Multi-omics data processing method and apparatus, storage medium, and program product

Abstract

A multi-omics data processing method and apparatus, a storage medium, and a program product. The method comprises: acquiring multi-omics data and pathway data; preprocessing the multi-omics data to obtain feature data of a plurality of omics; performing graph representation learning on the pathway data to obtain a pathway representation vector; and constructing a first data processing model on the basis of the feature data of the plurality of omics and the pathway representation vector, wherein an input layer of the first data processing model comprises the feature data of the plurality of omics, the first data processing model comprises at least a first structure, and the first structure comprises the pathway representation vector of a pathway related to the multi-omics data.

Description

Multi-omics data processing methods, devices, storage media and software products Technical Field

[0001] This disclosure relates to, but is not limited to, the field of bioinformatics, and particularly to a multi-omics data processing method and apparatus, storage medium, and program product. Background Technology

[0002] Cancer is a highly complex disease, and its occurrence and development involve intricate regulation of multiple levels of omics molecules, including gene mutations, abnormal gene expression, copy number variations, epigenetic alterations, and abnormal protein expression. This complexity makes single-omics data increasingly limited in identifying pathogenic factors. Multi-omics data can more comprehensively reveal the molecular mechanisms of cancer, providing precise diagnostic and treatment options for cancer patients. Therefore, integrating omics data from different levels, such as genomics, transcriptomics, methylomes, proteomics, and metabolomics, has become an important trend in cancer research.

[0003] In practical applications, multi-omics data from cancer patients often exhibit a significantly higher number of features than the number of samples. When using machine learning models for classification tasks, this can lead to overfitting and lower model accuracy. Therefore, it is necessary to integrate high-dimensional multi-omics data to achieve dimensionality reduction.

[0004] There are three main categories of multi-omics data integration methods in related technologies: (1) statistical integration methods, such as principal component analysis and matrix factorization; (2) integration methods based on traditional machine learning, such as clustering algorithms and random forests; and (3) integration methods based on deep learning, such as autoencoders, multilayer feedforward neural networks, and multi-view deep learning models.

[0005] However, these multi-omics data integration methods have the following drawbacks: these existing models are not specifically designed for multi-omics data integration, and therefore lack biological interpretability. Furthermore, because inter-omics associations are relatively abstract, they are often difficult to incorporate into the multi-omics data integration process. Therefore, existing multi-omics data integration methods rarely consider inter-omics associations, especially the important information of pathways. Summary of the Invention

[0006] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0007] This disclosure provides a multi-omics data processing method, comprising: acquiring multi-omics data and pathway data; preprocessing the multi-omics data to obtain feature data of multiple omics; performing graph representation learning on the pathway data to obtain pathway representation vectors; and constructing a first data processing model based on the feature data of the multiple omics and the pathway representation vectors, wherein the input layer of the first data processing model includes the feature data of the multiple omics, and the first data processing model includes at least a first structure, the first structure including pathway representation vectors of the pathways involved in the multi-omics data.

[0008] This disclosure also provides a multi-omics data processing apparatus, including a memory; and a processor connected to the memory, the memory being used to store instructions, the processor being configured to execute the steps of the multi-omics data processing method according to any embodiment of this disclosure based on the instructions stored in the memory.

[0009] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-omics data processing method described in any embodiment of this disclosure.

[0010] This disclosure also provides a program product including instructions that, when executed by a computer, perform a multi-omics data processing method as described in any embodiment of this disclosure.

[0011] This disclosure also provides a multi-omics data processing apparatus, including a data acquisition module, an omics processing module, a pathway learning module, and a model building module, wherein: the data acquisition module is configured to acquire multi-omics data and pathway data; the omics processing module is configured to preprocess the acquired multi-omics data to obtain feature data of multiple omics; the pathway learning module is configured to perform graph representation learning on the acquired pathway data to obtain pathway representation vectors; the model building module is configured to construct a first data processing model based on the feature data of the multiple omics and the pathway representation vectors, wherein the input layer of the first data processing model includes the feature data of the multiple omics, and the first data processing model includes at least a first structure, the first structure including pathway representation vectors of the pathways involved in the multi-omics data.

[0012] The multi-omics data processing method, apparatus, storage medium, and program product of this disclosure construct a first data processing model based on feature data and pathway representation vectors of multiple omics. The input layer of the first data processing model includes feature data of multiple omics, and the first data processing model includes at least a first structure. The first structure includes pathway representation vectors of pathways involved in the multi-omics data. The pathway information is used to guide the design and improvement of the first data processing model, which effectively solves the problem that existing multi-omics data integration models lack biological significance and interpretability, and improves the biological interpretability of the first data processing model.

[0013] Other features and advantages of this disclosure will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the disclosure. Other advantages of this disclosure may be realized and obtained by means of the methods described in the description and the accompanying drawings. Attached Figure Description

[0014] The accompanying drawings are used to provide an understanding of the technical solutions of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the technical solutions of this disclosure and do not constitute a limitation on the technical solutions of this disclosure.

[0015] Figure 1 is a flowchart illustrating a multi-omics data processing method provided by an exemplary embodiment of this disclosure.

[0016] Figure 2 is a flowchart illustrating another multi-omics data processing method provided by an exemplary embodiment of this disclosure.

[0017] Figure 3 is a schematic diagram of a quantification process of inter-pathway interaction provided by an exemplary embodiment of this disclosure.

[0018] Figure 4 is a schematic diagram of matrix decomposition of matrix M provided by an exemplary embodiment of the present disclosure.

[0019] Figure 5 is a schematic diagram of the structural design and loss function design of an autoencoder model based on path information guided by an exemplary embodiment of this disclosure.

[0020] Figure 6 is a schematic diagram of a structure design of a multilayer feedforward neural network based on path information, provided by an exemplary embodiment of this disclosure.

[0021] Figure 7 is a schematic diagram of the structural design and loss function design of a hypergraph neural network model guided by path information, provided by an exemplary embodiment of this disclosure.

[0022] Figure 8 is a schematic diagram of the structural design of a multi-view learning model provided by an exemplary embodiment of this disclosure.

