Cell type identification method based on multi-omics decoupling representation and graph embedding and identification system thereof

Through multi-omic decoupling representation and graph embedding methods, the sample similarity information is integrated, and the problem of failure to fully utilize similarity information in cell type recognition in the prior art is solved, and accurate identification and robustness of large data sets are achieved. Sex improvement.

CN119027941AActive Publication Date: 2024-11-26CENT SOUTH UNIV

Patent Information

Application Number
CN202410906499.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-08
Publication Date
2024-11-26
Estimated Expiration
2044-07-08

AI Technical Summary

Technical Problem

The prior art fails to fully utilize sample similarity information in cell type identification, resulting in difficulty in correct classification of cells when cell type boundaries are not obvious or cell states are continuously changed, and it is impossible to effectively capture nonlinear biological processes and expand to Large data sets.

Method used

Using a method based on multi-omic decoupling representation and graph embedding, cell spectral embedding representation is obtained through graph embedding model, sample similarity graph is constructed, and different omic data are mapped to the same dimension space using a variational autoencoder model, combined with Gaussian mixing The models are clustered and similarity information is integrated by maximum likelihood estimation and stochastic gradient descent methods.

Benefits of technology

It improves the accuracy and robustness of cell type recognition, can process large multiomics data sets, enhances the model's adaptability to complex data, and achieves more accurate cell type recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119027941A_ABST
    Figure CN119027941A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of cell type identification, and discloses a cell type identification method based on multi-omics decoupling representation and graph embedding, comprising the following steps: step 1, acquiring single-cell multi-omics data and preprocessing; 2, obtaining a cell spectrum embedding expression by using the graph embedding model, calculating the similarity of the single cell samples based on the spectrum embedding expression, and further constructing a sample similarity graph; 3, mapping different omics data to the same dimension space for data integration by using the variational auto-encoder model to obtain single cell sharing potential representation; 4, clustering the cells by using a Gaussian mixture model based on the shared potential representation to obtain clustering distribution of the cells; and 5, constructing a target function based on maximum likelihood estimation. Through the beneficial effects of extensible model architecture, high-quality cell type identification and the like, the performance efficiency of single-cell multi-omics data integration analysis is improved, and a new direction and possibility are provided for single-cell multi-omics data integration.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of cell type identification, and in particular to a cell type identification method and an identification system thereof based on multi-omics decoupling representation and graph embedding. Background Art

[0002] Single-cell multi-omics sequencing technology is a revolutionary biotechnology that allows scientists to simultaneously analyze multiple omics information, including transcriptome, epigenome, and proteome, at the single cell level. The development of this technology has greatly pushed the boundaries of biological research, enabling researchers to more accurately understand cell functions, life development processes, and the mechanisms of disease occurrence. In traditional omics research, it is difficult to obtain in-depth knowledge about individual cell differences and complex biological processes because only mixed cell populations can be analyzed. The emergence of single-cell multi-omics technology effectively solves this problem and provides more comprehensive and detailed biological data.

[0003] Although single-cell multi-omics technology has great potential, its data analysis still faces some challenges. At present, a variety of cell type identification methods based on multi-omics data have been developed, which can be mainly divided into three categories: matrix decomposition-based methods, graph structure-based methods, and deep learning-based methods. These methods have their own advantages. For example, matrix decomposition-based methods can extract the main components of omics data, graph structure-based methods can use the similarities between cells to build networks, and deep learning-based methods can learn predictive models from complex data.

[0004] However, these existing technologies still have some limitations in practical applications. For example, although deep learning-based integration methods are able to process large-scale data sets and learn complex patterns from them, most of these methods only focus on learning sample-independent potential representations and fail to fully utilize the similarity information between samples. This approach of ignoring sample similarity may lead to insufficient accuracy in cell type identification, especially when cell type boundaries are not obvious or cell states change continuously, the correct classification of cells becomes particularly difficult. Therefore, how to effectively integrate similarity information between samples in deep learning models to improve the accuracy and robustness of cell type identification has become an urgent problem that needs to be solved. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a cell type identification method and an identification system based on multi-omics decoupled representation and graph embedding, which solves the problems in the prior art that it cannot capture nonlinear biological processes, cannot be expanded to large data sets, and does not fully consider sample similarity information.

