Gene multi-omics data fusion analysis method and system based on graph network
The graph network method solves the heterogeneity and complexity problems in multi-omics data fusion analysis, provides a unified analysis framework and visualization tool, improves the quality and reliability of data analysis, reduces the interdisciplinary threshold, and reveals that Complex biological regulatory networks and disease mechanisms.
Patent Information
- Application Number
- CN202410921228.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-10
- Publication Date
- 2024-11-01
AI Technical Summary
The fusion analysis of multiomics data faces difficulties such as data heterogeneity, data scale and complexity, and biological interpretation. It lacks a unified and easy-to-use analytical framework, and existing tools require interdisciplinary knowledge and strong programming capabilities.
Using graph network method, different omics data are mapped into the same graph model, biological entities and their interactions are modeled through graph form, three-layer heterogeneous graph convolution network is constructed, and supervised learning is combined with clinical prognostic information, providing Unified analytical framework and visualization tools.
Effectively integrate and analyze large-scale multiomics data, improve the quality and reliability of data analysis, lower interdisciplinary thresholds, reveal complex biological regulatory networks and disease mechanisms, and provide intuitive explanations of biological information.
Smart Images

Figure CN118887992A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for fusion analysis of gene multi-omics data based on graph network. Background Art
[0002] Single-omics data can only provide information on a certain level of the biological system. By integrating multi-omics data, we can achieve a comprehensive understanding of the biological system from genes to functions, providing a more comprehensive and integrated biological perspective. The fusion analysis of genomic multi-omics data is a necessary means to understand biological complexity, disease mechanisms, and promote the development of precision medicine and biotechnology. It provides unprecedented depth and breadth for biomedical research and applications by integrating and analyzing a large amount of data at different omics levels.
[0003] However, the fusion analysis of multi-omics data faces difficulties such as data heterogeneity, data scale and complexity, and complexity of biological interpretation. First, different omics data types have significant differences in data format, dimension, and processing method, which brings challenges to data integration. How to effectively integrate these heterogeneous data is an important issue in the fusion analysis of multi-omics data. Second, with the development of high-throughput sequencing technology, the amount of omics data produced has increased dramatically, and the scale and complexity of the data have become bottlenecks for processing and analysis. How to efficiently store, manage, and analyze these large-scale data requires high-performance computing resources and optimized algorithms. Data generated by different platforms and experimental conditions may differ in quality and consistency, and these differences may affect the results and reliability of the final analysis. Therefore, how to design standardized experiments and data processing processes to ensure data quality and consistency is another key issue. Third, although there are many tools and software available for the analysis of single omics data, there is a lack of a unified and easy-to-use analysis framework to integrate and analyze multi-omics data. This requires researchers to have not only an interdisciplinary knowledge background, but also strong programming skills. Even if multi-omics data are successfully integrated, it is still a huge challenge to extract meaningful biological information from complex data and transform it into a deep understanding of biological processes or disease mechanisms, which requires deep biological knowledge and data interpretation capabilities.
[0004] Using graph networks to fuse multi-omics data is a cutting-edge method that can effectively address the difficulties of multi-omics data fusion analysis mentioned above. Graph networks model data in the form of graphs, where nodes represent biological entities (such as genes, proteins, etc.) and edges represent relationships between entities (such as interactions, regulatory relationships, etc.). The advantages of this method in multi-omics data fusion analysis are mainly reflected in the following four aspects: First, graph networks are naturally suitable for processing heterogeneous data because they can map different types of omics data into the same graph model. Biological entities at different omics levels and their interactions can be naturally represented as nodes and edges in the graph, thereby achieving seamless integration of different data types. Second, the powerful computing power of graph networks enables them to effectively process large-scale omics data. Through graph compression and subgraph sampling techniques, the consumption of computing resources can be reduced while retaining important biological information, thereby effectively processing and analyzing large-scale and complex multi-omics data. By integrating multi-source omics data, graph networks can naturally identify and correct inconsistencies in the data. For example, by comparing the differences between genomics and transcriptomics data, errors or deviations that may occur during data processing can be identified, thereby improving the quality and reliability of data analysis. Third, graph networks provide a unified analysis framework for multi-omics data. By defining common graph operations and algorithms, different biological problems can be flexibly addressed. In addition, with the development of technologies such as graph neural networks, more user-friendly tools are provided for non-professionals, lowering the threshold for interdisciplinary knowledge. Fourth, graph networks can reveal complex biological regulatory networks and disease mechanisms by simulating entities and their interactions in biological systems. Compared with traditional data analysis methods, graph networks provide a more intuitive way to understand and explain biological phenomena in multi-omics data.