[0023] Figure 9 is a schematic diagram of the structure of a multi-omics data processing device provided by an exemplary embodiment of the present disclosure.

[0024] Figure 10 is a schematic diagram of another multi-omics data processing apparatus provided by an exemplary embodiment of the present disclosure. Detailed Implementation

[0025] This disclosure describes several embodiments, but these descriptions are exemplary and not limiting, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0026] This disclosure includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this disclosure may also be combined with any conventional features or elements to form a unique inventive scheme as defined by the claims. Any feature or element of any embodiment may also be combined with features or elements from other inventive schemes to form another unique inventive scheme as defined by the claims. Therefore, it should be understood that any feature shown and / or discussed in this disclosure may be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes may be made within the scope of the appended claims.

[0027] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that the method or process does not depend on the specific order of steps described herein. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims relating to the method and / or process should not be limited to the steps performed in the order written, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments disclosed herein.

[0028] As shown in Figure 1, this disclosure provides a multi-omics data processing method, including:

[0029] Step 101: Obtain multi-omics data and pathway data;

[0030] Step 102: Preprocess the acquired multi-omics data to obtain feature data of multiple omics;

[0031] Step 103: Perform graph representation learning on the acquired pathway data to obtain pathway representation vectors;

[0032] Step 104: Construct a first data processing model based on the feature data and pathway representation vectors of multiple omics, wherein the input layer of the first data processing model includes feature data of multiple omics, and the first data processing model includes at least a first structure, which includes pathway representation vectors of the pathways involved in the multiple omics data.

[0033] The multi-omics data processing method of this disclosure constructs a first data processing model based on feature data and pathway representation vectors of multiple omics. The input layer of the first data processing model includes feature data of multiple omics. The first data processing model includes at least a first structure, which includes pathway representation vectors of pathways involved in the multi-omics data. The pathway information is used to guide the design and improvement of the first data processing model, effectively solving the problem that existing multi-omics data integration models lack biological significance and interpretability, and improving the biological interpretability of the first data processing model.

[0034] In some exemplary embodiments, the first data processing model may be a multi-omics data integration model or a disease analysis model. The multi-omics data integration model may be used to classify and integrate feature data from multiple omics. The disease analysis model may be used for at least one of the following: probability analysis of a patient’s disease, probability analysis of a patient’s expected treatment effectiveness, and probability analysis of residual lesions. However, this disclosure does not limit this.

[0035] The multi-omics data processing method disclosed herein can be used for classification and integration of multi-omics data, and can also be used for other application tasks based on multi-omics data. For example, it can be used for disease analysis tasks, including but not limited to: probability analysis of patients' disease, probability analysis of patients' expected treatment effectiveness, probability analysis of residual lesions, etc.

[0036] In some exemplary embodiments, the multi-omics data acquired in step 101 may include omics data such as genomics, transcriptomics, proteomics, and metabolomics.

[0037] For example, taking the use of multi-omics data for cancer classification, this step involves collecting multi-omics data, such as genomics, transcriptomics, proteomics, and metabolomics, from samples of cancer patients and healthy individuals. This data typically contains a wealth of biomolecular information, such as gene expression levels, protein abundance, and metabolite concentrations.

[0038] In some exemplary embodiments, the acquired pathway data includes multiple biomolecules and pathway types between the multiple biomolecules.

[0039] In step 101, known pathway data can be obtained from pathway databases (such as KEGG, Reactome, etc.). The pathway data includes biomolecules contained in the pathway, such as genes, proteins, metabolites, etc., as well as pathway annotation information. The type of pathway can be obtained based on the pathway annotation information.

[0040] In some exemplary embodiments, in step 102, the feature data of multiple omics include at least two of the following: gene mutation, copy number variation, DNA methylation, gene expression level, protein abundance, and metabolite concentration; however, this disclosure does not limit this.

[0041] In some exemplary embodiments, step 102 involves preprocessing the acquired multi-omics data, including:

[0042] Quality control processing was performed on the acquired multi-omics data;

[0043] Preliminary bioinformatics analysis was performed on the multi-omics data after quality control processing to extract multiple initial feature data;

[0044] The extracted initial feature data are standardized or normalized.

[0045] In this embodiment, bioinformatics tools can be used to perform quality control processing on the data to remove noise, outliers, etc. Then, preliminary bioinformatics analysis is performed on the multi-omics data, extracting multiple initial feature data such as gene mutations, copy number variations, DNA methylation, gene expression levels, protein abundance, and metabolite concentrations through sequence alignment, methylation information extraction, and quantitative gene expression analysis. Simultaneously, these initial feature data are standardized or normalized to ensure consistency in magnitude across different omics data.

[0046] In this embodiment of the disclosure, the feature data of multiple omics obtained through preprocessing can be represented as a matrix M(N*p), where N represents the number of samples, p represents the number of features of all omics, and N and p are both natural numbers greater than 1.

[0047] In some exemplary embodiments, in step 103, graph representation learning is performed on the acquired path data to obtain path representation vectors, including:

[0048] Construct a general graph or hypergraph using biomolecules as nodes and pathways as edges;

[0049] With pathway classification as the goal, graph representation learning is performed based on graph convolutional neural networks or hypergraph convolutional neural networks to obtain pathway representation vectors.

[0050] In a regular graph, an edge can only connect two nodes, while in a hypergraph, an edge can connect one or more nodes; such an edge is called a hyperedge. Therefore, a hypergraph can reflect higher-order information. In this embodiment, a regular graph or hypergraph is constructed using biomolecules such as genes, proteins, and metabolites as nodes and pathways as edges or hyperedges. With pathway classification as the goal, graph representation learning is performed based on a graph convolutional neural network or a hypergraph convolutional neural network to obtain the representation of the edges or hyperedges, i.e., pathway representation vectors. For example, the KEGG pathway database classifies pathways according to function, mainly including the following categories:

[0051] (1) Metabolic pathways: including carbohydrate metabolism, fat and ester metabolism, amino acid metabolism, nucleotide metabolism, etc.