[0006] To achieve the above objectives, the present invention is implemented by the following technical solutions: A cell type identification method based on multi-omics decoupling representation and graph embedding comprises the following steps:

[0007] Step 1: Obtain single-cell multi-omics data and perform preprocessing;

[0008] Step 2: Use the graph embedding model to obtain the cell spectrum embedding representation, calculate the similarity of single cell samples based on the spectrum embedding representation, and then construct a sample similarity graph;

[0009] Step 3: Use the variational autoencoder model to map different omics data into the same dimensional space for data integration to obtain the single-cell shared potential representation;

[0010] Step 4: Cluster the cells using a Gaussian mixture model based on the shared potential representation to obtain clustering assignments of the cells;

[0011] Step 5: Construct the objective function based on maximum likelihood estimation and incorporate similarity information by minimizing the JS divergence of the latent representation and cluster assignment of similar cells;

[0012] Step 6: Derived and simplified the objective function to obtain the final loss function, and used the stochastic gradient descent method to iteratively update the variational autoencoder model and Gaussian mixture model;

[0013] Step 7: Use the trained model to obtain the shared potential representation and cluster assignment of single cells.

[0014] Preferably, further, the training process of the graph embedding model includes the following steps:

[0015] Splice the preprocessed single-cell multi-omics data;

[0016] Use the autoencoder model to reduce the dimension of the concatenated data;

[0017] Use the twin neural network model to learn the sample similarity matrix of the reduced-dimensional data;

[0018] Based on spectral clustering, a neural network mapping function is learned from the dimensionality reduction space to the spectral embedding space, and the sample similarity is calculated according to the spectral embedding representation to obtain the sample similarity graph.

[0019] Preferably, further, the objective function of the graph embedding model is:

[0020]

[0021] where Y = [y1,y2,…,y N ], N is the number of data samples, x i and x jis the encoded representation after dimensionality reduction by the encoder, w(x i ,x j ) represents the sample x in the similarity matrix learned by the twin neural network i and x j The similarity of i and j The samples x are i and x j The spectral embedding of is a low-dimensional representation.

[0022] Preferably, in step 5, the objective function of the maximum likelihood estimation is:

[0023]

[0024] Among them, φ and θ represent the set of encoder parameters and the set of decoder parameters in each omics variational autoencoder model, respectively, {x (v)} is a universal multi-omics variable, {x (v)}:={x (1) ,…,x (v) ,…,x (V)}, where x (v) is the general variable of the v-th omics.

[0025] Preferably, similarity information constraints are added to the objective function, and the objective function becomes:

[0026]

[0027] Among them, w ij (∑ j w ij =1) represents the sample similarity graph learned by the graph embedding model and The similarity of represents the shared potential representation of the cell, M is the feature dimension of the latent variable z, c∈{0,1} K×1 represents the clustering assignment of cells, K is the number of cluster categories, and JS(·,·) represents the JS divergence between the two distributions.

[0028] Preferably, further, the objective function is simplified and estimated to obtain the final loss function, and the variational autoencoder model and the Gaussian mixture model are iteratively updated using the stochastic gradient descent method, wherein the simplified estimation of the objective function specifically includes:

[0029] Introduce the variational upper bound of JS divergence to replace JS divergence;

[0030] Will Decompose into and Two items related;

[0031] The objective function is estimated and solved to obtain the final loss function.

