics and computer science, and in particular relates to a method for predicting the prognosis of lung cancer based on multi-omics data fusion. Background Technology

[0002] Lung cancer is one of the most common malignant tumors worldwide and a significant threat to human health, with low prognosis and survival rates, severely impacting patients' quality of life and lives. Accurately predicting the prognostic risk of lung cancer patients is crucial for guiding clinical treatment decisions and improving survival rates. Traditional methods for predicting lung cancer prognosis mainly rely on clinical indicators such as age, tumor grade, and extent of surgical resection, but these indicators have limited predictive power. With the development of high-throughput sequencing technology, genomic data has been widely used in lung cancer prognostic research. Several predictive models based on molecular features such as gene expression and DNA methylation have been proposed. These models utilize machine learning to screen key prognostic features from high-dimensional molecular characteristics, significantly improving predictive accuracy.

[0003] Early lung cancer prediction models primarily used single-omics data, such as genomics, proteomics, and metabolomics, for tumor analysis and employed biochemical indicators for prediction. For example, HER2 and EGFR gene variants in genomics have been widely used in clinical testing, benefiting numerous patients. Xie et al. developed a lung cancer auxiliary diagnostic and prognostic assessment system based on exosomal protein biomarkers (Patent Publication No.: CN202411139237.8), which includes information acquisition, calculation, and diagnostic modules to improve the sensitivity and specificity of lung cancer diagnosis. SU et al. developed a prognostic risk scoring model for non-small cell lung cancer recurrence (Patent Publication No.: CN202410747151.7), constructing a risk scoring model using six genes (HLA-DOB, SPP1, TWIST2, etc.) and two clinical features, accurately distinguishing between high-risk and low-risk recurrence patients. Luo et al. proposed a method for detecting platelet markers in non-small cell lung cancer (NSCLC) (Patent Publication No.: CN202410299001.4). Based on differential gene analysis, they screened ten characteristic genes closely related to the prognosis of NSCLC, providing new clues for prognostic assessment. Wang et al. created a prognostic gene tag for radiotherapy in NSCLC (Patent Publication No.: CN202410846192.1), using LASSO logistic regression to construct gene tags based on five genes, including MDM2 and THBS1, to predict radiotherapy prognosis and support the development of personalized radiotherapy plans. Furthermore, Wang et al. developed a lung cancer malignancy risk stratification model (Patent Publication No.: CN202410793134.7), applying a Densenet-Self-Attention-DeepSurv neural network, combining medical imaging and clinical data to accurately predict the malignancy risk and prognosis of lung cancer. These systems and models provide important technical support for the diagnosis, prognostic assessment, and personalized treatment of lung cancer.

[0004] However, studies based on a single omics dataset cannot fully reveal the complete mechanisms of lung cancer, a complex disease. With advancements in bioscience and measurement techniques, researchers can acquire and analyze molecular data at many different levels, including genomics, transcriptomics, proteomics, and metabolomics—different "omics" datasets. These data reflect biological information at different levels, such as genes, RNA, proteins, and metabolites, offering the possibility of a comprehensive understanding of the pathogenic mechanisms of lung cancer. On the other hand, the occurrence and development of lung cancer is a complex process driven by the interaction of changes at multiple molecular levels. Highly interconnected regulatory relationships exist between molecules at different omics levels; for example, gene mutations may cause changes in RNA expression, which in turn affects the synthesis of specific proteins, ultimately leading to cellular metabolic disorders. Therefore, single omics datasets can only provide a partial understanding of the pathogenic mechanisms of lung cancer and cannot fully reveal the disease's complete picture and complexity. Summary of the Invention

[0005] To address the problems existing in the background art and improve the accuracy of lung cancer prognosis prediction, one aspect of the present invention provides a lung cancer prognosis prediction method based on multi-omics data fusion, comprising:

[0006] S1: Acquire multi-omics data and survival data of lung cancer patients, and preprocess the multi-omics data of each lung cancer patient;

[0007] S2: Concatenate the multi-omics data of each lung cancer patient to obtain the multi-omics integrated feature matrix of each training sample;

[0008] S3: The multi-omics integrated feature matrix of the training samples is mapped to a low-dimensional space by an autoencoder to obtain the multi-omics low-dimensional feature matrix of the training samples.