[0005] Therefore, this application proposes a method and system for fusion analysis of genetic multi-omics data based on graph networks to solve the problem of multi-omics data analysis in clinical research and biological research, and provide new methods and tools for research. Summary of the invention
[0006] The purpose of this application is to provide a method and system for fusion analysis of genomic multi-omics data based on graph networks, aiming to solve the above problems.
[0007] To achieve the above objectives, this application provides the following technical solutions:
[0008] The present application provides a method for fusion analysis of genomic multi-omics data based on a graph network, including:
[0009] Acquiring multi-omics data, and preprocessing the multi-omics data;
[0010] Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network.
[0011] Inputting nodes and edges into the graph network for training;
[0012] Verify the accuracy of the graph network in predicting tumor prognosis;
[0013] Node features and edge weights are extracted from the trained graph network, and the training results of the graph network are visualized based on the node features and edge weights to complete multi-omics data fusion analysis.
[0014] Furthermore, in the step of assigning node names and building edge relationships based on the preprocessed multi-omics data, and then building a graph network, the following steps are specifically included:
[0015] Obtaining a dataset including protein-protein interaction information from a database;
[0016] Screening the protein quantification files and phosphorylated protein quantification files in the data set, retaining the corresponding genes, and assigning corresponding node names;
[0017] For transcriptome data in multi-omics data, edges are constructed by mapping genes to corresponding pathways and determining the gene pair combinations within each pathway;
[0018] The edges of protein and phosphorylated protein data in multi-omics data are established based on the interactions recorded in the STRING database;
[0019] Edges are also constructed between RNA-protein pairs and protein-phosphorylated protein pairs.
[0020] Furthermore, in the step of inputting nodes and edges into the graph network for training, the following steps are specifically included:
[0021] The graph network is a three-layer heterogeneous graph convolutional network targeting transcriptome, proteome and phosphorylated proteome respectively; each layer corresponds to an omics entity, and each node represents a gene;
[0022] Based on the feature matrix x∈RN xp, the graph network averages the information around the node, where N is the number of nodes and p is the number of patients;
[0023] The graph network also includes a classification layer for mapping the aggregated and processed graph features to patient prognosis; the classification layer outputs the prediction result after normalization, linear transformation and nonlinear activation function.
[0024] Furthermore, the graph network training process also includes: using tumor prognosis data to supervise the graph network learning;
[0025] By using prognostic data as supervisory signals, the graph network learns multi-omics features corresponding to clinical outcomes.
[0026] Furthermore, in the step of verifying the accuracy of the graph network in predicting tumor prognosis, the following steps are specifically included:
[0027] Using the cross-validation method, patients were randomly assigned to the acquired training and validation sets;
[0028] Generate a deep learning score for the corresponding patient through the graph network to predict the prognosis information;
[0029] The network model generated by the cross-validation method was evaluated by calculating the C index, where the network model iteration with the highest C index in the validation phase was determined as the final model configuration;
[0030] Dividing the patients into a high-risk group and a low-risk group according to the prognostic information obtained by the deep learning score;
[0031] Through Kaplan-Meier survival analysis, whether there is a survival difference between the high-risk group and the low-risk group is tested; if there is a survival difference, it is judged that the graph network training is successful; if not, the number of training rounds is increased and training is continued.
[0032] Furthermore, the step of preprocessing the multi-omics data specifically includes the following steps:
[0033] Preprocess the mRNA expression, protein quantification and phosphorylated protein quantification files corresponding to the patients to eliminate the variability between the data;
[0034] During the preprocessing, a specific strategy was adopted for missing values, and the missing values were replaced with 0;
[0035] Z scores were calculated for the data of mRNA, protein and phosphorylated protein respectively, and the original data were converted into a distribution with 0 as the mean and 1 as the standard deviation.