[0032] A cell type identification system based on multi-omics decoupled representation and graph embedding, including:

[0033] Similarity graph building module, used to identify similar samples in the data, capture the interactions and similarities between cells, and provide accurate graph structure representation for subsequent analysis;

[0034] The decoupled representation learning module is used to learn the potential representation of each omics data in the same dimensional space;

[0035] The fusion module is used to construct the omics weight parameters and obtain the cell-shared potential representation by weighted fusion of the potential representations learned from each omics data;

[0036] Clustering module, used to perform cluster assignment on multi-omics data, and realize cluster assignment of cell types based on shared latent representation and Gaussian mixture model;

[0037] The pre-training module is used to pre-train the variational autoencoder model to ensure that the variational autoencoder can reconstruct multi-omics data, and initialize the Gaussian mixture model to ensure that the Gaussian mixture model can capture the category information of the data;

[0038] The training module further optimizes the variational autoencoder and Gaussian mixture model based on the graph embedding model to integrate and utilize similarity information.

[0039] The recognition module is used to identify cell types based on the trained variational autoencoder model and Gaussian mixture model.

[0040] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0041] A computer-readable storage medium having a computer program stored thereon.

[0042] The present invention provides a cell type identification method and an identification system based on multi-omics decoupling representation and graph embedding, which has the following beneficial effects:

[0043] 1. The present invention uses a combination of pre-training and training to reasonably integrate similarity information into decoupled representation learning while ensuring the quality of multi-omics data reconstruction. This phased training strategy can gradually optimize each component and provide a powerful multi-omics data clustering solution by integrating these components. This method not only improves the clustering performance of the model, but also enhances the model's adaptability to the complexity of multi-omics data.

[0044] 2. The present invention performs well on single-cell multi-omics data sets of multiple omics types and multiple omics quantities, achieving relatively accurate cell type identification, which means that the present invention has the ability to integrate most single-cell multi-omics data. This ability is of great significance for processing large and mixed single-cell multi-omics data. It can not only realize multi-omics cell type identification of different omics types and quantities, but also provide new directions and possibilities for the integration of single-cell multi-omics data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the method of the present invention;

[0046] Figure 2 The schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Embodiment 1:

[0049] Please see attached Figure 1 -Attached Figure 2 , an embodiment of the present invention provides a cell type identification method based on multi-omics decoupled representation and graph embedding, comprising the following steps:

[0050] Step 1: Obtain the multi-omics datasets commonly used in experiments and preprocess them;

[0051] In single-cell multi-omics research experiments, it is first necessary to obtain various types of omics data, such as transcriptome, proteome, and metabolome data. These data sets usually come from public databases or samples sequenced by the laboratory. After obtaining the data, necessary preprocessing steps are performed, which include but are not limited to data cleaning, standardization, and missing value processing. The purpose of preprocessing is to ensure data quality and provide accurate input for subsequent analysis. For example, in transcriptome data, gene expression levels may need to be normalized to eliminate systematic biases in experimental operations.

[0052] Step 2: For the preprocessed multi-omics dataset, the deep learning-based SpectralNet model is then used to learn the mapping of the data to its spectral embedding space. Spectral embedding is a powerful dimensionality reduction technique that can reveal the underlying structure of the data, and is particularly suitable for revealing intrinsic patterns and associations in complex biological data, including:

[0053] Step 2.1: Data Dimensionality Reduction. In the spectral embedding process, the multi-omics data first needs to be effectively reduced in dimensionality. This can be achieved by using an autoencoder model. An autoencoder is an unsupervised learning neural network that learns to encode high-dimensional input data into a low-dimensional hidden layer and then decodes it back to the original data space from the hidden layer. Through this process, the model is able to learn a compressed representation of the data. Applying an autoencoder to the concatenated multi-omics data not only reduces the dimensionality of the data, but also helps capture complex and abstract biological information across different omics levels.

[0054] Step 2.2: Similarity calculation. Spectral embedding learning requires sample similarity information, so the present invention uses a twin neural network model to calculate sample similarity. The K nearest neighbor algorithm is used to identify the k nearest neighbors of a data point and construct positive and negative sample pairs. Each sample and its k nearest neighbor samples constitute a positive sample pair, and each sample and another k randomly selected samples constitute a negative sample pair. By minimizing the contrast loss L Siamese Training Siamese neural network, contrast loss L Siamese The definition is as follows:

[0055]

[0056] in is the potential representation learned by the twin neural network, and c is the margin, which is usually set to 1. Based on the learned potential representation Calculate the Euclidean distance between samples and use the Gaussian kernel function to convert the distance into sample similarity information.