[0052] (2) Disease pathways: including health-related disease pathways such as obesity, diabetes, and cancer.

[0053] (3) Genetic information processing: including DNA replication, transcription, translation, repair and other pathways related to genetic information processing.

[0054] (4) Environmental information processing: including pathways related to environmental information processing such as bacterial sensing and signal transduction.

[0055] (5) Nervous system: including pathways related to the synthesis and breakdown of neurotransmitters.

[0056] (6) Immune system: including immune cell signal transduction, cytokine synthesis and other pathways related to the immune system.

[0057] (7) Metabolite transport system: including hormone synthesis, signal transduction and other pathways related to metabolite transport.

[0058] (8) Perception system: including pathways related to sensory systems such as vision, hearing, smell, and taste.

[0059] (9) Cell cycle and apoptosis: including cell cycle regulation, apoptosis and other pathways related to cell cycle and apoptosis.

[0060] (10) Organ systems: including pathways related to the cardiovascular system, respiratory system, digestive system, urinary system, etc.

[0061] We determine the category of each pathway based on the annotation information of each pathway, assign a corresponding category label to each pathway, and use the labeled pathway data to perform graph representation learning to obtain the representation vector of each node and pathway in the graph.

[0062] Pathways are complex and sophisticated regulatory networks within the human body, playing a crucial role in maintaining normal physiological functions. However, pathways do not function in isolation; they may exhibit synergistic, complementary, or substitutive relationships. Studying the strength of these interactions is essential for disease analysis. Existing disease analyses based on multi-omics data have failed to consider the strength of these interactions.

[0063] In some exemplary embodiments, the method further includes:

[0064] Calculate the strength of the interaction relationship between pathways based on the pathway representation vector;

[0065] When constructing the first data processing model, the first data processing model includes a constraint layer and / or constraint terms, which include the calculated strength of the interaction relationships between pathways.

[0066] This disclosure, through calculating the strength of inter-pathway interactions based on pathway representation vectors, effectively quantifies these interactions, addressing the issue that existing disease analysis studies based on multi-omics data have failed to consider inter-pathway interactions. Furthermore, by incorporating the strength of inter-pathway interactions into the constraint layer and / or constraint terms of the first data processing model, redundancy in multi-omics data is reduced, mitigating the overfitting problem caused by limited sample size and high dimensionality in existing studies.

[0067] As shown in Figure 2, in some exemplary embodiments, the multi-omics data processing method includes: acquiring multi-omics data and completing data preprocessing; acquiring known pathway data and designing a method for quantifying the interaction relationships between pathways; designing and improving a multi-omics data integration model based on the quantified interaction relationships between pathways (including using known pathways and internal molecular data to guide the design of the multi-omics data integration model, and incorporating the strength of the interaction relationships between pathways as constraint information into the model objective function), thereby realizing multi-omics data integration; using multi-omics data to complete the training and testing of the model, and applying the trained model to tasks such as disease analysis.

[0068] In some exemplary embodiments, the calculated strength of the interaction between pathways can be the similarity between pathways. This disclosure calculates the similarity between pathways, i.e., the strength of the interaction between pathways, using similarity calculations. The more common nodes two graphs contain, the higher the final similarity between the learned pathways will be.

[0069] For example, the similarity between pathways can be cosine similarity; however, this disclosure does not limit it. Based on the learned pathway representation vectors, the interaction relationship between pathways can be quantified using methods such as cosine similarity.

[0070] As shown in Figure 3, this embodiment of the present disclosure provides a method for quantifying the interaction relationships between pathways, including: obtaining known pathway data from a pathway database (such as KEGG, Reactome, etc.), including biomolecules contained in the pathway, such as genes, proteins, metabolites, etc., and pathway annotation information, to obtain the pathway type; constructing a general graph or hypergraph with biomolecules such as genes, proteins, metabolites, etc. as nodes and pathways as edges or hyperedges; learning graph representation based on graph convolutional neural networks or hypergraph convolutional neural networks with pathway classification as the goal, to obtain the representation of edges or hyperedges, i.e., pathway representation vectors; and quantifying the interaction relationships between pathways based on the learned pathway representation vectors using quantization methods such as cosine similarity.

[0071] In this embodiment of the disclosure, the first data processing model may be a matrix factorization, an autoencoder, a multilayer feedforward neural network, a graph neural network, a multimodal fusion network, a multi-view learning model, etc. However, this disclosure does not limit it.

[0072] When constructing the first data processing model based on the feature data and pathway representation vectors of multiple omics, a model can be selected from matrix factorization, autoencoders, multilayer feedforward neural networks, graph neural networks, multimodal fusion networks, multi-view learning, etc., and the selected model can be designed and improved based on pathway information. This design and improvement can include: (1) using known pathways and their internal biomolecule data to guide the design of the model, increasing the biological significance and interpretability of the model; (2) incorporating the strength of the interaction relationship between pathways as a constraint term into the model objective function, realizing the mining of the correlation information between multi-omics data to reduce the redundancy of multi-omics data, thereby improving the problem of overfitting caused by multi-omics data. The preprocessed multi-omics data is used as the input of the improved model to complete the classification and integration of multi-omics data or disease analysis.

[0073] In some exemplary embodiments, when the first data processing model is a multi-omics data fusion model, after completing the classification and fusion of multi-omics data, a second data processing model can be selected, and the fused multi-omics data can be used to train the second data processing model. For example, the second data processing model can be a machine learning or deep learning model, such as a random forest, support vector machine (SVM), neural network, etc.

[0074] When training the second data processing model using integrated multi-omics data, the integrated multi-omics data can be divided into training set data, validation set data, and test set data. The training set data is used to train the second data processing model, and the parameters of the second data processing model are adjusted by optimization algorithms to minimize the loss function. The model performance is evaluated on the validation set data, and hyperparameter tuning is performed to prevent overfitting or underfitting.

