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

Abstract

The application discloses a cancer subtype classification method based on multi-omics data, comprising the following steps: S1. Obtain several omics data, and pre-process each omics data to obtain a multi-omics data set; S2. Divide the omics data into a benchmark omics data and several supplementary omics data; S3. Independently construct an autoencoder for each omics data, and optimize through a reconstruction loss to realize primary feature extraction and obtain a multi-omics low-dimensional feature matrix; S4. Combine the benchmark omics autoencoder and each supplementary omics autoencoder to obtain a coupled autoencoder pair, perform secondary feature extraction on the multi-omics low-dimensional feature matrix to obtain a cross-omics shared feature matrix; S5. Perform a series splicing operation on the cross-omics shared feature matrix to obtain a multi-omics integrated feature matrix; and S6. Construct a subtype clustering block composed of a deep neural network, input the multi-omics integrated feature matrix into the subtype clustering block, and finally output a subtype category of the sample.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and big data analysis, specifically to a subtype classification method and system based on multi-omics data integration and analysis. Background Technology

[0002] In cancer research, the application of multi-omics technologies has become a trend. Its aim is to gain a more comprehensive understanding of the biological characteristics of cancer by integrating data from different levels, including genomics, proteomics, metabolomics, and transcriptomics. The integration of multi-omics data can provide a panoramic view of the occurrence and development of cancer, revealing its complexity. By analyzing various biomarkers and molecular events, researchers can identify key drivers of cancer, thus supporting precision medicine. Furthermore, multi-omics data can reveal molecular differences between different cancer subtypes, contributing to the development of targeted treatments and providing important information for early diagnosis, treatment, and prognostic assessment of cancer.

[0003] However, the integration and analysis of multi-omics data faces a series of challenges. First, the sources, types, and quantities of different omics data vary greatly, leading to data heterogeneity. Second, steps such as data preprocessing, feature selection, and handling missing values ​​require highly customized algorithms and expertise. Furthermore, effectively extracting useful information from multi-omics data and transforming it into interpretable biological knowledge remains a current research hotspot and challenge.

[0004] Accurate identification of cancer subtypes is crucial for personalized treatment. Traditional cancer classification methods often rely on single biomarkers or clinical features, which limits our understanding of the complexity of cancer. In contrast, cancer subtype classification based on multi-omics data comprehensively considers various biological information, improving the accuracy and reliability of classification. Furthermore, this approach helps discover new biomarkers and potential therapeutic targets, bringing new hope to cancer treatment. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies, to achieve a more comprehensive understanding of the occurrence and development of cancer and to improve the accuracy of subtype classification. It provides a cancer subtype classification method and system based on multi-omics data. By integrating and analyzing data from different omics groups, it overcomes the problems of data heterogeneity and analytical difficulty. Employing advanced autoencoders and contrastive learning strategies, it effectively extracts multi-level features from multi-omics data and accurately classifies cancer subtypes, realizing an end-to-end process for identifying cancer subtypes based on multi-omics data integration. This provides clinicians and researchers with a powerful tool to better understand and treat cancer. This invention is expected to promote the development of precision medicine for cancer and bring better treatment outcomes to patients.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A cancer subtype classification method based on multi-omics data is used to achieve end-to-end identification of cancer subtypes based on multi-omics data integration, specifically including the following steps:

[0008] S1. Obtain several omics data points for cancer, preprocess each omics data point to obtain a multi-omics dataset; preprocessing methods include missing value handling, variable selection, and standardization;

[0009] S2. Divide the omics data within the multi-omics dataset into a baseline omics dataset Omics t and several supplementary omics datasets Omics m;

[0010] S3. Construct an autoencoder independently for each omics data in the multi-omics dataset, and optimize it through reconstruction loss to achieve primary feature extraction and obtain a new set of multi-omics low-dimensional feature matrices;

[0011] All of these autoencoders together form a cross-omics integrated autoencoder; the autoencoder constructed from baseline omics data is called the baseline omics autoencoder, and the autoencoder constructed from supplementary omics data is called the supplementary omics autoencoder.

[0012] Primary feature extraction is performed on the supplementary omics autoencoder to obtain the supplementary omics low-dimensional feature matrix;

[0013] All autoencoders include both encoders and decoders;

[0014] S4. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs. Use the autoencoder pairs to perform secondary feature extraction on the low-dimensional feature matrix of multiple omics to obtain the cross-omics shared feature matrix.