[0057] Step 2.3: Spectral embedding mapping learning. In order to solve the scalability problem of the graph-based method, a deep neural network is used in the present invention to learn the spectral embedding mapping function. Based on the mapping function, the optimal spectral embedding low-dimensional representation {y n *}for:

[0058]

[0059] Step 2.4: Similarity graph construction. Based on the spectral embedding low-dimensional representation obtained from SpectralNet, we calculate the Euclidean distance between samples. This distance measures the relative position of samples in the low-dimensional space and reflects the direct similarity or difference between samples. In order to convert these Euclidean distances into more expressive similarity information, a Gaussian kernel function is applied for conversion. This process involves calculating the similarity between each pair of samples, strengthening the connection between neighboring samples through the exponential decay characteristics of the Gaussian kernel, while weakening the influence between distant samples. Finally, this similarity information is used to construct a comprehensive sample similarity graph, which serves as a key data structure in subsequent analysis to support more complex data-driven tasks such as clustering and classification.

[0060] Step 3: In order to integrate different omics data in a unified dimensional space, a variational autoencoder (VAE) model is used. VAE is a deep generative model that can not only learn the compressed representation of data, but also capture the uncertainty and variability in the data through probability distribution, including:

[0061] Step 3.1: For each omics data in the multi-omics dataset, an encoder is independently trained to learn the distribution of the omics data in the low-dimensional latent space. This step allows the model to capture the biological information and variability unique to each omics, ensuring that the key information of a specific omics is not lost during the integration process.

[0062] Step 3.2: The latent representations learned from each independent encoder are weighted and fused through a certain strategy. This fusion process takes into account that the contributions of different omics data may be different, so the selection of weighting coefficients may be based on the intrinsic quality of the data, the amount of information, or its importance in biological processes. In this way, a single-cell shared latent representation that comprehensively reflects multi-omics features can be obtained.

[0063] Step 3.3: Each omics data is equipped with an independent decoder to reconstruct its own original data from the shared latent representation. This step verifies the validity and integrity of the shared latent representation, ensuring that the integrated representation can restore the key features of various omics, while also supporting subsequent biological interpretation and analysis.

[0064] Step 4: Based on the shared potential representation obtained by integration, the cells are clustered using a Gaussian mixture model (GMM). GMM is a probabilistic model that assumes that the data consists of a mixture of multiple Gaussian distributions and is suitable for processing complex biological data with multiple potential subpopulations. The clustering results can reveal the distribution of different cell types and help researchers understand the heterogeneity between cells and the role of different omics features in cell function. This step is key in multi-omics data analysis and provides a basis for a deep understanding of the organizational structure and functional dynamics of biological systems.

[0065] Step 5: Construct the objective function based on the maximum likelihood estimation, and incorporate the similarity information by minimizing the JS divergence of the potential representation of similar cells and cluster assignment to obtain the final objective function:

[0066]

[0067] Step 6: Derived and simplified the objective function to obtain the final loss function, and the stochastic gradient descent method is used to iteratively update the variational autoencoder model and Gaussian mixture model, including:

[0068] Step 6.1: For the objective function (3), introduce the variational upper bound of the JS divergence Alternative JS divergence, Defined as:

[0069]

[0070] in

[0071] Step 6.2: Decompose into and Two related items:

[0072]

[0073] Will The two decomposed items are added and averaged and then input into the objective function, and the objective function becomes:

[0074]

[0075] Step 6.3: Further simplify the estimation of the objective function, where the first term in the objective function (7) can be decomposed as:

[0076]

[0077] in φ1 and φ2 represent trainable neural network parameters.

[0078] Further simplifying the terms in formula (8) yields:

[0079]

[0080] Where L represents the number of Monte Carlo samples in the stochastic gradient variational Bayes (SGVB) estimator, m represents the number of omics, and D represents and The characteristic dimension of yes The d-th element of is the mean of the omics data reconstructed by the variational autoencoder model, represents the dth element of the lth sample of *, λ (v) It is a predefined parameter and is usually set to 1.