[0075] In other exemplary embodiments, when the first data processing model is a multi-omics data fusion model, the multi-omics data fusion model and the selected second data processing model can also be trained as a whole model. That is, the preprocessed multi-omics data is used to train the overall model containing the multi-omics data fusion model and the second data processing model. For example, the second data processing model can be a machine learning or deep learning model, such as random forest, support vector machine (SVM), neural network, etc.

[0076] When training an overall model that includes a multi-omics data ensemble model and a second data processing model using preprocessed multi-omics data, the preprocessed multi-omics data can be divided into training set data, validation set data, and test set data. The entire model is trained using the training set data, and the parameters of the entire model are adjusted through optimization algorithms to minimize the loss function. The model performance is then evaluated on the validation set, and hyperparameter tuning is performed to prevent overfitting or underfitting.

[0077] For example, when the first data processing model is an unsupervised learning model, it can be trained separately from the second data processing model using the method described above, or it can be trained as a single model. For instance, matrix factorization and autoencoders are both unsupervised learning models; they do not require labeled data during training and can be trained unsupervised. Therefore, the first and second data processing models can be trained separately. Alternatively, labeled data can be used to adjust the trained model to improve its classification performance.

[0078] When the first data processing model is a supervised learning model, the first and second data processing models can be trained as a whole model using the method described above. For example, multilayer feedforward neural networks and graph neural networks are both supervised learning models, which require labeled data during training. Therefore, labels can be set for multi-omics data according to a preset task type (e.g., the training objective of the second data processing model), and the labeled multi-omics data can be used to train the whole model.

[0079] After model training is complete, performance evaluation can be performed. The trained secondary data processing model or the entire model can be used to perform classification predictions on test set data, and metrics such as accuracy, recall, F1 score, ROC curve, and AUC value can be used to evaluate the model's classification performance. Following performance evaluation, the trained model can be applied to disease analysis; for example, it can be used in areas such as early cancer diagnosis, prognostic assessment, and personalized treatment.

[0080] In some exemplary embodiments, the first data processing model is a matrix factorization model, wherein the factorization dimension of the matrix factorization model is the number of pathways involved in the multi-omics data.

[0081] In some exemplary embodiments, the objective function of the matrix factorization model is: agrmin||M-XY|| 2 +λ1z(X,Y)+λ2f(Y), where M=XY, M is a multi-omics data matrix with dimensions N*p. X is the first matrix obtained by decomposing the multi-omics data matrix M, and Y is the second matrix obtained by decomposing the multi-omics data matrix M. The first matrix X has dimensions N*K, and the second matrix Y has dimensions K*p, where K is the number of pathways involved in the multi-omics data. λ1 is the first hyperparameter, λ2 is the second hyperparameter, z(X,Y) is the first constraint term, f(Y) is the second constraint term, h is the total logarithm of the interaction strengths between pathways involved in the multi-omics data, and s j Let c be the known interaction strength between the j-th pair of pathways (a quantitative score of the interaction between pathways, which can be calculated from the aforementioned pathway representation vector). j Let represent the strength of the interaction relationship between the j-th pair of pathways in the second matrix Y.

[0082] This embodiment uses the first data processing model as a matrix factorization model to illustrate how to design and improve the current model. First, preprocessed multi-omics data, pathway data, and quantified inter-pathway interactions (inter-pathway interaction strength) are obtained. Then, a matrix factorization model for multi-omics data integration is designed based on pathway information.

[0083] The goal of matrix factorization is to decompose a large matrix M into two or more smaller matrices. This is typically done to reduce the dimensionality of the data, decrease storage space usage, improve computational efficiency, or better understand and interpret patterns and relationships within the data. Determining the factorization dimension K is a crucial step in matrix factorization, directly impacting the expressive power of the decomposed matrices, the complexity of the model, and its performance.

[0084] In this embodiment, the preprocessed multi-omics data is represented as a matrix M(N*p), where N represents the number of samples and p represents the number of features across all omics. To achieve the integration and dimensionality reduction of the multi-omics data, it can be decomposed into a matrix: M = XY, where X has a dimension of N*K and Y has a dimension of K*p. The goal is to solve for matrices X and Y, and the resulting matrix X is the non-redundant representation of the multi-omics data matrix M after dimensionality reduction. Figure 4 is a schematic diagram of matrix decomposition of matrix M.

[0085] This disclosure provides an embodiment that guides the dimensionality and objective function design of the matrix factorization model based on pathway information. First, the decomposition dimension K is set to the number of pathways involved in the multi-omics data. Then, each row of matrix Y(K*p) can be considered as the feature vector of each pathway. Next, the quantified inter-pathway interactions are used as constraints on matrix Y. Therefore, the objective function for matrix factorization of the multi-omics data matrix M can be expressed as: agrmin||M-XY|| 2 +λ1z(X,Y)+λ2f(Y), where, z(X,Y) is the L1 regularization term for matrices X and Y; f(Y) is a constraint term on matrix Y constructed based on the quantified inter-pathway interactions, where h represents the total logarithm of the inter-pathway interaction strengths involved, and s j Given the strength of the interaction between a known pair of pathways, c j λ1 represents the cosine similarity (strength of interaction between pathways) between the eigenvectors of the corresponding pathway pair in matrix Y. This constraint term allows prior knowledge of pathway data and interactions between pathways to be incorporated into the multi-omics data integration process. λ1 and λ2 are hyperparameters that determine the importance of the constraint term. agrmin represents the parameter value corresponding to the minimum value of the returned function.

[0086] After completing the design and improvement of the matrix factorization model, it can be used to decompose the multi-omics data M of N samples to obtain the integrated multi-omics data matrix X. Then, the multi-omics data matrix X is divided into training set, validation set, and test set. The second data processing model is trained using each dataset, thus realizing disease analysis based on multi-omics data.

[0087] In some other exemplary embodiments, the first data processing model is an autoencoder model; the autoencoder model includes an encoder and a decoder, and the number of neurons in the last layer of the encoder is the number of pathways involved in the multi-omics data.