[0009] S4: Based on the survival data of lung cancer patients, the VSOEnetbag algorithm, which combines elastic networks and the bagging concept, is used to perform feature selection on the multi-omics low-dimensional feature matrix of the training samples to obtain the salient expression feature matrix of the training samples.

[0010] S5: Use K-means clustering to perform risk subtype clustering on the training samples based on the salient feature expression matrix of the training samples, and obtain the subtype clustering results;

[0011] S6: Use the subtype clustering results of the training samples as labels and the RNA-Seq expression profile data of lung cancer patients as independent variables to construct a multivariate Cox prognostic model;

[0012] S7: Input the RNA-Seq expression profile data of the lung cancer patients to be tested into the multivariate Cox prognostic model to obtain the prognostic results of the lung cancer patients.

[0013] Another aspect of the present invention provides a lung cancer prognosis prediction device based on multi-omics data fusion, comprising a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the lung cancer prognosis prediction device based on multi-omics data fusion performs the lung cancer prognosis prediction method based on multi-omics data fusion.

[0014] Another aspect of the present invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the aforementioned method for predicting the prognosis of lung cancer based on multi-omics data fusion.

[0015] The present invention has at least the following beneficial effects

[0016] This invention develops a multi-omics data integration framework for performing cancer subtype clustering. Through multi-level feature extraction, it achieves progressive dimensionality reduction and efficient integration of high-dimensional multi-omics data. Compared to single-omics analysis, multi-omics integration improves the accuracy and reliability of the framework's predictions. Simultaneously, this invention proposes an innovative multi-level extraction strategy, effectively addressing the computational challenges of large-scale multi-omics data and reducing the computational load of the model. This strategy, through hierarchical processing and optimized data representation, not only improves data processing efficiency but also helps reduce storage requirements and processing time, making large-scale multi-omics data analysis more feasible and efficient. This invention also proposes a feature selection method, VSOEnetbagging, based on a combination of Elastic Net regression and bagging. Based on multiple iterations and cross-validation, it accurately selects features significantly related to lung cancer prognosis. By combining LASSO regression with multiple subspace iterations, overfitting can be effectively avoided, and the most representative and predictive feature variables are selected through statistical frequency selection. Compared to traditional single feature selection methods, this method has significant advantages in improving the stability and accuracy of feature selection. Based on the feature matrix, this disclosure uses K-means clustering to subtype lung cancer samples. This clustering process can divide the samples into multiple subtypes, thereby revealing the potential biological heterogeneity of lung adenocarcinoma. This classification not only provides a new perspective for the biological understanding of cancer subtypes but also provides an important basis for the development of personalized treatment plans. Compared with traditional simple classification methods, this method, through cluster analysis of multi-omics data, can more accurately capture the complex characteristics of cancer, improving the identification and stratification capabilities of cancer subtypes. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0018] Figure 2To evaluate the optimal number of clusters based on the Silhouette Score;

[0019] Figure 3 Visualize the K-means clustering results;

[0020] Figure 4 Kaplan-Meier survival curves (high risk vs. low risk) for a multi-omics lung cancer prognostic model based on autoencoders and VSOenetbagging; where predicted_survival=0 indicates high risk and predicted_survival=1 indicates low risk. Detailed Implementation

[0021] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] Please see Figure 1 One aspect of the present invention provides a method for predicting the prognosis of lung cancer based on multi-omics data fusion, comprising:

[0023] S1: Acquire multi-omics data and survival data of lung cancer patients, and preprocess the multi-omics data of each lung cancer patient;

[0024] In this embodiment, multiple omics data for cancer and corresponding patient survival data are obtained from a reliable data source (e.g., the publicly available database Cancer Genome Atlas (TCGA)). Missing values ​​are processed for each omics dataset, and then statistical methods and computational tools are used to screen out key variables closely related to cancer occurrence and development. Next, standardization is performed to integrate these datasets into a comprehensive multi-omics dataset, which serves as input to a cross-omics ensemble autoencoder. In a specific embodiment, the lung adenocarcinoma (LUAD) dataset from TCGA is used. Considering the complexity and heterogeneity of cancer itself, four types of omics data are selected, including mRNA expression data from lung adenocarcinoma patients (gene expression, GE), miRNA expression data (miRNA expression, miRNA), DNA methylation data (DNA methylation, ME), and protein expression data (Protein Expression, PE). These four multi-omics datasets are generated using different platforms: GE and miRNA data are generated using Illumina HiSeq, ME data is generated using Illumina Infinium HumanMethylation450, and PE data is generated using Reverse Phase Protein Array (RPPA). These data are all TCGA level 3 data.