[0015] S5. The cross-omics shared feature matrix obtained in step S4 is concatenated and concatenated to obtain the multi-omics integrated feature matrix, which is used as the input for the subtype clustering block; the formula for this process is as follows:

[0016] z = concat(z) 1→t ,z 2→t ,...z m→t ,...,z M→t (m≠t)

[0017] Where concat(·) represents the concatenation operator, M is the number of omics data, and z m→t z is the cross-omics shared feature matrix between omics data m and omics data t; z is the multi-omics ensemble feature matrix, where rows represent samples and columns represent features.

[0018] S6. Construct a subtype clustering block composed of a deep neural network. Input the multi-omics integrated feature matrix z obtained in step S5 into the subtype clustering block. Optimize the inter-class relationship of samples by performing contrastive learning in the column space of the data pairs to obtain the subtype soft label matrix.

[0019] Furthermore, in step S3: for each omics data x m Build an autoencoder and reconstruct the loss By applying constraints, we achieve primary feature extraction, resulting in a new set of low-dimensional feature matrices from multi-omics learning; the specific formula is shown below:

[0020] z m =enc m (x m )

[0021]

[0022] Where, x m Represents the m-th omics data, enc m (·) represents the encoder of the m-th omics data autoencoder, dec m (·) represents the decoder of the m-th omics data autoencoder, z m The low-dimensional feature matrix representing the m-th omics data. The primary reconstructed representation of the m-th omics data;

[0023] The reconstruction loss formula is expressed as follows:

[0024]

[0025] Where N is the number of samples, i∈[1,N];

[0026] Furthermore, in step S4:

[0027] S401. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; for any autoencoder pair, input the supplementary omics low-dimensional feature matrix obtained in step S3 into the baseline omics decoder, and then perform secondary reconstruction through the baseline omics encoder to obtain the cross-omics shared feature matrix.

[0028] The fusion process of the m-th supplementary omics data into the baseline omics data t is expressed by the following formula:

[0029]

[0030]

[0031] Among them, enc t(·) represents the encoder of the baseline omics data autoencoder, dec t (·) represents the decoder of the baseline omics data autoencoder. z represents the quadratic reconstructed feature matrix of the m-th omics data. m→t The shared feature matrix representing the m-th omics data; z m x represents the low-dimensional feature matrix of the m-th omics data. m Represents omics data; enc m (·) represents the encoder of the m-th omics data autoencoder;

[0032] S402. The process in step S401 involves cross-omics cyclic loss. The constraints are applied, and the formula is expressed as follows:

[0033]

[0034] in, The cyclic loss between omics data m and omics data t;

[0035] Furthermore, in step S6:

[0036] S601. Construct a deep neural network to form subtype clustering blocks. Input the multi-omics ensemble feature matrix z obtained in S5 into the subtype clustering blocks and perform two data augmentations to form data pairs {h}. a ,h b}, where h a h b The two symbols represent the augmented omics data, with a and b corresponding to two data augmentations respectively. For numerical omics data, Gaussian noise is added to the original multi-omics ensemble feature matrix for data augmentation.

[0037] S602. A nonlinear transformation is performed on the data pairs using a projection block composed of two fully connected layers, and the output layer of the subtype clustering block is set to the number of subtypes, with the activation function being Softmax. Softmax transforms the final output of the subtype clustering block into the probability that a sample belongs to each subtype, obtaining a soft label matrix containing subtype category information. The formula for this process is as follows:

[0038] y a =g(h a )

[0039] y b =g(h b )

[0040] Where g(·) represents two nonlinear transformations, {y a ,y b} represents the soft label matrix output by the subtype clustering block;

[0041] S603. The clustering process relies on contrastive learning of the columns of the soft-label matrix; the similarity between different clusters is calculated using cosine similarity, and then the contrastive loss L is calculated. con Optimization is performed; the formula for this process is as follows:

[0042]

[0043]

[0044]

[0045] Where j, k∈[1,C], y n,j This represents the probability that the nth sample is assigned to the jth subtype; τ is the contrast loss between the j-th subtype and the other subtypes; τ is the temperature coefficient, q is the hyperparameter controlling the shape of the loss function, and C is the number of subtype categories;

[0046] Finally, the subtype categories of all samples were obtained.

