Multi-omics causal structure relation learning method based on comparative learning

Abstract

The invention discloses a multi-omics causal structure relation learning method based on comparative learning, which comprises the following steps: firstly, respectively constructing corresponding encoders for preprocessed gene mutation and gene expression data, and respectively carrying out feature extraction on two kinds of omics data; then, constructing a projection head with shared parameters to realize cross-modal feature alignment; then, using the aligned features as nodes, and constructing causal graph data through a learnable causal graph structure; constructing a graph neural network to learn causal graph representation, and constructing a contrast loss function; and finally, a model prediction result is obtained through a multi-layer perceptron, a survival prediction loss function is constructed, and a total loss function is obtained for multi-omics causal structure model training. Based on gene mutation and gene expression data, a cross-omics causal structure relationship is constructed and learned through comparative learning, more accurate prognosis prediction is provided for diseases such as acute myelogenous leukemia and the like, and potential biomarkers and key regulatory factors are helped to be found.

Description

Technical Field

[0001] The present invention relates to the learning of causal structure relationships, and in particular to a multi-omics causal structure relationship learning method based on contrastive learning. Background Art

[0002] Acute myeloid leukemia (AML) is a malignant blood disease originating from the bone marrow, with complex clinical manifestations and poor prognosis. Traditional prognosis prediction methods often rely on limited clinical indicators and single gene markers. However, due to the complex pathogenesis of AML, involving abnormalities in multiple genes and pathways, the analysis of single-omics data often fails to comprehensively reveal the occurrence and development of the disease. Therefore, multi-omics analysis, especially the combination of gene mutations and gene expression, has become an important direction in AML prognosis research in recent years. Gene mutations are closely related to the occurrence and prognosis of AML. Different gene mutations not only affect the proliferation and differentiation of leukemia cells, but also are closely related to treatment response and patient survival. For example, mutations such as FLT3, NPM1, IDH1 / 2, etc. have been widely recognized as prognostic markers for AML. Compared with gene mutations, changes in gene expression levels are often closely related to functions such as cell proliferation, differentiation, and immune escape, and thus have important significance in the diagnosis and prognosis assessment of AML. Multi-omics analysis can understand the complexity of AML from different levels by integrating gene mutation and gene expression data.

[0003] The multi-omics causal structure relationship, especially the causal structure analysis of gene mutations and gene expression on the prognosis prediction of acute myeloid leukemia (AML), has important clinical significance. Causal structure relationship analysis can reveal the interactions and causal mechanisms between different genes and their expression products through multi-level and multi-dimensional data fusion. By constructing a multi-omics causal structure model, the potential causal relationship between gene mutations and gene expression can be more accurately captured, thereby revealing the deep mechanism of AML and providing new ideas for clinical prognosis prediction. Causal structure analysis not only helps to improve the accuracy of prognosis prediction, but also provides a scientific basis for the early diagnosis, risk assessment, and treatment plan optimization of AML.

[0004] Traditional methods for causal structure relationships usually require explicitly defining the causal structure between variables, relying on expert knowledge or assumed models. For example, Structural Equation Modeling (SEM) requires setting causal paths in advance. If the settings are incorrect, it will lead to result biases. In addition, a large amount of data is needed to estimate causal relationships. In some fields (such as medicine and bioinformatics), high-quality data is difficult to obtain, which may lead to overfitting or unstable estimation. Many traditional causal methods default that the relationships between variables are linear (such as the linear regression assumption in SEM). However, in reality, causal relationships are usually non-linear and even involve complex interactions. However, most existing studies are based on single omics or modalities. In omics data fusion (such as genomics, transcriptomics, proteomics), how to establish cross-modal causal structures remains a challenge. Therefore, how to obtain a good representation of different omics data and how to learn cross-modal causal structure relationships not only helps improve the accuracy of disease-related task prediction but also enables in-depth understanding of disease mechanisms. Summary of the Invention

[0005] An object of the present invention is to provide a method for learning multi-omics causal structure relationships based on contrastive learning in view of the deficiencies of the prior art.