[0025] Preferably, the multi-omics data of the lung cancer patients includes: RNA-Seq expression profile data, DNA methylation data, miRNA expression data, and protein expression data of the lung cancer patients; the survival data of the lung cancer patients includes: survival time and survival outcome of the lung cancer patients.

[0026] S2: Concatenate the multi-omics data of each lung cancer patient to obtain the multi-omics integrated feature matrix of each training sample;

[0027] In this embodiment, based on the dataset collected in step S1, the omics information of lung cancer patients is obtained. These matrices are concatenated to obtain a multi-omics integrated feature matrix, which serves as the input to the feature fusion module (autoencoder).

[0028] S3: The multi-omics integrated feature matrix of the training samples is mapped to a low-dimensional space by an autoencoder to obtain the multi-omics low-dimensional feature matrix of the training samples.

[0029] Based on the multi-omics ensemble feature matrix obtained in step S2, an autoencoder is used to perform data fusion on the high-dimensional feature dataset. The autoencoder consists of an input layer, a first convolutional layer, a bottleneck layer, a second convolutional layer, and an output layer. For each layer, the number of nodes in the input layer is equal to the dimension of the feature dataset obtained in S2 (44581), the first convolutional layer has 1000 nodes, the bottleneck layer has 500 nodes, the second convolutional layer has 1000 nodes, and the output layer has the same number of nodes as the input layer (44581). The autoencoder learns effective data by minimizing the error between the reconstructed output data and the original input data. In this example, the bottleneck layer of the autoencoder is extracted to represent new features of the original omics data.

[0030] In this embodiment, except for the output layer, the activation function used between each layer is the ReLU activation function, specifically including the activation function between the input layer and the first convolutional layer, the activation function between the first convolutional layer and the bottleneck layer, the activation function between the bottleneck layer and the second convolutional layer, and the activation function between the second convolutional layer and the output layer.

[0031] The specific formula is shown below:

[0032] f(x) = max(0,x)

[0033] In the formula, x is the input real value or vector.

[0034] For real values ​​x, the ReLU function converts negative values ​​to zero, while for non-negative values ​​x, the ReLU function remains unchanged. Therefore, if x is greater than or equal to zero, the output is x; if x is less than zero, the output is zero. The output of the ReLU function for real values ​​x can be expressed as: if...

[0035] For a vector x, the ReLU function applies the ReLU operation to each element of the vector individually, resulting in the corresponding output vector. Therefore, x in the ReLU function can represent a single real value or a vector composed of multiple real values.

[0036] In this embodiment, the activation function used in the autoencoder output layer is the Sigmoid, as shown in the following formula:

[0037]

[0038] Among them, y i is the output of the i-th neuron in the output layer, is the weight matrix of the output layer, is the input of the output layer, and is the bias term of the output layer.

[0039] Specifically, data fusion is performed in S3, and the number of hidden layer nodes and bottleneck layer nodes are tuned. In this example, the number of hidden layer nodes is configured between 1000 and 1500, and the number of bottleneck layer nodes is between 500 and 700. The process involves randomly dividing the input dataset into a training set and a validation set, with the validation set accounting for 70%. For each combination of hidden layer and bottleneck layer nodes, an autoencoder model is constructed. For each model configuration, an early stopping strategy is used to prevent overfitting. Training automatically stops when the validation loss no longer decreases over 5 consecutive epochs, thus avoiding excessive training time and the risk of overfitting. After training, the best loss value on the validation set is recorded for each model configuration. The validation losses of all model configurations are compared, and the configuration with the smallest loss is selected as the optimal model. In this example, 1000 hidden layer nodes and 500 bottleneck layer nodes are chosen as the optimal configuration.

[0040] S4: Based on the survival data of lung cancer patients, the VSOEnetbag algorithm, which combines elastic networks and the bagging concept, is used to perform feature selection on the multi-omics low-dimensional feature matrix of the training samples to obtain the salient expression feature matrix of the training samples.