[0036] The present application provides a genomic multi-omics data fusion analysis system based on a graph network, comprising:
[0037] Acquisition module: acquiring multi-omics data and preprocessing the multi-omics data;
[0038] Construction module: Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network;
[0039] Training module: inputting nodes and edges into the graph network for training;
[0040] Verification module: verifying the accuracy of the graph network in predicting tumor prognosis;
[0041] Visualization module: extract node features and edge weights from the trained graph network, visualize the training results of the graph network based on the node features and edge weights, and complete multi-omics data fusion analysis.
[0042] The present application provides a device, comprising a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing a method for fusion analysis of genomic multi-omics data based on a graph network; the processor is used to execute the program instructions stored in the memory to implement a method for fusion analysis of genomic multi-omics data based on a graph network.
[0043] The present application provides a storage medium storing program instructions executable by a processor, wherein the program instructions are used to execute a method for fusion analysis of genomic multi-omics data based on a graph network.
[0044] The present application provides a method and system for fusion analysis of genomic multi-omics data based on graph networks, which has the following beneficial effects:
[0045] (1) The use of graph network structures, through the representation of nodes and edges, can naturally simulate these complex relationships between biomolecules, intuitively represent the interactions between biomolecules such as genes and proteins, and make multi-omics fusion more logical;
[0046] (2) Biological prior knowledge is used to construct node-edge relationships, ensuring that the network structure constructed by the model is not only data-driven but also reflects the real biological process, thus enhancing the biological interpretability of the model;
[0047] (3) Combined with clinical prognostic information supervised model learning, multi-omics data and clinical prognostic information are combined to guide the training of graph networks through supervised learning to discover disease-related biomarkers and potential therapeutic targets, thereby improving the efficiency and accuracy of this application. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A schematic diagram of a flow chart of a method for fusion analysis of genomic multi-omics data based on a graph network according to Example 1 of the present application;
[0049] Figure 2 This is a schematic diagram of the multi-dataset experimental verification results of Example 1 of the present application;
[0050] Figure 3 This is a schematic diagram of the structure of a genomic multi-omics data fusion analysis system based on a graph network according to Example 2 of the present application;
[0051] Figure 4 This is a schematic diagram of the device structure of Example 3 of the present application;
[0052] Figure 5 This is a schematic diagram of the storage medium structure of Example 4 of the present application. DETAILED DESCRIPTION
[0053] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0054] The following is an analysis of the existing technical solutions in combination with relevant technologies.
[0055] The fusion analysis of genomic multi-omics data involves integrating data from different omics levels to obtain a more comprehensive understanding of biological systems. Currently, a variety of methods and strategies have been developed for the fusion analysis of multi-omics data, each with its own specific application scenarios and advantages. The most direct fusion method is the serial fusion method, which simply splices data from different sources by features to form a large feature matrix. The advantage of this method is that it is simple to operate, but it may cause the information of one type of omics data to mask the information of other omics data. Similarity network fusion is a network fusion technique used to integrate multiple types of omics data. It constructs a similarity network for each type of omics data and then fuses these networks into a comprehensive network, thereby comprehensively considering the similarities and differences between various data. This method is particularly suitable for finding disease subtypes or classifying diseases. The subspace integration method fuses data by finding shared or complementary information subspaces in different data sets. This method aims to identify common change patterns across omics data and can effectively deal with data heterogeneity and dimensionality disaster problems. Multi-level integrated analysis is a flexible statistical framework designed to extract potential factors from multi-omics data sets. These factors are able to capture variation across multiple omics levels, providing a powerful tool for understanding complex biological processes.
[0056] Model-driven fusion is a major direction in multi-omics fusion methods. In recent years, machine learning and deep learning technologies have been increasingly used in multi-omics data fusion. By designing complex neural network structures, such as autoencoders or deep neural networks, complex relationships and patterns between different omics data can be automatically learned. In model-driven fusion, prior biological knowledge and models are used to guide the data fusion process. For example, transcriptome and proteome data can be integrated based on known metabolic pathways or protein interaction networks. This approach can improve the biological interpretability of data fusion. As mentioned earlier, graph networks are a cutting-edge method that uses graph network algorithms to analyze and integrate information by mapping different omics data to nodes and edges in a graph. This method is particularly suitable for revealing complex interactions and regulatory networks between biological entities.