[0081]

[0082] in K is the number of cluster categories, M is μ k and The characteristic dimension, μ k and represents the mean and variance of the kth component of the Gaussian mixture model, and represents the mean and variance of the shared latent representation of the i-th sample, *| m Represents the mth element of *.

[0083]

[0084] where π ik represents the prior probability of cluster k and satisfies

[0085]

[0086] Integrating formulas (9), (10), (11), and (12), formula (8) can be rewritten as:

[0087]

[0088] Similarly, the second term in formula (7) can be estimated as:

[0089]

[0090] In order to achieve more flexible alignment in the latent space and obtain more accurate clustering assignment, the present invention adjusts the coefficients of alignment loss and clustering loss to α and β. Therefore, the final loss function is

[0091] The number is:

[0092]

[0093] L (i) =L (i),rec +α*(L (i),align1 +L (i),align2 )+β*L (i),cluster (16)

[0094] L (i,j)=L (i,j),rec +α*(L (i,j),align1 +L (i,j),align2 )+β*L (i,j),cluster (17)

[0095] Step 7: Use the trained variational autoencoder model to obtain the shared potential representation of single cells, and use the trained Gaussian mixture model based on the shared potential representation to obtain the clustering distribution of single cells.

[0096] Step 8: Use clustering tasks to evaluate the performance of feature selection methods. Three widely used clustering evaluation indicators are used to evaluate the clustering effect: clustering accuracy (ACC), normalized mutual information (NMI) and purity. Among them, clustering accuracy represents the proportion of correctly predicted samples to the total number of samples; normalized mutual information is used to measure the degree of correlation between two random variables; purity represents the proportion of the main categories of samples contained in each cluster. The numerical range of these three evaluation indicators is between 0 and 1, and the larger the value, the better the performance.

[0097] Embodiment 2:

[0098] The present invention provides a cell type identification system based on multi-omics decoupled representation and graph embedding, comprising:

[0099] The similarity graph construction module is used to identify similar samples in the data, capture the interactions and similarities between cells, and provide an accurate graph structure representation for subsequent analysis. The core function of this module is to identify similar samples in multi-omics data and build a detailed graph structure representation based on this. By calculating the similarity between samples (for example, using the Gaussian kernel function to convert the Euclidean distance), this module can effectively capture the interactions and similarities between cells and form a similarity graph. This graph not only reflects the direct relationship between samples, but also reveals the underlying group structure, providing a basis for subsequent data analysis such as clustering and classification;

[0100] The decoupled representation learning module is used to learn the potential representation of each omics data in the same dimensional space. In this module, independent encoders are used to learn the potential representation of different omics data in the same dimensional space. This decoupled representation learning method allows each omics data to express its information independently while maintaining the relevance between the data through a unified framework. This is particularly important for processing complex interactions and intrinsic connections between omics data;

[0101] The fusion module is used to construct omics weight parameters. By weighted fusion of the potential representations learned from each omics data, the cell-shared potential representation is obtained. The task of the fusion module is to weighted fuse the potential representations learned from each omics data by constructing omics weight parameters. This process involves determining the contribution of each omics data to the comprehensive biological representation, and then generating a single-cell shared potential representation. This weighted fusion not only takes into account the uniqueness of each omics data, but also optimizes the information integrity and expressiveness of the integrated data representation;

[0102] Clustering module, which is used to cluster multi-omics data and realize clustering assignment of cell types based on shared latent representation and Gaussian mixture model. The clustering module uses the fused shared latent representation to cluster cells through Gaussian mixture model. This model assumes that the data is composed of multiple Gaussian distributions, and can consider the probability distribution of data in cell type identification, thereby more accurately reflecting the natural classification of cells. Clustering results help identify and interpret biologically meaningful cell populations;