[0047] This invention also provides a cancer subtype classification system based on multi-omics data, comprising:

[0048] The multi-omics data acquisition and preprocessing module is used to collect multi-omics data of cancer samples from data sources and perform preprocessing operations on the multi-omics data;

[0049] The primary feature extraction module is used to construct an autoencoder for single-omics data and extract primary features to obtain a multi-omics low-dimensional feature matrix.

[0050] The secondary feature extraction module is used to divide the omics data into a baseline omics data and several supplementary omics data, and combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; and perform secondary feature extraction on the low-dimensional feature matrix of multiple omics through the autoencoder pairs to obtain a cross-omics shared feature matrix.

[0051] The multi-omics integration module is used to integrate a set of cross-omics shared features into a matrix and input it into the subtype clustering module;

[0052] The subtype clustering module consists of a projection block composed of a deep neural network and a soft label output layer. By performing data augmentation on the input feature matrix and implementing a contrastive learning strategy in the column space, it automatically completes the optimization process of subtype clustering and outputs subtype labels. Then, it transmits the subtype label of the sample to the display module.

[0053] The display module is used to display the subtype classification results of the samples;

[0054] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the steps of the cancer subtype classification method based on multi-omics data.

[0055] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cancer subtype classification method based on multi-omics data.

[0056] Compared with the prior art, the beneficial effects of the technical solution of the present invention are:

[0057] 1) 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.

[0058] 2) Applying deep neural networks to multi-scale cancer multi-omics datasets, combined with contrastive learning strategies, can significantly improve the accuracy and efficiency of cancer subtype classification. Multiple coupled autoencoder pairs, through their deep structure, can effectively capture complex nonlinear relationships and deep-level features in the data. Furthermore, the introduction of contrastive learning strategies further optimizes the learning of relationships between sample subtypes, enabling the model to better distinguish between different cancer subtypes. This combination leads to more accurate classification results, helping doctors make more precise diagnoses and develop personalized treatment plans.

[0059] 3) The technical framework and processing strategy of this invention are not only applicable to cancer subtype classification, but also have broad versatility and scalability. This means that the same technology can be applied to study the occurrence and development mechanisms of other complex diseases (such as Alzheimer's disease, diabetes, etc.). Furthermore, it can be extended to other studies that require processing and analyzing complex multi-omics data containing multiple scales and types. Attached Figure Description

[0060] Figure 1 This is a flowchart of the classification method of the present invention;

[0061] Figure 2 This is a schematic diagram of the feature extraction process and subtype clustering blocks of the multi-level coupled autoencoder in Example 1;

[0062] Figure 3 This is a Kaplan-Meier survival analysis curve of different subtype samples in an embodiment of the present invention. Detailed Implementation

[0063] To provide a clearer explanation of the objectives, technical solutions, and advantages of this invention, a more detailed description will be given below in conjunction with the accompanying drawings and embodiments. It should be noted that the specific implementation methods described herein are merely for illustrative purposes and do not constitute a limitation on the scope of protection of this invention.

[0064] Example 1

[0065] This embodiment provides a cancer subtype classification method based on multi-omics data, realizing a complete end-to-end process from data input to cancer subtype identification. (See [link]) Figure 1 and Figure 2 The specific process includes the following steps:

[0066] S1. Data Collection and Preprocessing:

[0067] Multiple omics data on cancer were obtained from reliable data sources (e.g., the publicly available database Cancer Genome Atlas (TCGA)). Missing values ​​were removed from each omics dataset, and then statistical methods and computational tools were used to screen for key variables closely related to cancer occurrence and development. Next, standardization was performed, and these datasets were integrated into a comprehensive multi-omics dataset, which served as input to a cross-omics ensemble autoencoder.

[0068] In a specific embodiment, the Breast Cancer Association (BRCA) dataset from TCGA was used. Considering the complexity and heterogeneity of cancer itself, four types of omics data were selected: gene expression (GE), miRNA expression (miRNA), DNA methylation (ME), and copy number variation (CNV) data at the gene level from breast cancer patients. These four multi-omics datasets were generated using different platforms: GE and miRNA data were generated using Illumina HiSeq, ME data using Illumina Infinium HumanMethylation450, and CNV data using Affymetrix Genome-Wide Human SNP Array 6.0. All of these data are TCGA level 3 data.

[0069] S2. Selection of baseline omics data:

[0070] Based on the prior knowledge of domain experts and the interaction analysis results of existing omics data, the selected omics data are divided into baseline omics data (Omics t) and multiple supplementary omics data (Omics m). The baseline omics data will serve as the core of the analysis, while the supplementary omics data will be used to perform cross-omics fusion to enrich the information. In this embodiment, GE is selected as the baseline omics data, and miRNA, ME, and CNV are selected as supplementary omics data.