[0088] In some exemplary embodiments, the loss function of the autoencoder model is: L = L1 + λL2, where, L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, N is the number of samples, and X... i Let X be the multi-omics data matrix of the i-th sample. i ′ represents the reconstructed multi-omics data matrix of the i-th sample, h represents the total logarithm of the inter-pathway interaction strengths involved in the multi-omics data, and s represents the multi-omics data matrix of the i-th sample. j Given the known interaction strength (quantified score of inter-pathway interaction) of the j-th pair, c jThe strength of the interaction relationship between the j-th pair of pathways learned by the autoencoder model.

[0089] This embodiment uses the first data processing model as an autoencoder model to illustrate how to design and improve the current model. First, preprocessed multi-omics data, pathway data, and quantified inter-pathway interactions are obtained. Then, an autoencoder model for multi-omics data integration is designed based on the pathway information.

[0090] An autoencoder is an unsupervised model typically consisting of an encoder and a decoder. The encoder compresses data into a low-dimensional representation, while the decoder recovers the original data from the low-dimensional representation. An autoencoder can be used to reconstruct the feature matrix M of multi-omics data, thereby achieving the integration and dimensionality reduction of multi-omics data. The latent representation X output by the bottleneck layer in the middle of the autoencoder is the integrated multi-omics data feature matrix.

[0091] This embodiment of the disclosure guides the structural design and loss function design of the autoencoder model based on pathway information. As shown in Figure 5, the number of neurons in the last layer of the autoencoder is set to the number of pathways K involved in the multi-omics data. Then, the decoder part can be regarded as the representation vector of each pathway learned by the autoencoder. Then, the quantized inter-pathway interaction relationship (inter-pathway interaction strength) is used as the constraint information in the autoencoder learning process. Thus, the loss function of the autoencoder can be expressed as:

[0092] L = L1 + λL2,

[0093] Where L is the total loss function of the newly designed autoencoder; L1 is the loss of the autoencoder reconstructing multi-omics data; N represents the number of samples; and X... i X′ represents the multi-omics data feature matrix of the current sample. i This represents the multi-omics data feature matrix of the reconstructed sample. λ is a hyperparameter, L2 is a constraint term imposed on the model based on the strength of inter-pathway interactions, h represents the total logarithm of the inter-pathway interactions involved, and s j For a known pair of pathways, the strength of their interaction relationship (quantified score of the pathway interaction relationship), c j This is the cosine similarity between the feature vectors of the pathway pair learned by the autoencoder. This constraint term allows the quantified inter-pathway interactions to be incorporated into the multi-omics data integration process.

[0094] After designing and improving the autoencoder model, unsupervised training of the model can be performed using N samples of multi-omics data M, resulting in an integrated multi-omics data feature matrix X. This integrated multi-omics data X is then divided into training, validation, and test sets. A classifier (i.e., the second data processing model) is then selected for training and testing to obtain a disease analysis model based on multi-omics data. Alternatively, the preprocessed N samples of multi-omics data M can be divided into training, validation, and test sets, and then both the autoencoder model and the classifier model can be trained simultaneously. The trained model can then be used for disease analysis based on multi-omics data.

[0095] In some other example implementations, the first data processing model is a multi-layer feedforward neural network model, which includes an input layer, a pathway layer, a hidden layer and an output layer arranged sequentially. The input layer includes multiple first nodes, which are biomolecules corresponding to the multi-omics data. The pathway layer includes multiple second nodes, which are pathways involved in the multi-omics data.

[0096] In some example implementations, the multilayer feedforward neural network model further includes a constraint layer positioned between the pathway layer and the hidden layer. The constraint layer includes multiple third nodes, which represent the strength of inter-pathway interactions involved in the multi-omics data.

[0097] This embodiment uses a multi-layer feedforward neural network model as an example to illustrate how to design and improve the current model. First, preprocessed multi-omics data, pathway data, and quantified inter-pathway interactions (inter-pathway interaction strength) are obtained. Then, a multi-layer feedforward neural network model for multi-omics data integration is designed based on the pathway information.

[0098] Existing methods for integrating multi-omics data based on multi-layer feedforward neural networks exist. These networks are fully connected and are trained on data to obtain a weighted network structure. However, due to the high dimensionality and limited sample size of multi-omics data, fully connected networks often fail to achieve good results. Therefore, a well-designed and effective neural network structure is crucial.

[0099] This disclosure embodiment guides the structural design of a multilayer feedforward neural network based on pathway information. As shown in Figure 6, the first layer is the input layer, which includes multiple first nodes. In this disclosure embodiment, the genes, proteins, and other biomolecules corresponding to the multi-omics data are used as the first nodes of the input layer of the neural network. The input of each first node is the omics data corresponding to the corresponding biomolecule, and the input matrix is ​​M(N*p). The second layer is the pathway layer, which includes multiple second nodes. The pathways involved in the multi-omics data are used as second nodes. The connection between the input layer and the pathway layer is determined by the relationship between the genes and other biomolecules and each pathway. The numerical matrix output by this layer is M'(N*K), where K is the number of pathways. The third layer is the constraint layer, which includes multiple third nodes. The strength of the interaction relationships between pathways involved in the multi-omics data is used as the third node. The connection between the second layer and this layer is determined by the interaction relationships between pathways, and the quantified interaction relationships between pathways are used as the initial weights of the edges. The output matrix of this layer is M(N*h), where h is the total number of interaction strengths between the pathways involved. The fourth layer is the hidden layer, which further extracts higher-order features from the multi-omics data. The output matrix is ​​X(N*t). The last layer is the output layer, which is used to characterize the classification of the multi-omics data.

[0100] After completing the design and improvement of the multi-layer feedforward neural network model, the N labeled samples are divided into training set, validation set and test set to complete the training of the multi-layer feedforward neural network model. When it is necessary to perform disease analysis on new test samples, the feature data of multiple omics of the test samples can be input into the trained model to realize disease analysis based on multi-omics data.

[0101] In some other exemplary embodiments, the first data processing model is a hypergraph neural network model, which includes multiple supernodes and multiple hyperedges. The multiple supernodes are biomolecules corresponding to the multi-omics data, and the multiple hyperedges are pathways involved in the multi-omics data.