[0041] Preferably, step S4 includes:

[0042] S41: Construct T sub-sample subspaces by sampling from the training set using the bagging algorithm;

[0043] Preferably, constructing M sample subspaces includes: sampling with replacement from the training set, with the same number of samples sampled each time, and repeating the sampling T times to obtain T sample subspaces.

[0044] In this embodiment, several sample subspaces are generated. Each sample subspace is filtered through multiple filters, including a univariate Cox model, and LASSO regression is performed on each subset using cross-validation to filter variables for features. After repeatedly applying LASSO regression to multiple subsets, the selection frequency of each feature is recorded; important features are selected based on a cutoff threshold determined by the feature selection frequency. A certain proportion (80%) of the samples are randomly selected from 314 samples to form a subspace. This operation is repeated 100 times to generate 100 subspaces. A certain proportion (80%) of the samples are randomly selected from the training set as a subspace; this subset usually contains duplicate samples. This process is repeated 100 times, each time selecting a different sample subset, with each subspace containing approximately 80% of the samples, or approximately 251 samples. In this way, through Bagging, 100 sample subspaces are generated, each containing different samples from the training set.

[0045] S42: Traverse each sample subspace, take each feature element in the multi-omics low-dimensional feature matrix of the training samples as the independent variable, take the survival data of lung cancer patients as the dependent variable, and construct a univariate Cox regression model for each feature element.

[0046] In this embodiment, the first subspace is used as an example. This subspace has 251 samples and 500 features. For each of these 500 features, it is used as the independent variable, and the survival data of lung cancer patients corresponding to the 251 samples in the subspace is used as the dependent variable to construct a univariate Cox regression model.

[0047] For example, for feature 1, a Cox regression model is constructed using the values ​​of feature 1 of all samples in the subspace as independent variables and the corresponding survival data as dependent variables; then the same operation is performed on feature 2 to construct another univariate Cox regression model, until all 500 features have been constructed with corresponding models.

[0048] S43: Calculate the likelihood ratio test p-value for each feature element based on the Cox regression model for each feature element;

[0049] The likelihood ratio test p-value of each feature element is calculated based on the Cox regression model constructed for each feature element.

[0050] For example, for the 500 univariate Cox regression models constructed above, the likelihood ratio test p-value can be calculated for each model's corresponding feature, resulting in 500 p-values ​​corresponding to each of the 500 features. In the context of univariate Cox regression models, the likelihood ratio test is used to compare the goodness of fit between two nested models.

[0051] S44: Select the feature elements whose likelihood ratio test P-value is less than a set threshold as initial features to obtain the initial feature set for each sample subspace.

[0052] Assuming a threshold of 0.05, features with a likelihood ratio test p-value less than 0.05 are used as initial features. For example, among the 500 calculated p-values, 30 features have p-values ​​less than 0.05; these 30 features constitute the initial feature set for that subspace. This operation is performed on all 100 subspaces, resulting in their own initial feature sets for each subspace.

[0053] S45: The initial feature set of each sample subspace is further selected using the cross-validation elastic network to obtain the candidate feature set after each bagging by the cross-validation elastic network;

[0054] Preferably, step S45 includes: during each bagging process, dividing the samples in each sample subspace into two parts according to a preset ratio, wherein the part with fewer samples is used as a sub-validation set and the part with more samples is used as a sub-training set; using the survival data of the training samples in the sub-training set as labels, using the selected initial features as input, training the elastic network regression model using the sub-training set, calculating the regression coefficient of the elastic network regression model for each initial feature through the sub-validation set after training, and using the initial features with non-zero regression coefficients as candidate features to construct a candidate feature set.

[0055] Preferably, the loss function during training of the elastic network regression model includes:

[0056]

[0057] Where L represents the loss function, y i Let x represent the survival data of the i-th sample. ij Let β represent the j-th initial feature of the i-th sample. j Let λj represent the regression coefficient of the j-th initial feature, λ1 and λ2 represent the coefficients of the regularization terms of L1 loss and L2 loss, respectively, J represents the number of initial features, and n represents the number of samples.

[0058] In this embodiment, it is assumed that the preset ratio is to divide the samples into 70% and 30%. In the first subspace, 251 samples are divided according to this ratio, with approximately 176 samples serving as the sub-training set and 75 samples serving as the sub-validation set.