[0071] S3. Construction of the primary feature extractor and primary feature extraction:

[0072] First, it is necessary to process each omics dataset x m An autoencoder is constructed independently to form the primary feature extractor. Each independent autoencoder consists of an encoder (enc(·)) and a decoder (dec(·)). The encoder's role is to transform the input data (in this embodiment, the various omics data) into a lower-dimensional latent representation. This process can be viewed as a compression of the data, simplifying the data structure by extracting important features. The encoder consists of multiple hidden layers, each capable of learning different levels of abstraction of the data. The decoder attempts to recover data that is as similar as possible to the original input by decompressing the latent representation back into the original data space. In this embodiment, the decoder's structure corresponds to that of the encoder, but the goal is to accurately reconstruct the input data, thereby ensuring that the latent representation captures sufficient information.

[0073] The formula is shown below:

[0074] z m =enc m (x m )

[0075]

[0076] Where, x m Represents the m-th omics data, enc m (·) represents the encoder of the m-th omics data autoencoder, dec m (·) represents the decoder of the m-th omics data autoencoder, z m The low-dimensional feature matrix representing the m-th omics data. The primary reconstructed representation of the m-th omics data;

[0077] By reconstructing the loss The above process is constrained. This step achieves the initial feature extraction, outputting a new set of multi-omics low-dimensional feature matrices; the reconstruction loss formula is expressed as follows:

[0078] The reconstruction loss formula is expressed as follows:

[0079]

[0080] Where N is the number of samples, i∈[1,N];

[0081] The autoencoder constructed from the baseline omics data in the above process is called the baseline omics autoencoder, and the autoencoder constructed from the supplementary omics data is called the supplementary omics autoencoder; primary feature extraction is performed on the supplementary omics autoencoder to obtain the supplementary omics low-dimensional feature matrix.

[0082] S4. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain multiple coupled autoencoder pairs.

[0083] S401. For any autoencoder pair, input the supplementary omics low-dimensional feature matrix obtained in S3 into the baseline omics decoder, and then perform secondary reconstruction through the baseline omics encoder to obtain the cross-omics shared feature matrix.

[0084] The fusion process of the m-th supplementary omics to the baseline omics t is expressed by the following formula:

[0085] The fusion process of the m-th supplementary omics data into the baseline omics data t is expressed by the following formula:

[0086]

[0087]

[0088] Among them, enc t (·) represents the encoder of the baseline omics data autoencoder, dec t (·) represents the decoder of the baseline omics data autoencoder. z represents the quadratic reconstructed feature matrix of the m-th omics data. m→t The shared feature matrix representing the m-th omics data; z m x represents the low-dimensional feature matrix of the m-th omics data. m Represents omics data; enc m (·) represents the encoder of the m-th omics data autoencoder;

[0089] S402. This process involves cross-omics cyclic loss. The constraints are applied, and the formula is expressed as follows:

[0090]

[0091] in, The cyclic loss between omics m and omics t represents the cyclic loss.

[0092] S5. The cross-omics shared feature matrix obtained from the secondary feature extraction in step S4 is concatenated and spliced ​​to form a comprehensive multi-omics integrated feature matrix, which serves as the input for the subtype clustering block; the formula is expressed as follows:

[0093] z = concat(z) 1→t ,z 2→t ,...z m→t ,...,z M→t (m≠t)

[0094] Where concat(·) represents the concatenation operator, M is the number of omics, and z m→t z is the cross-omics shared feature matrix between omics data m and omics data t; z is the multi-omics ensemble feature matrix, where rows represent samples and columns represent features.

[0095] S6. Construct a deep neural network to form subtype clustering blocks, and perform subtype clustering;

[0096] S601. Input the multi-omics ensemble feature matrix z obtained in step S5 into the subtype clustering block and perform two data augmentations to form positive and negative data pairs {h}. a ,h b}, where h a h b The symbols a and b represent the augmented omics data, respectively, with a and b corresponding to two data augmentations. For numerical omics data, Gaussian noise is added to the original multi-omics ensemble feature matrix for data augmentation.