[0057] Methods such as serial fusion, similarity network fusion, subspace integration, multi-level integrated analysis, and machine learning model-driven fusion are the main methods for dealing with multi-omics data fusion. The following discusses the shortcomings of these methods and their deficiencies relative to graph network fusion: Serial fusion may ignore the complex interactions between data by simply connecting different types of data, resulting in information loss. The magnitude and distribution of different omics data may vary greatly. Simple concatenation may cause certain data types to dominate the analysis, thereby ignoring other important information. Similarity network construction and fusion Similarity networks require highly complex calculations, especially when dealing with large-scale data sets. For non-network structured data, converting them into similarity networks may cause information distortion. The selection of subspaces is often based on mathematical or statistical criteria, which may make the final results difficult to interpret from a biological perspective. The integration of different subspaces may ignore the consistency issues between data, resulting in reduced reliability of the integrated results. Multi-level integrated analysis requires data processing and integration at different levels, which greatly increases the computational cost and time. Designing effective multi-level integration strategies requires in-depth expertise and complex algorithm design. Ordinary machine learning model-driven fusion has the risk of overfitting, especially when the amount of data is relatively small. Complex machine learning models may cause overfitting and affect the generalization ability of the model.
[0058] Compared with the above methods, graph network fusion can effectively cope with the complexity and heterogeneity of multi-omics data. Graph networks can naturally represent different types of data and their interactions, providing a more comprehensive view of the data. Graph networks can adapt to data sets of different sizes and complexities, and are particularly suitable for processing large-scale multi-omics data.
[0059] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.
[0060] Example 1
[0061] See also Figure 1 , is a flow chart of a method for fusion analysis of multi-omics data based on a graph network in Example 1 of the present application; the steps include:
[0062] S1: Acquire multi-omics data, and pre-process the multi-omics data.
[0063] In this example, to ensure the reliability of comparisons between different biological data types, the mRNA expression, protein quantification, and phosphorylated protein quantification files corresponding to the patients were first preprocessed to eliminate the variability between the data. During the processing, a specific strategy was adopted for missing values: missing values were replaced with 0, and for those columns with a missing value ratio of more than 20%, it was decided to completely remove these columns from the data set based on the consideration of ensuring the overall quality and reliability of the data.
[0064] Subsequent data processing steps include strict data deduplication operations to ensure that there are no duplicate indexes within each data set and to maintain the uniqueness and neatness of the data. In order to effectively standardize the data, Z-scores are calculated for mRNA, protein, and phosphorylated protein data. By converting the raw data into a distribution with a mean of 0 and a standard deviation of 1, it aims to eliminate the influence of data dimensions under different experimental conditions and improve the comparability between data, thus laying a solid foundation for subsequent statistical analysis and biological interpretation. This method not only enhances the professionalism of data processing, but also provides high-quality data support for the integrated analysis of complex biological information.
[0065] S2: Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network.
[0066] In this example, a dataset including protein-protein interaction information was obtained from the STRING database (version 11.0b); and only data with interaction scores of 700 or above were taken into consideration.
[0067] Subsequently, the protein quantification files and phosphorylated protein quantification files in the data set were screened, and only those genes that appeared in the PPI network were retained and assigned corresponding node names to facilitate subsequent analysis. This step is crucial for constructing a graphical structure that can accurately reflect biological interactions at the molecular level.
[0068] The criteria for constructing edges between nodes are mainly based on two aspects: one is based on the biological pathways defined by the Kyoto Encyclopedia of Genes and Genomes (KEGG); the other is based on the direct interactions provided by the STRING database. For transcriptome data in multi-omics data, edges are constructed by mapping genes to corresponding pathways and determining the gene pair combinations within each pathway; this process reflects the complex interactions between genes involved in specific biological processes.
[0069] Edges for protein and phosphoprotein data in multi-omics data are established based on interactions recorded in the STRING database; these interactions indicate direct physical or functional connections between molecules.
[0070] In order to connect different omics levels, edges are also constructed between RNA-protein pairs and protein-phosphorylated protein pairs. This highlights the multi-level regulatory mechanisms and signaling pathways that exist in the regulation of cancer cell function. This cross-level edge construction is crucial for developing a comprehensive network model that integrates multiple molecular data types, providing deep insights into the multifaceted nature of cancer biology.