[0103] The pre-training module is used to pre-train the variational autoencoder model to ensure that the variational autoencoder can reconstruct multi-omics data, and to initialize the Gaussian mixture model to ensure that the Gaussian mixture model can capture the category information of the data. The purpose of the pre-training module is to ensure that the variational autoencoder model can effectively reconstruct multi-omics data and provide support for the initialization of the Gaussian mixture model. Through pre-training, model parameters can be adjusted in advance to ensure that the model has the basic ability to process complex data before more detailed training. This step is crucial to improving the stability and performance of the model;

[0104] In the training module, the variational autoencoder and Gaussian mixture model are further optimized based on the graph constructed by the graph embedding model to realize the integration and utilization of similarity information. In the training module, the variational autoencoder and Gaussian mixture model are further optimized using the similarity graph constructed based on the graph embedding model. In this process, the model improves its learning and representation of data structure by integrating and utilizing similarity information, thereby improving the accuracy of cell type recognition and the generalization ability of the model.

[0105] The recognition module is used to identify cell types based on the trained variational autoencoder model and Gaussian mixture model. The final recognition module identifies cell types based on the trained variational autoencoder and Gaussian mixture model. This module is the end point of the entire analysis process, and its output can be directly used in biological research and clinical applications to provide support for disease diagnosis, biomarker discovery, and the development of treatment strategies.

[0106] Embodiment three:

[0107] The purpose of this embodiment is to provide an electronic device: comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the cell type identification method based on multi-omics decoupled representation and graph embedding as in the first embodiment is implemented to identify the cell type;

[0108] Memory: Electronic devices are equipped with high-speed, large-capacity memory for storing a variety of omics data (such as transcriptome data, proteome data, etc.) and related analysis tools and programs. This memory can handle large data sets from multiple data sources and ensure the security and integrity of the data.

[0109] Processor: The processor in the device is high-performance and capable of processing complex algorithms and models, including but not limited to variational autoencoders and graph embedding models. The main function of the processor is to perform computationally intensive tasks such as data preprocessing, model training, and identification of cell types.

[0110] Computer program: The computer program stored on the memory is designed for multi-omics data analysis. The program contains multiple modules, including data processing module, decoupled representation learning module, fusion module, clustering module, etc., which can be run on the processor to realize automatic identification and classification of cell types.

[0111] Embodiment 4:

[0112] The purpose of this embodiment is to provide a computer scale storage medium:

[0113] A computer program is stored thereon, and when the program is executed by the processor, the cell type identification method based on multi-omics decoupling representation and graph embedding as in the first embodiment is implemented;

[0114] The computer-readable storage medium may be a solid-state drive (SSD), a hard disk drive (HDD), a USB flash drive, an optical disk, or any other form of digital data storage device that has a high data transfer rate and sufficient storage capacity to store complex computer programs and their required data sets.

[0115] The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1. For the specific implementation, please refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium, any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0116] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0117] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A cell type identification method based on multi-omics decoupled representation and graph embedding, characterized in that: The following steps are involved: Step 1: Obtain single-cell multi-omics data and perform preprocessing; Step 2: Use the graph embedding model to obtain the cell spectrum embedding representation, calculate the similarity of single cell samples based on the spectrum embedding representation, and then construct a sample similarity graph; Step 3: Use the variational autoencoder model to map different omics data into the same dimensional space for data integration to obtain the single-cell shared potential representation; Step 4: Cluster the cells using a Gaussian mixture model based on the shared potential representation to obtain clustering assignments of the cells; Step 5: Construct the objective function based on maximum likelihood estimation and incorporate similarity information by minimizing the JS divergence of the latent representation and cluster assignment of similar cells; Step 6: Derived and simplified the objective function to obtain the final loss function, and iteratively updated the variational autoencoder model and Gaussian mixture model using the stochastic gradient descent method; Step 7: Use the trained model to obtain the shared potential representation and cluster assignment of single cells.

2. The cell type identification method based on multi-omics decoupled representation and graph embedding according to claim 1, characterized in that: Furthermore, the training process of the graph embedding model includes the following steps: Splice the preprocessed single-cell multi-omics data; Use the autoencoder model to reduce the dimension of the concatenated data; Use the twin neural network model to learn the sample similarity matrix of the reduced-dimensional data; Based on spectral clustering, a neural network mapping function is learned from the dimensionality reduction space to the spectral embedding space, and the sample similarity is calculated according to the spectral embedding representation to obtain the sample similarity graph.