[0097] S602. A nonlinear transformation is performed on the data pairs using a projection block composed of two fully connected layers, and the output layer of the subtype clustering block is set to the number of subtype categories (in this embodiment, the number of output subtype categories of BRCA is set to 5), with the activation function being Softmax. Softmax transforms the final output of the model into the probability that a sample belongs to each subtype, obtaining a soft label matrix containing subtype category information; the formula for this process is as follows:

[0098] y a =g(h a )

[0099] y b =g(h b )

[0100] Where g(·) represents two nonlinear transformations, {y a ,y b} represents the soft label matrix output by the clustering block;

[0101] S603. The clustering process relies on contrastive learning of the column space of the soft-label matrix. The similarity between different clusters is calculated using cosine similarity, and then optimized using contrastive loss; the formula for this process is as follows:

[0102]

[0103]

[0104]

[0105] Where j, k∈[1,C], y n,j This represents the probability that the nth sample is assigned to the jth subtype; τ is the contrast loss between the j-th subtype and the other subtypes; τ is the temperature coefficient, q is the hyperparameter controlling the shape of the loss function, and C is the number of subtype categories;

[0106] Ultimately, the subtype categories of all samples from 10 TCGA cancer multi-omics datasets were obtained, and Kaplan-Meier survival analysis was performed on the five newly reclassified subtype categories. Figure 3 The results of the survival analysis and the level of significant differences in subtypes are presented. Furthermore, compared with current advanced multi-omics ensemble subtype classification methods, this study achieved the highest log10 P-value and identified the most significantly relevant clinical parameters, as shown in Table 1.

[0107] Table 1. Performance comparison of this invention with other advanced multi-omics integration methods for subtype classification tasks.

[0108]

[0109] The effectiveness of this invention was verified through application on the TCGA BRCA multi-omics dataset. For subtypes that are difficult to distinguish, this invention can achieve more accurate classification. This invention employs coupled networks and contrastive learning strategies to integrate multi-omics data to mine important molecular characteristics of cancer, effectively classifying cancer samples into significantly different subtypes. It has the following significant advantages:

[0110] (1) Efficient data integration framework: By adopting a hierarchical feature extraction and integration method, this invention significantly reduces the computational burden of the model while achieving effective multi-omics integration. This multi-level processing strategy and the deployment of coupled networks not only improve the efficiency of data processing, but also reduce the requirements for storage and computation time, making the analysis of large-scale multi-omics data more efficient and easier to implement.

[0111] (2) Improved Cancer Subtype Classification Performance: By utilizing deep neural networks to process multi-scale cancer data and combining them with a contrastive learning strategy to implement end-to-end subtype clustering, this invention significantly improves the accuracy and efficiency of cancer subtype classification. The deep learning model captures complex patterns and deep features in the data, and contrastive learning further optimizes the model's understanding of the relationships between different subtypes, thereby achieving more accurate classification and assisting doctors in making more precise diagnoses and treatment plans.

[0112] (3) Wide applicability and scalability: Due to the flexibility and scalability of the framework, the technical framework and strategy of this invention are not limited to cancer subtype classification, but are also applicable to the study of other complex diseases, such as Alzheimer's disease and diabetes. In addition, by appropriately modifying the backbone of the feature extractor, it is also applicable to other fields that require the analysis and processing of multi-scale and multi-type data, showing good versatility and scalability.

[0113] Example 2

[0114] This embodiment proposes a cancer subtype classification system based on multi-omics data, including a multi-omics data acquisition and preprocessing module, a primary feature extraction module, a secondary feature extraction module, a multi-omics integration module, a subtype clustering module, and a display module;

[0115] The multi-omics data acquisition and preprocessing module is used to collect multi-omics data of cancer samples from reliable data sources and perform preprocessing operations such as cleaning, variable screening and standardization on the multi-omics data.

[0116] The primary feature extraction module is used to construct an autoencoder for single-omics data and extract primary features to obtain a multi-omics low-dimensional feature matrix.

[0117] The secondary feature extraction module is used to divide the omics data into a baseline omics data and several supplementary omics data, and to combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; the autoencoder pairs are used to perform secondary feature extraction on the low-dimensional feature matrices of multiple omics to obtain cross-omics shared feature matrices.

[0118] The multi-omics integration module is used to combine a set of cross-omics shared features into a matrix and input it into the subtype clustering module;

[0119] The subtype clustering module consists of a projection block composed of a deep neural network and a soft label output layer. By performing data augmentation on the input feature matrix and implementing a contrastive learning strategy in the column space, it automatically completes the optimization process of subtype clustering and outputs subtype labels. Then, it transmits the subtype label of the sample to the display module.