[0006] The object of the present invention is achieved through the following technical solutions: A method for learning multi-omics causal structure relationships based on contrastive learning, including:

[0007] First, construct corresponding encoders for preprocessed gene mutation and gene expression data respectively, and extract features from the two types of omics data respectively;

[0008] Then, construct a parameter-sharing projection head to achieve cross-modal feature alignment;

[0009] Subsequently, use the aligned features as nodes to construct causal graph data through a learnable causal graph structure;

[0010] Furthermore, construct a graph neural network to learn the causal graph representation and construct a contrastive loss function;

[0011] Finally, obtain the model prediction result through a multi-layer perceptron, and construct a survival prediction loss function to obtain the total loss function for multi-omics causal structure model training; the multi-omics causal structure model includes the encoder, the projection head, the graph neural network, and the multi-layer perceptron.

[0012] Further, the constructing corresponding encoders for preprocessed gene mutation and gene expression data respectively, and extracting features from the two types of omics data respectively, includes:

[0013] For omics data of gene mutations, the encoder includes a mutation embedding layer and an inter-gene interaction layer. The mutation embedding layer makes each locus independent and maps discrete mutation types into vectors, with each locus having its own set of embedding vectors. The inter-gene interaction layer uses a Transformer encoder to capture the interactions between loci and enhance the feature representation.

[0014] For omics data of gene expression, the encoder includes a gene independent embedding layer and an inter-gene interaction layer. Among them, the gene independent embedding layer includes a linear layer and an activation function, which converts the scalar of each gene into high-dimensional features to ensure that the position of the gene features is consistent with the input. The inter-gene interaction layer uses a Transformer-based encoder and the self-attention mechanism to allow the gene features to influence each other, enabling each gene feature to fuse global information.

[0015] Furthermore, the construction of a parameter-sharing projection head to achieve cross-modal feature alignment includes:

[0016] The projection head includes a shared layer and a private layer. The shared layer processes common features, and the private layer processes modality-specific features. The shared layer shares some parameters to facilitate knowledge transfer between modalities. The private layer does not share parameters and retains modality-specific information. Both are based on a multi-layer perceptron and configured with layer normalization to stabilize the training process.

[0017] Then, a cosine loss function is applied to the features passing through the projection head for calculation, directly maximizing the cosine of the angle between feature vectors to enhance feature alignment through feature similarity:

[0018]

[0019] where and are the gene mutation and gene expression features after being processed by the projection head, respectively.

[0020] Furthermore, taking the aligned features as nodes, a causal graph data is constructed through a learnable causal graph structure, including:

[0021] The learnable causal graph structure is characterized by a learnable adjacency matrix A, and a directed acyclic graph constraint is introduced ; the adjacency matrix is a binary matrix, and A[i, j]=1 indicates that there is a causal edge from mutant gene i to expression gene j;

[0022]

[0023] where ∗ represents the Hadamard product of two matrices, and n is the number of features;

[0024] The causal graph data is constructed for each sample, with the nodes being the aligned features, and the node features being the aligned features.

[0025] Furthermore, the graph neural network is constructed to learn the causal graph representation, and a contrastive loss function is constructed, including:

[0026] The graph neural network adopts a graph convolutional neural network, and the update of each node is based on the features of its neighbor nodes, through a weight matrix and normalization processing; the adjacency matrix only contains the connections from mutant nodes to expression nodes. During the message passing process, only the features of mutant nodes are aggregated to the expression nodes, that is, the features of mutant nodes remain unchanged between layers, and the features of expression nodes are updated based on the features of mutant nodes and the adjacency weights;

[0027] Construct the contrastive loss function is constructed based on the causal graph representation z learned by the graph neural network:

[0028]

[0029] where is the set of positive samples, is the set of negative samples, is the temperature coefficient, which controls the sharpness of the distribution.