[0071] In general, three types of nodes were defined for each dataset, corresponding to three types of omics data, and five types of edges (including three intra-group and two inter-group edges), in order to construct a comprehensive and precise biological network model, revealing the intricate intrinsic connections related to various cancers in multi-omics datasets containing mRNA, protein, and phosphorylated protein quantification.
[0072] S3: Input the nodes and edges into the graph network for training.
[0073] In this embodiment, the graph network is a three-layer heterogeneous graph convolutional network for the transcriptome, proteome and phosphorylated proteome respectively; each layer corresponds to an omics entity, and each node represents a gene; the transcriptome is labeled as hri, the proteome is labeled as hpi, and the phosphorylated proteome is labeled as hphi.
[0074] Based on the feature matrix x∈RN xp, the graph network averages the information around the nodes, where N is the number of nodes and p is the number of patients; this graph construction method is deeply rooted in biology and helps to accurately understand the dynamics of gene interactions. The graph learning process is supervised by recurrence-free survival (DFS) or cancer-related death (OS) in order to explore tumor prognosis and extract clinical information closely related to tumor outcomes, avoiding the interference effect caused by large-scale omics variations.
[0075] The flow of information in the graph is facilitated by the HeteroRGCN model in the Deep Graph Library, which utilizes two layers of relational graph convolutions. This model allows features to be transformed across nodes in the network, propagating and aggregating information through edge-specific transformations, aiming to respect the heterogeneity of the data and the biological basis behind it.
[0076] Attention mechanisms are implicitly incorporated into the graph convolution process through nn.Sequential and nn.Conv2d layers, customized for each edge type. This design enables the model to focus on key interactions by dynamically weighting the importance of edges during the feature aggregation stage. The convolution operations on the edges enable the network to adaptively learn which connections are most critical to the current task, thereby improving the model's ability to capture key features in the graph.
[0077] The graph network also includes a classification layer that maps the aggregated and processed graph features to patient prognosis; the classification layer outputs the prediction after operating through normalization, linear transformation, and nonlinear activation functions. The optimization process of the graph network uses a custom negative log-likelihood (NLL) loss function designed specifically for processing censored survival data. The loss calculation includes sorting the predicted risks in descending order, calculating the hazard ratio (i.e., an index of the predicted risk), and the cumulative log risk.
[0078] The graph network training process also includes: using tumor prognosis data to supervise the graph network learning. Using prognostic data to supervise the learning of multi-omics network relationships has multiple advantages. This method combines multi-omics data and clinical prognosis information to guide the training of network models through supervised learning to discover disease-related biomarkers and potential therapeutic targets. Specifically, by using prognostic data as a supervisory signal, the graph network can learn multi-omics features that are directly related to clinical outcomes such as disease progression, treatment response, and patient survival rate. This targeted learning helps improve the accuracy of the model in predicting disease prognosis. Supervised learning models can identify biological molecules and pathways that are closely related to disease prognosis, thereby revealing potential disease mechanisms. This is of great significance for understanding the pathogenesis and development of complex diseases.
[0079] S4: Validate the accuracy of the graph network in predicting tumor prognosis.
[0080] In this embodiment, a five-fold cross-validation strategy was adopted to ensure the stability of the model performance evaluation by randomly assigning patients to the acquired training set and validation set. In order to maintain a sufficient gradient of the prognostic labels between the subsets, the label homogeneity problem was effectively addressed.
[0081] The graph network can generate a deep learning score (DLS) for each patient to predict their prognostic information. The network model generated by the cross-validation method is evaluated by calculating the C index, where the network model iteration with the highest C index in the validation phase is determined as the final model configuration. The prognostic information provided by the DLS is then used to divide the patients into high-risk and low-risk groups. A Kaplan-Meier survival analysis is performed to test whether there is a significant survival difference between the two risk groups; if there is a survival difference, the graph network training is judged to be successful; if not, the number of training rounds is increased and training continues.
[0082] To ensure the reliability and repeatability of the model, the Pearson correlation coefficient was calculated for the edge weights obtained for each validation group, whether in the five-fold or three-fold cross-validation framework, to verify the consistency of the edge weight distribution in different model iterations, thereby demonstrating the stability of the model in capturing the underlying patterns related to generalized patient prognosis.
[0083] S5: Extract node features and edge weights from the trained graph network, visualize the training results of the graph network based on the node features and the edge weights, and complete multi-omics data fusion analysis.