3. The cell type identification method based on multi-omics decoupled representation and graph embedding according to claim 1, characterized in that: Furthermore, the objective function of the graph embedding model is: where Y = [y1,y2,…,y N ], N is the number of data samples, x i and x j is the encoded representation after dimensionality reduction by the encoder, w(x i ,x j ) represents the sample x in the similarity matrix learned by the twin neural network i and x j The similarity of i and j The samples x are i and x j The spectral embedding of is a low-dimensional representation.

4. The cell type identification method based on multi-omics decoupled representation and graph embedding according to claim 1, characterized in that: In step 5, the objective function of the maximum likelihood estimation is: Among them, φ and θ represent the set of encoder parameters and the set of decoder parameters in each omics variational autoencoder model, respectively, {x (v) } is a universal multi-omics variable, {x (v) }:={x (1) ,…,x (v) ,…,x (V) }, where x (v) is the general variable of the v-th omics.

5. The cell type identification method based on multi-omics decoupled representation and graph embedding according to claim 1, characterized in that: By adding similarity information constraints to the objective function, the objective function becomes: Among them, w ij (∑ j w ij =1) represents the sample similarity graph learned by the graph embedding model and The similarity of represents the shared potential representation of the cell, M is the feature dimension of the latent variable z, c∈{0,1} K×1 represents the clustering assignment of cells, K is the number of cluster categories, and JS(·,·) represents the JS divergence between the two distributions.

6. The cell type identification method based on multi-omics decoupled representation and graph embedding according to claim 5, characterized in that: Furthermore, the objective function is simplified and estimated to obtain the final loss function. The variational autoencoder model and the Gaussian mixture model are iteratively updated using the stochastic gradient descent method. The simplified estimation of the objective function specifically includes: Introduce the variational upper limit of JS divergence to replace JS divergence; Will Decompose into and Two items related; The objective function is estimated and solved to obtain the final loss function.

7. A cell type recognition system based on multi-omics decoupled representation and graph embedding, characterized by: The cell type identification method based on multi-omics decoupled representation and graph embedding applied to claims 1-6 comprises: Similarity graph building module, used to identify similar samples in the data, capture the interactions and similarities between cells, and provide accurate graph structure representation for subsequent analysis; The decoupled representation learning module is used to learn the potential representation of each omics data in the same dimensional space; The fusion module is used to construct the omics weight parameters and obtain the cell-shared potential representation by weighted fusion of the potential representations learned from each omics data; Clustering module, used to perform cluster assignment on multi-omics data, and realize cluster assignment of cell types based on shared latent representation and Gaussian mixture model; The pre-training module is used to pre-train the variational autoencoder model to ensure that the variational autoencoder can reconstruct multi-omics data, and initialize the Gaussian mixture model to ensure that the Gaussian mixture model can capture the category information of the data; The training module further optimizes the variational autoencoder and Gaussian mixture model based on the graph embedding model to integrate and utilize similarity information. The recognition module is used to identify cell types based on the trained variational autoencoder model and Gaussian mixture model.

8. An electronic device, characterized in that: The cell type identification method based on multi-omics decoupled representation and graph embedding applied to claims 1-6 comprises a memory, a processor, and a computer program stored in the memory and executable on the processor.

9. A computer-readable storage medium, characterized in that: An electronic device as claimed in claim 8, wherein a computer program is stored thereon.

Citation Information

Patent Citations

  • Cell type identification method, system and equipment based on multi-omics data integration

    CN116226698A

  • Single-cell multi-omics data cross-modal generation method and system

    CN118173173A

  • End-to-end signalized intersection transition state estimator with scene graphs over semantic keypoints

    US20220261583A1

  • Attention-based multiple instance learning for whole slide images

    US20230419491A1

Cited By

  • Single cell data multi-clustering method and system based on multi-omics data fusion

    CN119741978A