[0030] Furthermore, the model prediction result is obtained through a multi-layer perceptron, and a survival prediction loss function is constructed to obtain the total loss function for training the multi-omics causal structure model, specifically:

[0031] The multi-layer perceptron includes three linear layers, two activation functions, and layer normalization, and the model prediction result, that is, the risk score, is obtained through the last linear layer;

[0032] The constructed survival prediction loss function is based on the risk score predicted by the model and the corresponding survival status and survival time of the sample, that is, the Cox loss function:

[0033] For each sample i, its survival time is , the censoring indicator is , the predicted risk score is , denotes all samples still in the risk set after time , is the predicted risk score corresponding to sample j, N is the total number of all samples, and the Cox loss function is:

[0034]

[0035] Furthermore, during the model training process, the total loss function is:

[0036]

[0037] wherein is the cosine loss function, is the directed acyclic graph constraint, , and are learnable parameters respectively.

[0038] The present invention also provides an electronic device, including a memory and a processor, the memory is coupled to the processor; wherein, the memory is used for storing program data, and the processor is used for executing the program data to implement the above-mentioned method for learning multi-omics causal structure relationship based on contrast learning.

[0039] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned method for learning multi-omics causal structure relationship based on contrast learning is implemented.

[0040] The present invention also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the above-mentioned method for learning multi-omics causal structure relationship based on contrast learning is implemented.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can effectively integrate multi-omics data such as gene mutations and gene expressions in the causal structure learning task, achieve cross-modal feature alignment through a parameter-sharing projection head, and improve data consistency. Based on the learnable causal graph structure, the causal relationships between different omics features are mined, and combined with the graph neural network to enhance causal representation learning, the interaction between multi-omics data can be deeply understood, the explanatory ability of disease mechanisms can be improved, and thus strong support can be provided for the formulation of personalized treatment plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0043] Figure 1 is a schematic flowchart of a method for learning multi-omics causal structure relationship based on contrast learning provided by an embodiment of the present invention;

[0044] Figure 2 is a schematic structural diagram of a gene mutation encoder provided by an embodiment of the present invention;

[0045] Figure 3 Schematic diagram of the gene expression encoder structure provided by the embodiment of the present invention;

[0046] Figure 4 Schematic diagram of the projection head structure provided by the embodiment of the present invention;

[0047] Figure 5 Schematic diagram of the multi-layer perceptron structure provided by the embodiment of the present invention;

[0048] Figure 6 Schematic diagram of an electronic device provided by the embodiment of the present invention. Specific implementation manners

[0049] The present invention will be described in detail below with reference to the accompanying drawings. Without conflict, the features in the following embodiments and implementation manners can be combined with each other.

[0050] A multi-omics causal structure relationship learning method based on contrastive learning of the present invention, as Figure 1 shown, includes the following steps:

[0051] (1) First, construct corresponding encoders for the preprocessed gene mutations and gene expression data respectively, and extract features from the two types of omics data respectively;

[0052] Specifically, preprocess the gene mutation and gene expression data respectively: for the gene mutation data, first exclude genes with a mutation frequency lower than 1%, and finally select genes with a higher mutation frequency as the input of the gene mutation data, where the mutation is recorded as 1 and the non-mutated gene is recorded as 0; for the gene expression data, the gene expression data is obtained by RNA-seq and standardized using reads per kilobase per million mapped reads (HTseq-FPKM). Further processing includes converting the FPKM value to TPM (transcripts per million), adding 1 to the TPM value, and then applying a logarithmic transformation to stabilize the variance and improve the reliability of statistical analysis. Next, calculate the variance of gene expression, select the top 100 genes with the largest variance, and standardize their expression levels.

[0053] Subsequently, as Figure 2 shown, for the omics data of gene mutations, an encoder is constructed, which mainly includes a mutation embedding layer and an inter-gene interaction layer. The mutation embedding layer makes each site independent and maps discrete mutation types (such as 0 / 1) into higher-dimensional vectors, and each site has its own set of embedding vectors; the inter-gene interaction layer uses a Transformer encoder to capture the interactions between sites and enhance the feature representation.