[0059] The survival data of samples in the sub-training set are used as labels, and the selected initial features (such as the 30 initial features in the previous example) are used as input. The elastic network regression model is trained using the sub-training set. The loss function used during training is:

[0060]

[0061] Assuming λ1 = 0.5 and λ2 = 0.3, after training, the regression coefficients of the elastic network regression model for each initial feature are calculated using a sub-validation set. Initial features with non-zero regression coefficients are used as candidate features to construct a candidate feature set. For example, after calculation, 20 of the 30 initial features have non-zero regression coefficients; these 20 features constitute the candidate feature set for this subspace after this bagging process. This operation is performed on each of the 100 subspaces during each bagging process. After multiple bagging processes across the 100 subspaces, the number of times each feature element is selected as a candidate feature is counted. For example, if feature A is selected into the candidate feature set 30 times during the bagging process across the 100 subspaces, then its frequency of selection as a candidate feature, f, is calculated. A=30÷100÷P, where P represents the number of times each subspace is bagged.

[0062] S46: Calculate the frequency at which each feature element is selected as a candidate feature based on the candidate feature set of all sample subspaces in all bagging processes;

[0063] S47: Determine the salient expression feature matrix of the training samples using the PST method based on the frequency of each feature element being selected as a candidate feature;

[0064] Preferably, step S47 includes:

[0065] S471: For each feature element j, based on its frequency f of being selected as a candidate feature. j Calculate the corresponding number of choices, k = T × f j T represents the number of subspaces;

[0066] S472: Calculate the significance of feature element j:

[0067]

[0068] Where, d j X represents the significance of feature element j. j P(X) represents the number of times feature element j is selected in the T subspaces; j ≥k) represents the probability that feature element j is selected more than k times in the T subspaces; p=0.5 represents the probability that each feature element is randomly selected in a subspace; P i This represents p raised to the power of i;

[0069] S473: Sort all feature elements in ascending order of significance, and calculate the significance of feature element j after sorting. Where m represents the number of feature elements, l j The position of feature element j after sorting;

[0070] S474: If the significance of characteristic element j If the value is less than the set threshold, feature element j is taken as a salient expression feature, and thus a salient expression feature matrix of the training sample is constructed based on all the salient expression features of the training sample.

[0071] In this embodiment, it is assumed that the threshold is set to be such that the significance of feature A after sorting is... If the value is less than the set threshold of 0.01, then feature A is considered a salient feature. In this way, all salient features are determined, and a salient feature matrix of the training samples is constructed based on all the salient features of the training samples.

[0072] S5: Use K-means clustering to perform risk subtype clustering on the training samples based on the salient feature expression matrix of the training samples, and obtain the subtype clustering results;

[0073] Suppose we have obtained the salient feature representation matrix of the training samples through the previous feature selection steps. This matrix contains 200 lung cancer patient samples (training samples), each described by 10 salient features. We aim to classify these patient samples into different risk subtypes using the K-means clustering algorithm.

[0074] S6: Use the subtype clustering results of the training samples as labels and the RNA-Seq expression profile data of lung cancer patients as independent variables to construct a multivariate Cox prognostic model;

[0075] S7: Input the RNA-Seq expression profile data of the lung cancer patients to be tested into the multivariate Cox prognostic model to obtain the prognostic results of the lung cancer patients.

[0076] Another aspect of the present invention provides a lung cancer prognosis prediction device based on multi-omics data fusion, comprising a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the lung cancer prognosis prediction device based on multi-omics data fusion performs the lung cancer prognosis prediction method based on multi-omics data fusion.

[0077] Another aspect of the present invention provides a computer-readable storage medium storing a program that, when executed by a processor, implements the aforementioned method for predicting the prognosis of lung cancer based on multi-omics data fusion.

[0078] In this embodiment, we employ a combination of K-means clustering, LASSO feature selection, and Cox survival analysis to construct a model for predicting the survival of lung cancer patients. First, the top 30 key genes were extracted from the multi-omics data and standardized to suit subsequent analysis. Then, the suitability of different cluster numbers was evaluated based on the Silhouette Score, ultimately determining the optimal number of clusters as k=2. (See also...) Figure 2 Using K-means clustering, we divided patients into high-risk and low-risk groups, and validated the rationality of this clustering through t-SNE dimensionality reduction visualization. Figure 4 As shown in the figure. We also calculated the Jaccard similarity index (0.78), demonstrating that this method can capture the distribution characteristics of survival states effectively.