[0120] The display module is used to display the subtype classification results of the samples;

[0121] Preferably, embodiments of this application also provide a specific implementation of an electronic device capable of implementing all steps in the cancer subtype classification method based on multi-omics data described in the above embodiments. The electronic device specifically includes the following:

[0122] Processor, memory, communications interface, and bus;

[0123] The processor is the core component of an electronic device, responsible for interpreting and executing computer programs stored in memory. A processor can be a general-purpose central processing unit (CPU), a dedicated application-specific integrated circuit (ASIC), or other types of processors such as a graphics processing unit (GPU) or a field-programmable gate array (FPGA). The choice of processor depends on the complexity and performance requirements of the task to be processed. The processor and memory are connected via a high-speed communication interface, which can be PCI Express, USB 3.0 / 3.1, or other high-speed transmission standards.

[0124] The memory is used to store the code of computer programs and necessary data; it can be any type of computer-readable storage medium, such as, but not limited to, solid-state drives, dynamic random access memory, flash memory cards, optical discs, or any other form of digital storage device.

[0125] The communication interface is used to enable information transmission between server-side devices, metering devices, and user-side devices.

[0126] A computer program contains a series of instructions and algorithms, which are encoded in memory and executed one by one by the processor.

[0127] Preferably, embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the cancer subtype classification method based on multi-omics data in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the cancer subtype classification method based on multi-omics data in the above embodiments.

[0128] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0129] While this application provides method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed sequentially as shown in the embodiments or drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0130] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0132] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0133] This invention is not limited to the embodiments described above. The above description of specific embodiments is intended to illustrate and explain the technical solutions of this invention. The specific embodiments described above are merely illustrative and not restrictive. Without departing from the spirit and scope of the claims, those skilled in the art can make many specific modifications based on the teachings of this invention, and these modifications all fall within the scope of protection of this invention.

Claims

1. A cancer subtype classification method based on multi-omics data, characterized in that, The process for achieving end-to-end cancer subtype identification based on multi-omics data integration includes the following steps: S1. Obtain several omics data points related to cancer, preprocess each omics data point to obtain a multi-omics dataset; preprocessing methods include missing value handling, variable selection, and standardization; S2. Divide the omics data within the multi-omics dataset into a baseline omics dataset. and several supplementary omics data ; S3. Construct an autoencoder independently for each omics data in the multi-omics dataset, and optimize it through reconstruction loss to achieve primary feature extraction and obtain a new set of multi-omics low-dimensional feature matrices; All of these autoencoders together form a cross-omics integrated autoencoder; the autoencoder constructed from baseline omics data is called the baseline omics autoencoder, and the autoencoder constructed from supplementary omics data is called the supplementary omics autoencoder. Primary feature extraction is performed on the supplementary omics autoencoder to obtain the supplementary omics low-dimensional feature matrix; All autoencoders include both encoders and decoders; S4. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs. Perform secondary feature extraction on the multi-omics low-dimensional feature matrix using these autoencoder pairs to obtain a cross-omics shared feature matrix; including: S401. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; for any autoencoder pair, input the supplementary omics low-dimensional feature matrix obtained in step S3 into the baseline omics decoder, and then perform secondary reconstruction through the baseline omics encoder to obtain the cross-omics shared feature matrix. For the m Supplemental omics data to baseline omics data t The fusion process is expressed by the following formula: in, The encoder representing the autoencoder of baseline omics data. The decoder represents the autoencoder of baseline omics data. Representing the m The secondary reconstruction feature matrix of omics data Representing the m Shared feature matrix of omics data; Representing the m Low-dimensional feature matrix of omics data Representing omics data; Representing the m An encoder for an autoencoder of omics data; S402. The process in step S401 involves cross-omics cyclic loss. The constraints are applied, and the formula is expressed as follows: in, Representative omics data m With omics data t Cyclic losses between them; S5. The cross-omics shared feature matrix obtained in step S4 is concatenated and concatenated to obtain the multi-omics integrated feature matrix, which is used as the input for the subtype clustering block; the formula for this process is as follows: in Represents the chain operator, M For the number of omics data, Representative omics data m With omics data t Cross-omics sharing of feature matrices; z This is a multi-omics integrated feature matrix, where rows represent samples and columns represent features; S6. Construct subtype clustering blocks composed of deep neural networks, and integrate the multi-omics feature matrix obtained in step S5. z The subtype clustering blocks are input, and the inter-class relationships of the samples are optimized by performing contrastive learning in the column space of the data pairs to obtain the subtype soft label matrix.