[0054] As Figure 3As shown in the figure, for the omics data of gene expression, an encoder is constructed. The encoder mainly includes a gene independent embedding layer and an inter-gene interaction layer. Among them, the gene independent embedding layer includes a linear layer and an activation function, which can convert the scalar of each gene into a high-dimensional feature to ensure that the position of the gene feature is consistent with the input; the inter-gene interaction layer uses an encoder based on Transformer and the self-attention mechanism to allow each gene feature to interact with each other, so that each gene feature fuses global information.

[0055] (2) Then, a projection head with shared parameters is constructed to achieve cross-modal feature alignment;

[0056] Specifically, as Figure 4 shown in the figure, the projection head includes a shared layer and a private layer. The shared layer processes common features, and the private layer processes modality-specific features. The shared layer includes a linear layer, an activation function (such as Leaky ReLU), layer normalization, a linear layer, an activation function (such as Leaky ReLU), and a linear layer, which are connected in sequence. The private layer includes a linear layer, an activation function (such as Leaky ReLU), layer normalization, and a linear layer, which are connected in sequence; the shared layer shares some parameters to promote knowledge transfer between modalities; the private layer does not share parameters and retains modality-specific information. Both are based on a multi-layer perceptron and configured with layer normalization, which can help stabilize the training process, especially when the data distributions of different modalities are quite different.

[0057] Then, the features passing through the projection head are calculated using the cosine loss function to directly maximize the cosine of the angle between the feature vectors, so as to enhance feature alignment through feature similarity:

[0058]

[0059] where and are the gene mutation and gene expression features respectively after being processed by the projection head.

[0060] (3) Subsequently, the aligned features are used as nodes, and causal graph data is constructed through a learnable causal graph structure;

[0061] Specifically, the learnable causal graph structure is represented as a learnable adjacency matrix A, and a directed acyclic graph constraint is introduced. The adjacency matrix is a binary matrix, and A[i,j]=1 indicates that there is a causal edge from the mutant gene i to the expression gene j. The causal graph data is constructed for each sample, the nodes are the aligned features, and the node features are the aligned features.

[0062]

[0063] where ∗ denotes the Hadamard product of two matrices, and n is the number of features. is the trace function.

[0064] (4) Furthermore, construct a graph neural network to learn the causal graph representation and construct a contrastive loss function.

[0065] The graph neural network uses a graph convolutional neural network, and the update of each node is based on the features of its neighbor nodes through a weight matrix and normalization. The adjacency matrix only contains the connections from mutant nodes to expression nodes. During the message passing process, only the features of mutant nodes are aggregated to the expression nodes, that is, the features of mutant nodes remain unchanged between layers, and the features of expression nodes are updated based on the features of mutant nodes and adjacency weights.

[0066] The construction of the contrastive loss function is based on the causal graph representation z learned by the graph neural network:

[0067]

[0068] where is the set of positive samples, is the set of negative samples, and are the causal graph representations of sample i and sample j respectively, is the temperature coefficient, which controls the sharpness of the distribution.

[0069] It should be noted that samples with similar survival times are regarded as positive pairs (similar prognoses), and samples with large time differences are regarded as negative pairs (different prognoses). At the same time, consider the event status. That is, for each sample i, define the time window where can be set to 3 months. Therefore, positive samples are those whose survival times fall within the same window as sample i, regardless of the survival status; negative samples are those whose time differences exceed the window, such as or . However, if sample j is censored, that is, alive and not dead, it is only regarded as a negative sample when considering that its actual survival time may be longer.

[0070] (5) Finally, obtain the model prediction results through a multi-layer perceptron and construct a survival prediction loss function to obtain the total loss function for training the multi-omics causal structure model.