[0102] In some exemplary embodiments, the loss function of the hypergraph neural network model is: L = L1 + λL2, L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, h is the total logarithm of the strength of inter-pathway interactions involved in multi-omics data, and s j Given the known interaction strength between the j-th pathway pair, c j Let be the strength of the interaction relationship between the j-th pair of pathways in the hypergraph.

[0103] This embodiment uses a hypergraph neural network model as an example to illustrate how to design and improve the current model. First, preprocessed multi-omics data, pathway data, and quantified inter-pathway interactions (inter-pathway interaction strength) are obtained. Then, a hypergraph neural network model for multi-omics data integration is designed based on the pathway information.

[0104] This embodiment of the disclosure also incorporates pathway information into the multi-omics data integration process based on a hypergraph neural network to achieve disease analysis tasks. The overall model is shown in Figure 7. First, a multi-omics data hypergraph is constructed for each sample based on known pathway data. Genes, proteins, and other biomolecules corresponding to the multi-omics data are used as supernodes, and the multi-omics data corresponding to the biomolecules are used as the initial features of the supernodes. Pathways are used as hyperedges to construct the multi-omics data hypergraph. Each sample is represented as a separate hypergraph, and the sample category is labeled.

[0105] Then, hypergraph representation learning is performed using a hypergraph convolutional neural network. Simultaneously, the quantized interactions between pathways are used as constraint information for hyperedges during the hypergraph learning process, updating the node and hyperedge representation vectors in the hypergraph. The updated hypergraph node representation vectors are then input into the classifier to complete sample classification. The loss function of the entire model can be expressed as:

[0106] L = L1 + λL2,

[0107] Where L1 is the loss function for the sample classification task, which can be the cross-entropy loss function; λ is a hyperparameter; L2 is the constraint term of the hyperedge; h is the total number of interaction strengths between the involved pathways; and s j c is the known strength of the interaction between the j-th pathway pair. j It is the cosine similarity between corresponding path pairs in the hypergraph.

[0108] After designing and improving the hypergraph convolutional neural network model, N labeled samples are divided into training, validation, and test sets to train the entire model. When disease analysis is needed on new test samples, the multi-omics feature data of the test samples can be input into the trained model to achieve disease analysis based on multi-omics data.

[0109] The above embodiments all involve concatenating multiple omics data into a multi-omics data matrix M(N*p), which is then input into a pre-designed model for multi-omics data integration and disease analysis. This disclosure embodiment can also be based on the concept of multi-view learning, treating each omics data point as a view for multi-omics data integration.

[0110] In some other exemplary embodiments, the first data processing model can be a multi-view learning model, which includes multiple view learning modules and a feature fusion module. Each view learning module constructs a view learning model using feature data of an omics and the pathway representation vectors of the pathways involved in the current omics. The feature fusion module fuses the feature vectors output by multiple view learning models. The objective function of the multi-view learning model is a weighted sum of the objective functions of multiple view learning models.

[0111] In this embodiment, the view learning model in each view learning module can use any of the models described in the foregoing embodiments. The objective function of the entire multi-view learning model is a weighted sum of the objective functions of multiple view learning models. The weights can be set empirically and adjusted during training based on the performance of the validation set. Furthermore, the models for each view do not necessarily have to be the same model; they can be models based on different principles, methods, and techniques. Since the core idea of ​​multi-view learning is to utilize data from multiple different sources or representations (i.e., multi-view data) to improve the performance of the learning task, each view can be considered an independent data representation or model, and they may contain complementary or redundant information. The goal of the multi-view learning method is to capture and fuse information from these different views to form a more comprehensive and accurate data representation.

[0112] As shown in Figure 8, taking the multi-view autoencoder model as an example, the omics data feature matrix corresponding to each omics data is M. i (N*p i For each omics dataset, a separate autoencoder model is designed. Each autoencoder model can be designed and improved using the aforementioned design method. For specific model design methods, please refer to the aforementioned embodiments, which will not be repeated here. Next, the objective functions of the designed multiple view autoencoders are superimposed and trained to obtain the intermediate outputs of each encoder. Then, the intermediate outputs of multiple autoencoders are fused (which can be done by direct concatenation, weighted averaging, principal component analysis, etc.) to obtain the integrated multi-omics dataset.

[0113] As shown in Figure 9, this embodiment of the present disclosure also provides a multi-omics data processing device, including a data acquisition module 910, an omics processing module 920, a pathway learning module 930, and a model building block 940, wherein:

[0114] The data acquisition module 910 is configured to acquire multi-omics data and pathway data;

[0115] The omics processing module 920 is configured to preprocess multi-omics data to obtain feature data of multiple omics.

[0116] The pathway learning module 930 is configured to perform graph representation learning on pathway data to obtain pathway representation vectors.

[0117] Model building block 940 is configured to construct a first data processing model based on feature data and pathway representation vectors of multiple omics, wherein the input layer of the first data processing model includes feature data of multiple omics, and the first data processing model includes at least a first structure, the first structure including pathway representation vectors of pathways involved in the multiple omics data.

[0118] In some exemplary embodiments, the pathway learning module 930 is further configured to calculate the strength of the interaction relationship between pathways based on the pathway representation vector;

[0119] When constructing the first data processing model, the model building block 940 makes the first data processing model include a constraint layer and / or constraint terms, wherein the constraint layer and / or the constraint terms include the strength of the interaction relationship between the pathways.

[0120] In some exemplary embodiments, the strength of the inter-pathway interaction relationship is the inter-pathway similarity.

[0121] In some exemplary embodiments, the first data processing model is a multi-omics data integration model or a disease analysis model. The multi-omics data integration model is used to classify and integrate the feature data of the multiple omics. The disease analysis model is used for at least one of the following: probability analysis of patients having the disease, probability analysis of patients expecting effective treatment, and probability analysis of residual lesions.

[0122] In some exemplary embodiments, the first data processing model is a matrix factorization model, wherein the decomposition dimension of the matrix factorization model is the number of pathways involved in the multi-omics data.