[0084] In this embodiment, after the model is trained, each edge learns a corresponding edge weight, which can evaluate the importance of the relationship between the two nodes connected by the edge. The edge weights of all edges and the feature values of all nodes can be extracted from the model. Based on the visualization of edge weights and node features, the visualization of multi-omics fusion network can be completed.
[0085] To gain a deeper understanding of the dynamic processes of transcription and post-transcriptional modifications in trans-omics networks, we focused on analyzing the interactions between RNA and protein (i.e., RP edges) and between protein and phosphorylated protein (i.e., PP edges). We performed a detailed statistical comparison of the RP and PP edge weights of each gene in all patients across different cancer types. When the RP and PP weights of genes in the training and validation sets showed statistically significant differences (p<0.05), these genes were considered significant at the trans-omics level.
[0086] See also Figure 2 , which is a schematic diagram of the experimental verification results of multiple data sets of Example 1 of this application. This example was experimented and verified on colorectal cancer, glioblastoma, pediatric low-grade glioma, and pancreatic cancer data sets, and the average C-index of the cross-validation of the prognosis prediction of this application reached 0.91, 0.71, 0.74, and 0.74, respectively.
[0087] In summary, Example 1 of the present application preprocesses multi-omics data to remove redundant information and invalid information and reduce the differences between data; assigns node names based on multi-omics data, constructs edge relationships, and builds a graph network; imports nodes and edges to train the graph network, and uses tumor prognosis data to supervise model learning to ensure that the graph network learns intra-group and inter-group network relationships that are strongly correlated with tumor prognosis; then uses cross-validation to test the graph network effect, and uses C-index to test the accuracy of the graph network in predicting tumor prognosis to ensure the effectiveness of the network fusion of multi-omics data; finally, extracts node features and edge weights from the trained graph network, visualizes the learning results of the graph network, sorts the edge weights, determines the intra-group and inter-group network relationships that are strongly correlated with tumor prognosis, and completes multi-omics fusion analysis.
[0088] Example 2
[0089] See also Figure 3 , is a schematic diagram of the structure of a genomic multi-omics data fusion analysis system based on a graph network in Example 2 of the present application; the specific contents include:
[0090] Acquisition module: acquiring multi-omics data and preprocessing the multi-omics data;
[0091] Construction module: Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network;
[0092] Training module: inputting nodes and edges into the graph network for training;
[0093] Verification module: verifying the accuracy of the graph network in predicting tumor prognosis;
[0094] Visualization module: extract node features and edge weights from the trained graph network, visualize the training results of the graph network based on the node features and edge weights, and complete multi-omics data fusion analysis.
[0095] Example 3
[0096] See also Figure 4 , is a schematic diagram of the device structure of Embodiment 3 of the present application. The device 50 includes a processor 51 and a memory 52 coupled to the processor 51 .
[0097] The memory 52 stores program instructions for implementing the above-mentioned graph network-based genomic multi-omics data fusion analysis method.
[0098] The processor 51 is used to execute program instructions stored in the memory 52 to implement a genomic multi-omics data fusion analysis based on a graph network.
[0099] The processor 51 may also be referred to as a CPU (Central Processing Unit).
[0100] The processor 51 may be an integrated circuit chip with signal processing capabilities. The processor 51 may also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0101] Example 4
[0102] See also Figure 5, which is a schematic diagram of the structure of the storage medium of Example 4 of the present application. The storage medium of the embodiment of the present application stores a program file 61 that can implement all the above methods, wherein the program file 61 can be stored in the above storage medium in the form of a software product, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or a computer, a server, a mobile phone, a tablet, and other devices.
[0103] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
[0104] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.
[0105] Although the embodiments of the present application have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the appended claims and their equivalents.
[0106] Of course, the present invention may have many other implementations. Based on this implementation, other implementations obtained by ordinary technicians in this field without any creative work are all within the scope of protection of the present invention.
Claims
1. A method for fusion analysis of genomic multi-omics data based on graph network, characterized in that: include: Acquiring multi-omics data, and preprocessing the multi-omics data; Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network. Inputting nodes and edges into the graph network for training; Verify the accuracy of the graph network in predicting tumor prognosis; Node features and edge weights are extracted from the trained graph network, and the training results of the graph network are visualized based on the node features and edge weights to complete multi-omics data fusion analysis.