2. The cancer subtype classification method based on multi-omics data according to claim 1, characterized in that, In step S3: For each omics data Build an autoencoder and reconstruct the loss By applying constraints, we achieve primary feature extraction, resulting in a new set of low-dimensional feature matrices for multi-omics learning. The specific formula is shown below: in, Representing the m omics data, Representing the m The encoder of an autoencoder for omics data. Representing the m Decoder for omics data autoencoders Representing the m Low-dimensional feature matrix of omics data Representing the m A preliminary reconstruction representation of omics data; The reconstruction loss formula is expressed as follows: in N For the sample size, .

3. The cancer subtype classification method based on multi-omics data according to claim 1, characterized in that, In step S6: S601. Construct a deep neural network to form subtype clustering blocks. Input the multi-omics ensemble feature matrix z obtained in S5 into the subtype clustering blocks and perform two data augmentations to form data pairs. ,in h a 、h b The two symbols represent the augmented omics data, with a and b corresponding to two data augmentations respectively. For numerical omics data, Gaussian noise is added to the original multi-omics ensemble feature matrix for data augmentation. S602. A nonlinear transformation is performed on the data pairs using a projection block consisting of two fully connected layers, and the output layer of the subtype clustering block is set to the number of subtypes, with the activation function being Softmax. Softmax transforms the final output of the subtype clustering block into the probability that a sample belongs to each subtype, obtaining a soft label matrix containing subtype category information. The formula for this process is as follows: in Represents two nonlinear transformations. The soft label matrix representing the output of the subtype clustering blocks; S603. The clustering process relies on contrastive learning of the columns of the soft-label matrix; the similarity between different clusters is calculated using cosine similarity, and then the contrastive loss is calculated. Optimization is performed; the formula for this process is as follows: in , This represents the probability that the nth sample is assigned to the jth subtype; For the first j Contrast loss between one subtype and the other subtypes; For temperature coefficient, q These are hyperparameters that control the shape of the loss function. C Number of subtype categories; Finally, the subtype categories of all samples were obtained.

4. A system for cancer subtype classification based on multi-omics data according to any one of claims 1-3, characterized in that, include: The multi-omics data acquisition and preprocessing module is used to collect multi-omics data of cancer samples from data sources and perform preprocessing operations on the multi-omics data; The primary feature extraction module is used to construct an autoencoder for single-omics data and extract primary features to obtain a multi-omics low-dimensional feature matrix. The secondary feature extraction module is used to divide the omics data into a baseline omics dataset and several supplementary omics datasets, and to combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; secondary feature extraction is performed on the multi-omics low-dimensional feature matrix through the autoencoder pairs to obtain a cross-omics shared feature matrix; including: S401. Combine the baseline omics autoencoder and each supplementary omics autoencoder to obtain several coupled autoencoder pairs; for any autoencoder pair, input the supplementary omics low-dimensional feature matrix obtained in step S3 into the baseline omics decoder, and then perform secondary reconstruction through the baseline omics encoder to obtain the cross-omics shared feature matrix. For the m Supplemental omics data to baseline omics data t The fusion process is expressed by the following formula: in, The encoder representing the autoencoder of baseline omics data. The decoder represents the autoencoder of baseline omics data. Representing the m The secondary reconstruction feature matrix of omics data Representing the m Shared feature matrix of omics data; Representing the m Low-dimensional feature matrix of omics data Representing omics data; Representing the m An encoder for an autoencoder of omics data; S402. The process in step S401 involves cross-omics cyclic loss. The constraints are applied, and the formula is expressed as follows: in, Representative omics data m With omics data t Cyclic losses between them; The multi-omics integration module is used to integrate a set of cross-omics shared features into a matrix and input it into the subtype clustering module; The subtype clustering module consists of a projection block composed of a deep neural network and a soft label output layer. By performing data augmentation on the input feature matrix and implementing a contrastive learning strategy in the column space, it automatically completes the optimization process of subtype clustering and outputs subtype labels. Then, it transmits the subtype label of the sample to the display module. The display module is used to show the subtype classification results of the samples.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the cancer subtype classification method based on multi-omics data as described in any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the cancer subtype classification method based on multi-omics data as described in any one of claims 1 to 3.

Images