[0123] In some exemplary embodiments, the objective function of the matrix factorization model is: agrmin||M-XY|| 2 +λ1z(X,Y)+λ2f(Y), where M=XY, M is a multi-omics data matrix with dimensions N*p. X is a first matrix obtained by decomposing the multi-omics data matrix M. Y is a second matrix obtained by decomposing the multi-omics data matrix M. The first matrix X has dimensions N*K, and the second matrix Y has dimensions K*p, where K is the number of pathways involved in the multi-omics data. λ1 is a first hyperparameter, λ2 is a second hyperparameter, z(X,Y) is a first constraint term, f(Y) is a second constraint term, h is the total logarithm of the interaction strengths between pathways involved in the multi-omics data, and s j Given the known interaction strength between the j-th pathway pair, c jLet represent the strength of the interaction relationship between the j-th pair of pathways in the second matrix Y.

[0124] In some exemplary embodiments, the first data processing model is an autoencoder model; the autoencoder model includes an encoder and a decoder, and the number of neurons in the last layer of the encoder is the number of pathways involved in the multi-omics data.

[0125] In some exemplary embodiments, the loss function of the autoencoder model is: L = L1 + λL2, where, L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, N is the number of samples, and X... i Let X be the multi-omics data matrix of the i-th sample. i ' is the reconstructed multi-omics data matrix of the i-th sample, h is the total logarithm of the inter-pathway interaction strengths involved in the multi-omics data, and s is the multi-omics data matrix of the i-th sample. j Let c be the strength of the interaction between the j-th pathway pair. j The strength of the interaction relationship between the j-th pair of pathways learned by the autoencoder model.

[0126] In some example implementations, the first data processing model is a multi-layer feedforward neural network model, which includes an input layer, a pathway layer, a hidden layer and an output layer arranged sequentially. The input layer includes multiple first nodes, which are biomolecules corresponding to the multi-omics data. The pathway layer includes multiple second nodes, which are pathways involved in the multi-omics data.

[0127] In some example implementations, the multilayer feedforward neural network model further includes a constraint layer positioned between the pathway layer and the hidden layer. The constraint layer includes multiple third nodes, which represent the strength of inter-pathway interactions involved in the multi-omics data.

[0128] In some exemplary embodiments, the first data processing model is a hypergraph neural network model, which includes multiple supernodes and multiple hyperedges. The multiple supernodes are biomolecules corresponding to the multi-omics data, and the multiple hyperedges are pathways involved in the multi-omics data.

[0129] In some exemplary embodiments, the loss function of the hypergraph neural network model is: L = L1 + λL2, L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, h is the total logarithm of the inter-pathway interaction strengths involved in the multi-omics data, and s j Let c be the strength of the interaction between the j-th pathway pair. j Let be the strength of the interaction relationship between the j-th pair of pathways in the hypergraph.

[0130] In some exemplary embodiments, the first data processing model is a multi-view learning model, which includes multiple view learning modules and a feature fusion module. Each view learning module constructs a view learning model using feature data from an omics and the pathway representation vectors of the pathways involved in the current omics. The feature fusion module fuses the feature vectors output by multiple view learning models. The objective function of the multi-view learning model is a weighted sum of the objective functions of multiple view learning models.

[0131] In some exemplary embodiments, the characteristic data of multiple omics include at least two of the following: gene mutation, copy number variation, DNA methylation, gene expression level, protein abundance, and metabolite concentration.

[0132] This disclosure also provides a multi-omics data processing apparatus, including a memory; and a processor connected to the memory, the memory being used to store instructions, the processor being configured to perform steps of the multi-omics data processing method as described in any embodiment of this disclosure based on the instructions stored in the memory.

[0133] As shown in Figure 10, in one example, a multi-omics data processing device may include a processor 1010, a memory 1020, a bus system 1030, and a transceiver 1040. The processor 1010, memory 1020, and transceiver 1040 are connected via the bus system 1030. The memory 1020 stores instructions, and the processor 1010 executes the instructions stored in the memory 1020 to control the transceiver 1040 to transmit and receive signals. Specifically, the transceiver 1040 can acquire multi-omics data and pathway data under the control of the processor 1010. The processor 1010 preprocesses the multi-omics data to obtain feature data of multiple omics; performs graph representation learning on the pathway data to obtain pathway representation vectors; and constructs a first data processing model based on the feature data of the multiple omics and the pathway representation vectors. The input layer of the first data processing model includes the feature data of the multiple omics, and the first data processing model includes at least a first structure, which includes pathway representation vectors of the pathways involved in the multi-omics data.

[0134] It should be understood that processor 1010 can be a Central Processing Unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0135] Memory 1020 may include read-only memory and random access memory, and provides instructions and data to processor 1010. A portion of memory 1020 may also include non-volatile random access memory. For example, memory 1020 may also store device type information.

[0136] In addition to the data bus, the bus system 1030 may also include a power bus, a control bus, and a status signal bus. However, for clarity, all buses are labeled as bus system 1030 in Figure 10.

[0137] In implementation, the processing performed by the processing device can be accomplished through integrated logic circuits in the hardware of the processor 1010 or through software instructions. That is, the method steps of this embodiment can be executed by a hardware processor, or by a combination of hardware and software modules within the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other storage media. This storage medium is located in memory 1020. The processor 1010 reads information from memory 1020 and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, further details are omitted here.

[0138] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the multi-omics data processing method as described in any embodiment of this disclosure. The multi-omics data processing method driven by executing executable instructions is essentially the same as the multi-omics data processing method provided in the above embodiments of this disclosure, and will not be described in detail here.

[0139] In some possible implementations, various aspects of the multi-omics data processing method provided in this disclosure can also be implemented as a program product comprising program code that, when run on a computer device, causes the computer device to perform the steps of the multi-omics data processing method according to various exemplary embodiments of this disclosure as described above. For example, the computer device can execute the multi-omics data processing method described in the embodiments of this disclosure.

[0140] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0141] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0142] It should be noted that the above embodiments or implementation methods are merely exemplary and not restrictive. Therefore, this disclosure is not limited to the content specifically shown and described herein. Various modifications, substitutions, or omissions can be made to the form and details of the implementations without departing from the scope of this disclosure.