[0079] To further optimize the model, we used LASSO regression for feature selection, ultimately choosing 12 vectors with the highest predictive value. After re-performing K-means clustering based on the selected features, we found that the Jaccard index increased to 0.9382716 and the Silhouette Score improved to 0.1338907. Furthermore, we used Kaplan-Meier survival analysis to further validate the survival significance of the clustering results, finding that the survival rate of the high-risk group was significantly lower than that of the low-risk group (p=0.022), indicating that this clustering method can effectively distinguish patients with different survival outcomes. Figure 3 As shown.

[0080] To further quantify the survival prediction ability, we used a Cox proportional hazards regression model to evaluate the survival prediction effect of the selected genes. Cox analysis showed that the model's C-index (consistency index) was as high as 0.78, indicating that the model has a strong predictive ability for patient survival time. Simultaneously, to further evaluate the model's classification performance, we used ROC curves to calculate the AUC (area under the curve) of survival prediction, ultimately obtaining an AUC of 0.81. Figure 4 This further demonstrates that the model can predict patient survival outcomes with relatively high accuracy. In summary, the K-means+LASSO+Cox survival analysis model constructed in this study can not only effectively stratify patients by risk but also provide reliable survival prediction information, which is helpful for the development of precision medicine and personalized treatment plans.

[0081] In summary, this invention develops a multi-omics data integration framework for performing cancer subtype clustering, achieving progressive dimensionality reduction and efficient integration of high-dimensional multi-omics data through multi-level feature extraction. Compared to single-omics analysis, multi-omics integration can improve the accuracy and reliability of the framework's predictions. Simultaneously, this invention proposes an innovative multi-level extraction strategy, effectively addressing the computational challenges of large-scale multi-omics data and reducing the computational load of the model. This strategy, through hierarchical processing and optimized data representation, not only improves data processing efficiency but also helps reduce storage requirements and processing time, making large-scale multi-omics data analysis more feasible and efficient. This invention proposes a feature selection method, VSOEnetbagging, based on a combination of ElasticNet regression and bagging. Based on multiple iterations and cross-validation, it accurately selects features significantly related to lung cancer prognosis. By combining LASSO regression with multiple subspace iterations, overfitting can be effectively avoided, and the most representative and predictive feature variables are selected through statistical frequency selection. Compared with traditional single feature selection methods, this method has significant advantages in improving the stability and accuracy of feature selection. Based on the feature matrix, this disclosure uses K-means clustering to subtype lung cancer samples. This clustering process can divide the samples into multiple subtypes, thereby revealing the potential biological heterogeneity of lung adenocarcinoma. This classification not only provides a new perspective for the biological understanding of cancer subtypes but also provides an important basis for the development of personalized treatment plans. Compared with traditional simple classification methods, this method, through cluster analysis of multi-omics data, can more accurately capture the complex characteristics of cancer, improving the identification and stratification capabilities of cancer subtypes.

[0082] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0083] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for predicting the prognosis of lung cancer based on multi-omics data fusion, characterized in that, include: S1: Acquire multi-omics data and survival data of lung cancer patients, and preprocess the multi-omics data of each lung cancer patient; S2: Concatenate the multi-omics data of each lung cancer patient to obtain the multi-omics integrated feature matrix of each training sample; S3: The multi-omics integrated feature matrix of the training samples is mapped to a low-dimensional space by an autoencoder to obtain the multi-omics low-dimensional feature matrix of the training samples. S4: Based on the survival data of lung cancer patients, the salient expression feature matrix of the training samples is obtained by combining elastic networks and bagging ideas to perform feature selection on the multi-omics low-dimensional feature matrix of the training samples. S5: Use K-means clustering to perform risk subtype clustering on the training samples based on the salient feature expression matrix of the training samples, and obtain the subtype clustering results; S6: Use the subtype clustering results of the training samples as labels and the RNA-Seq expression profile data of lung cancer patients as independent variables to construct a multivariate Cox prognostic model; S7: Input the RNA-Seq expression profile data of the lung cancer patients to be tested into the multivariate Cox prognostic model to obtain the prognostic results of the lung cancer patients.

2. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 1, characterized in that, The multi-omics data of the lung cancer patients include: RNA-Seq expression profile data, DNA methylation data, miRNA expression data, and protein expression data of the lung cancer patients; the survival data of the lung cancer patients include: survival time and survival outcome of the lung cancer patients.

3. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 1, characterized in that, The autoencoder comprises an encoder and a decoder. The encoder consists of an input layer, a first hidden layer, and a bottleneck layer, and is used to map high-dimensional features of the input to a low-dimensional space. The decoder consists of a second hidden layer and an output layer, and is used to restore the features mapped to the low-dimensional space to the original high-dimensional features.

4. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 1, characterized in that, Step S4 includes: S41: Construct T sample subspaces by sampling from the training set using the bagging algorithm; S42: Traverse each sample subspace, take each feature element in the multi-omics low-dimensional feature matrix of the training samples as the independent variable, take the survival data of lung cancer patients as the dependent variable, and construct a univariate Cox regression model for each feature element. S43: Calculate the likelihood ratio test p-value for each feature element based on the Cox regression model for each feature element; S44: Select the feature elements whose likelihood ratio test P-value is less than a set threshold as initial features to obtain the initial feature set for each sample subspace. S45: The initial feature set of each sample subspace is further selected using the cross-validation elastic network to obtain the candidate feature set after each bagging by the cross-validation elastic network; S46: Calculate the frequency at which each feature element is selected as a candidate feature based on the candidate feature set of all sample subspaces in all bagging processes; S47: Determine the salient expression feature matrix of the training samples using the PST method based on the frequency with which each feature element is selected as a candidate feature.

5. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 4, characterized in that, The construction of T sample subspaces includes: sampling with replacement from the training set, with the same number of samples sampled each time, and repeating the sampling T times to obtain T sample subspaces.

6. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 4, characterized in that, Step S45 includes: during each bagging process, dividing the samples in each sample subspace into two parts according to a preset ratio, wherein the part with fewer samples is used as a sub-validation set and the part with more samples is used as a sub-training set; using the survival data of the training samples in the sub-training set as labels, using the selected initial features as input, training the elastic network regression model using the sub-training set, calculating the regression coefficient of the elastic network regression model for each initial feature through the sub-validation set after training, and using the initial features with non-zero regression coefficients as candidate features to construct a candidate feature set.

7. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 6, characterized in that, The loss function used during training of the elastic network regression model includes: in, Represents the loss function. Indicates the first Survival data for each sample Indicates the first The first sample An initial feature, Indicates the first The regression coefficients of the initial features, and Let L1 and L2 represent the coefficients of the regularization terms, respectively. Indicates the number of initial features. Indicates the number of samples.

8. The lung cancer prognosis prediction method based on multi-omics data fusion according to claim 4, characterized in that, Step S47 includes: S471: For each feature element Based on the frequency with which it is selected as a candidate feature Calculate the corresponding number of selections T represents the number of subspaces; S472: Calculate characteristic elements Significance: in, Representing characteristic elements The significance of Representing characteristic elements The number of selections in T subspaces; Representing characteristic elements The probability that the number of selections in the T subspaces is greater than k; This represents the probability that each feature element is randomly selected within a subspace; express power of i; S473: Sort all feature elements in ascending order of significance. For feature elements... Calculate the significance of its sorting Where m represents the number of feature elements. Sorted feature elements Location; S474: If the characteristic element Significance If the value is less than the set threshold, then the feature element will be... As salient expression features, a salient expression feature matrix of the training samples is constructed based on all salient expression features of the training samples.

9. A lung cancer prognostic prediction device based on multi-omics data fusion, characterized in that, It includes a processor and a memory; the memory is used to store a computer program; the processor is connected to the memory and is used to execute the computer program stored in the memory, so that the lung cancer prognosis prediction device based on multi-omics data fusion performs the lung cancer prognosis prediction method based on multi-omics data fusion as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a program, characterized in that, When the program is executed by the processor, it implements the lung cancer prognosis prediction method based on multi-omics data fusion as described in any one of claims 1 to 8.

Images