2. The method for fusion analysis of genomic multi-omics data based on graph network according to claim 1, characterized in that: In the step of assigning node names and building edge relationships based on the preprocessed multi-omics data, and then building a graph network, the following steps are specifically included: Obtaining a dataset including protein-protein interaction information from a database; Screening the protein quantification files and phosphorylated protein quantification files in the data set, retaining the corresponding genes, and assigning corresponding node names; For transcriptome data in multi-omics data, edges are constructed by mapping genes to corresponding pathways and determining the gene pair combinations within each pathway; The edges of protein and phosphorylated protein data in multi-omics data are established based on the interactions recorded in the STRING database; Edges are also constructed between RNA-protein pairs and protein-phosphorylated protein pairs.
3. The method for fusion analysis of genomic multi-omics data based on graph network according to claim 1, characterized in that: The step of inputting nodes and edges into the graph network for training specifically includes the following steps: The graph network is a three-layer heterogeneous graph convolutional network targeting transcriptome, proteome and phosphorylated proteome respectively; each layer corresponds to an omics entity, and each node represents a gene; Based on the feature matrix x∈RN xp, the graph network averages the information around the node, where N is the number of nodes and p is the number of patients; The graph network also includes a classification layer for mapping the aggregated and processed graph features to patient prognosis; the classification layer outputs the prediction result after normalization, linear transformation and nonlinear activation function.
4. The method for fusion analysis of genomic multi-omics data based on graph network according to claim 3, characterized in that: The graph network training process also includes: using tumor prognosis data to supervise the graph network learning; By using prognostic data as supervisory signals, the graph network learns multi-omics features corresponding to clinical outcomes.
5. The method for fusion analysis of genomic multi-omics data based on graph network according to claim 1, characterized in that: The step of verifying the accuracy of the graph network in predicting tumor prognosis specifically includes the following steps: Using the cross-validation method, patients were randomly assigned to the acquired training and validation sets; Generate a deep learning score for the corresponding patient through the graph network to predict the prognosis information; The network model generated by the cross-validation method was evaluated by calculating the C index, where the network model iteration with the highest C index in the validation phase was determined as the final model configuration; Dividing the patients into a high-risk group and a low-risk group according to the prognostic information obtained by the deep learning score; By using Kaplan-Meier survival analysis, whether there is a survival difference between the high-risk group and the low-risk group is tested; if there is a survival difference, it is judged that the graph network training is successful; If not, increase the number of training rounds and continue training.
6. The method for fusion analysis of genomic multi-omics data based on graph network according to claim 1, characterized in that: The step of preprocessing the multi-omics data specifically includes the following steps: Preprocess the mRNA expression, protein quantification and phosphorylated protein quantification files corresponding to the patients to eliminate the variability between the data; During the preprocessing, a specific strategy was adopted for missing values, and the missing values were replaced with 0; Z scores were calculated for the data of mRNA, protein and phosphorylated protein respectively, and the original data were converted into a distribution with 0 as the mean and 1 as the standard deviation.
7. A system for a method for fusion analysis of genomic multi-omics data based on graph network according to claim 1, characterized in that: include: Acquisition module: acquiring multi-omics data and preprocessing the multi-omics data; Construction module: Based on the preprocessed multi-omics data, node names are assigned and edge relationships are constructed to build a graph network; Training module: inputting nodes and edges into the graph network for training; Verification module: verifying the accuracy of the graph network in predicting tumor prognosis; Visualization module: extract node features and edge weights from the trained graph network, visualize the training results of the graph network based on the node features and edge weights, and complete multi-omics data fusion analysis.
8. A device, characterized in that The device includes a processor and a memory coupled to the processor, wherein the memory stores program instructions for implementing a method for fusion analysis of genomic multi-omics data based on a graph network as described in any one of claims 1-6; and the processor is used to execute the program instructions stored in the memory to implement a method for fusion analysis of genomic multi-omics data based on a graph network.
9. A storage medium, characterized in that: The method stores program instructions executable by a processor, wherein the program instructions are used to execute a method for fusion analysis of genomic multi-omics data based on a graph network as described in any one of claims 1 to 6.
Citation Information
Cited By
Plant phenotype multi-omics prediction method and system
CN118983005A