[0071] Specifically, as Figure 5As shown in the figure, the multi-layer perceptron includes linear layers, activation functions (such as Leaky ReLU), layer normalization, linear layers, activation functions (such as Leaky ReLU), layer normalization, and linear layers, which are connected in sequence, and the model prediction result, that is, the death risk score of the patient (the death risk score of an individual, which can predict the risk of death events in the future), is obtained through the last linear layer. The constructed survival prediction loss function is based on the risk score predicted by the model and the corresponding survival status and survival time of the samples, that is, the CoxLoss loss function:

[0072] For each sample i, its survival time is , the censoring indicator is (1 indicates that the event occurs, i.e., death, and 0 indicates right censoring, i.e., survival), and the predicted risk score is , represents all samples still in the risk set after time , is the predicted risk score corresponding to sample j, N is the total number of all samples, and the Cox loss function is:[[]]

[0073]

[0074] During the model training process, the total loss function is:

[0075]

[0076] Among them, , and are learnable parameters respectively.

[0077] The present invention also provides a multi-omics causal structure relationship learning device based on contrast learning, including:

[0078] A feature extraction module, which is used to construct corresponding encoders for the preprocessed gene mutation and gene expression data respectively, and extract features from the two types of omics data respectively;

[0079] A feature alignment module, which is used to construct a projection head with shared parameters to achieve cross-modal feature alignment;

[0080] A causal graph construction module, which is used to use the aligned features as nodes and construct causal graph data through a learnable causal graph structure;

[0081] A network learning module, which is used to construct a graph neural network to learn the causal graph representation and construct a contrast loss function;

[0082] A model training module, which is used to obtain a model prediction result through a multi-layer perceptron, construct a survival prediction loss function, and obtain a total loss function for training a multi-omics causal structure model; the multi-omics causal structure model includes the encoder, the projection head, the graph neural network, and the multi-layer perceptron.

[0083] It should be noted that the system embodiment shown in this embodiment matches the content of the above method embodiment. For the content of the above method embodiment, reference can be made and will not be elaborated here.

[0084] Figure 6 This is a schematic structural diagram of an electronic device provided by an embodiment of the present invention. Please refer to Figure 6 , the electronic device provided in this embodiment includes: a memory and a processor. Among them, the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, the above-mentioned method for learning the multi-omics causal structure relationship based on contrast learning is implemented.

[0085] It should be noted that in addition to Figure 6 the shown memory and processor, according to its actual functions, the electronic device may further include other hardware, which will not be elaborated here.

[0086] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the above-mentioned method for learning the multi-omics causal structure relationship based on contrast learning is implemented.

[0087] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, the above-mentioned method for learning the multi-omics causal structure relationship based on contrast learning is implemented.

[0088] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0089] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing device produce means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0090] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce a manufacture including instruction means for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0091] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more of the flows Figure 1 one or more of the flows and / or blocks Figure 1 or means for implementing the functions specified in one or more of the blocks.

[0092] The above embodiments are only used to illustrate the design concept and features of the present invention, and the purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made according to the principles and design concepts disclosed by the present invention are within the protection scope of the present invention.

Claims

1. A method for learning the causal structure relationship of multi-omics based on contrastive learning, characterized in that, Including: First, corresponding encoders are constructed for the preprocessed gene mutation and gene expression data respectively, and feature extraction is performed on the two types of omics data respectively; Then, a projection head with shared parameters is constructed to achieve cross-modal feature alignment; Subsequently, the aligned features are used as nodes, and causal graph data is constructed through a learnable causal graph structure; Furthermore, a graph neural network is constructed to learn the causal graph representation, and a contrast loss function is constructed; Finally, the model prediction result is obtained through a multi-layer perceptron, and a survival prediction loss function is constructed to obtain the total loss function for training the multi-omics causal structure model; the multi-omics causal structure model includes the encoder, the projection head, the graph neural network, and the multi-layer perceptron.

2. The multi-omics causal structure relationship learning method based on contrastive learning according to claim 1, characterized in that, The step of constructing corresponding encoders for the preprocessed gene mutation and gene expression data respectively and performing feature extraction on the two types of omics data respectively includes: For the omics data of gene mutation, the constructed encoder includes a mutation embedding layer and an inter-gene interaction layer. The mutation embedding layer makes each locus independent and maps the discrete mutation types into vectors, and each locus has its own set of embedding vectors; the inter-gene interaction layer uses a Transformer encoder to capture the interactions between loci and enhance the feature representation; For the omics data of gene expression, the constructed encoder includes a gene independent embedding layer and an inter-gene interaction layer. Among them, the gene independent embedding layer includes a linear layer and an activation function, which converts the scalar of each gene into a high-dimensional feature to ensure that the position of the gene feature is consistent with the input; the inter-gene interaction layer adopts an encoder based on Transformer, and uses the self-attention mechanism to make the gene features affect each other, so that the gene features fuse global information.