Claims

1. A multi-omics data processing method, comprising: Acquire multi-omics and pathway data; The multi-omics data are preprocessed to obtain feature data of multiple omics. Graph representation learning is performed on the path data to obtain path representation vectors; A first data processing model is constructed based on the feature data of the multiple omics and the pathway representation vectors. The input layer of the first data processing model includes the feature data of the multiple omics. The first data processing model includes at least a first structure, which includes pathway representation vectors of the pathways involved in the multiple omics data.

2. The method according to claim 1, wherein, The method further includes: Calculate the strength of the interaction relationship between pathways based on the pathway representation vector; When constructing the first data processing model, the first data processing model includes a constraint layer and / or constraint terms, wherein the constraint layer and / or the constraint terms include the strength of the interaction relationship between the pathways.

3. The method according to claim 2, wherein, The strength of the interaction between pathways is the similarity between pathways.

4. The method according to claim 1, wherein, The first data processing model is a multi-omics data integration model or a disease analysis model. The multi-omics data integration model is used to classify and integrate the feature data of the multiple omics. The disease analysis model is used for at least one of the following: probability analysis of patients' disease, probability analysis of patients' expected treatment effectiveness, and probability analysis of residual lesions.

5. The method according to claim 1, wherein, The first data processing model is a matrix factorization model, and the decomposition dimension of the matrix factorization model is the number of pathways involved in the multi-omics data.

6. The method according to claim 5, wherein, The objective function of the matrix factorization model is: agrmin‖M-XY‖ 2 +λ1z(X,Y)+λ2f(Y), where M=XY, M is a multi-omics data matrix with dimensions N*p. X is a first matrix obtained by decomposing the multi-omics data matrix M. Y is a second matrix obtained by decomposing the multi-omics data matrix M. The first matrix X has dimensions N*K, and the second matrix Y has dimensions K*p, where K is the number of pathways involved in the multi-omics data. λ1 is a first hyperparameter, λ2 is a second hyperparameter, z(X,Y) is a first constraint term, f(Y) is a second constraint term, h is the total logarithm of the interaction strengths between pathways involved in the multi-omics data, and s j Let c be the strength of the interaction between the j-th pathway pair. j Let represent the strength of the interaction relationship between the j-th pair of pathways in the second matrix Y.

7. The method according to claim 1, wherein, The first data processing model is an autoencoder model; The autoencoder model includes an encoder and a decoder, and the number of neurons in the last layer of the encoder is equal to the number of pathways involved in the multi-omics data.

8. The method according to claim 7, wherein, The loss function of the autoencoder model is: L = L1 + λL2, where, L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, N is the number of samples, and X... i Let X′ be the multi-omics data matrix of the i-th sample. i Let h be the total logarithm of the inter-pathway interaction strengths involved in the reconstructed i-th sample multi-omics data matrix, and s be the multi-omics data matrix. j Let c be the strength of the interaction between the j-th pathway pair. j The strength of the interaction relationship between the j-th pair of pathways learned by the autoencoder model.

9. The method according to claim 1, wherein, The first data processing model is a multi-layer feedforward neural network model, which includes an input layer, a pathway layer, a hidden layer and an output layer arranged sequentially. The input layer includes multiple first nodes, which are biomolecules corresponding to the multi-omics data. The pathway layer includes multiple second nodes, which are pathways involved in the multi-omics data.

10. The method according to claim 9, wherein, The multilayer feedforward neural network model further includes a constraint layer, which is set between the pathway layer and the hidden layer. The constraint layer includes multiple third nodes, which represent the strength of the inter-pathway interaction relationships involved in the multi-omics data.

11. The method according to claim 1, wherein, The first data processing model is a hypergraph neural network model, which includes multiple hypernodes and multiple hyperedges. The multiple hypernodes are the biomolecules corresponding to the multi-omics data, and the multiple hyperedges are the pathways involved in the multi-omics data.

12. The method according to claim 11, wherein, The loss function of the hypergraph neural network model is: L = L1 + λL2. L is the total loss function, L1 is the first loss function, L2 is the second loss function, λ is the hyperparameter, h is the total logarithm of the inter-pathway interaction strengths involved in the multi-omics data, and s j Let c be the strength of the interaction between the j-th pathway pair. j Let be the strength of the interaction relationship between the j-th pair of pathways in the hypergraph.

13. The method according to claim 1, wherein, The first data processing model is a multi-view learning model, which includes multiple view learning modules and a feature fusion module. Each view learning module uses feature data from an omics and the pathway representation vectors of the pathways involved in the current omics to construct a view learning model. The feature fusion module fuses the feature vectors output by multiple view learning models. The objective function of the multi-view learning model is a weighted sum of the objective functions of multiple view learning models.

14. The method according to claim 1, wherein, The feature data of the multiple omics include at least two of the following: gene mutation, copy number variation, DNA methylation, gene expression level, protein abundance, and metabolite concentration.

15. A multi-omics data processing apparatus, comprising a memory; and a processor connected to the memory, the memory for storing instructions, the processor being configured to perform the steps of the multi-omics data processing method as claimed in any one of claims 1 to 14 based on the instructions stored in the memory.

16. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-omics data processing method as described in any one of claims 1 to 14.

17. A computer program product comprising instructions that, when executed by a computer, perform the multi-omics data processing method as described in any one of claims 1 to 14.

18. A multi-omics data processing device, comprising a data acquisition module, an omics processing module, a pathway learning module, and a model building module, wherein: The data acquisition module is configured to acquire multi-omics data and pathway data; The omics processing module is configured to preprocess the acquired multi-omics data to obtain feature data of multiple omics. The path learning module is configured to perform graph representation learning on the acquired path data to obtain path representation vectors; The model building module is configured to build a first data processing model based on the feature data of the multiple omics and the pathway representation vectors, wherein the input layer of the first data processing model includes the feature data of the multiple omics, and the first data processing model includes at least a first structure, the first structure including pathway representation vectors of the pathways involved in the multiple omics data.

Images