3. A method for learning the causal structure relationship of multi-omics based on contrastive learning according to claim 1, characterized in that, The step of constructing a projection head with shared parameters to achieve cross-modal feature alignment includes: The projection head includes a shared layer and a private layer. The shared layer processes the common features, and the private layer processes the modality-specific features; The shared layer shares some parameters to promote knowledge transfer between modalities; the private layer does not share parameters and retains modality-specific information; both are based on a multi-layer perceptron and are configured with layer normalization to stabilize the training process; Then, perform cosine loss function calculation on the features passing through the projection head, directly maximizing the cosine of the angle between feature vectors to enhance features through feature similarity ​ Alignment: Among them, and are the gene mutation and gene expression characteristics after being processed by the projection head, respectively.

4. A method for learning the causal structure relationship of multi-omics based on contrastive learning according to claim 1, characterized in that, The step of using the aligned features as nodes and constructing causal graph data through a learnable causal graph structure includes: The learnable causal graph structure is represented as a learnable adjacency matrix A, and a directed acyclic graph constraint is introduced. ; The adjacency matrix is a binary matrix, where A[i, j]=1 indicates that there is a causal edge from mutant gene i to expression gene j. Where, ∗ represents the Hadamard product of two matrices, and n is the number of features; The causal graph data is constructed for each sample. The nodes are the aligned features, and the node features are the aligned features.

5. A method for learning multi-omics causal structure relationships based on contrastive learning according to claim 1, characterized in that, The step of constructing a graph neural network to learn the causal graph representation and constructing a contrast loss function includes: The graph neural network adopts a graph convolutional neural network. The update of each node is based on the features of its neighbor nodes through a weight matrix and normalization processing; the adjacency matrix only contains the connections from mutation nodes to expression nodes. During the message passing process, only the features of mutation nodes are aggregated to expression nodes, that is, the features of mutation nodes remain unchanged between layers, and the features of expression nodes are updated based on the features of mutation nodes and the adjacency weights; Construct the contrastive loss function It is constructed based on the causal graph representation z learned by the graph neural network: Among them, is the positive sample set, is the negative sample set, is the temperature coefficient, which controls the sharpness of the distribution.

6. The multi-omics causal structure relationship learning method based on contrastive learning according to claim 1, wherein, The model prediction result is obtained through a multi-layer perceptron, and a survival prediction loss function is constructed to obtain a total loss function for training the multi-omics causal structure model. Specifically: The multi-layer perceptron includes three linear layers, two activation functions, and layer normalization, and the model prediction result, i.e., the risk score, is obtained through the last linear layer. Constructed survival prediction loss function It is based on the risk score predicted by the model and the corresponding survival status and survival time of the sample, that is, the Cox loss function: For each sample i, its survival time is , the censoring indicator is , the predicted risk score is , denotes all samples still in the risk set after time , is the predicted risk score corresponding to sample j, N is the number of all samples, and the Cox loss function is:

7. A method for learning the causal structure relationship of multi-omics based on contrastive learning according to claim 6, characterized in that, During the model training process, the total loss function is as follows: Among them is the cosine loss function, is the directed acyclic graph constraint, , and are learnable parameters respectively.

8. An electronic device, comprising a memory and a processor, characterized in that, The memory is coupled to the processor; wherein, the memory is used to store program data, and the processor is used to execute the program data to implement a method for learning multi-omics causal structure relationships based on contrastive learning according to any one of claims 1-7 above.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a method for learning multi-omics causal structure relationships based on contrastive learning according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for learning multi-omics causal structure relationships based on contrastive learning according to any one of claims 1-7.

Images