Single-cell multi-omics data integration method

Abstract

The invention discloses a single-cell multi-omics data integration method, which relates to the technical field of biological information, and is technically characterized by comprising the following steps of: S1, preprocessing single-cell RNA (Ribonucleic Acid) sequencing data and single-cell ATAC (Attenuated Atmospheric Converter) sequencing data, including two parts of data screening and gene selection; s2, constructing a low-dimensional shared space, and constructing the low-dimensional shared space of two kinds of omics data by using an auto-encoder; s3, training a deep pyramid convolutional neural network by using single-cell RNA sequencing data in the low-dimensional shared space; and S4, classifying the single-cell ATAC sequencing data by using the trained deep pyramid convolutional neural network to obtain a cell type tag, and realizing data integration of the single-cell RNA sequencing data and the single-cell ATAC sequencing data. According to the method, a low-dimensional shared space is constructed by using an auto-encoder network, the similarity between different omics data is maximized, the distribution between different cell types of data is learned through a text convolutional network, and finally the purpose of integrating multiple omics data is achieved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a single-cell multi-omics data integration method. Background Art

[0002] In recent years, single-cell multi-omics technology is no longer limited to transcriptomes, but also includes multiple omics such as epigenomes and proteomes. This technology is able to detect two or more molecules at the same time, such as the combined detection of single cell RNA sequencing (scRNA-seq) and single cell ATAC sequencing (scATAC-seq), as well as the simultaneous detection of single cell transcriptomes, immune repertoires, and surface proteins. This comprehensive approach provides a more comprehensive, refined, and complete analysis strategy for studies at the single-cell level. In particular, scATAC-seq (single-cell ATAC sequencing) allows us to explore heterogeneity between cells at the single-cell level. By detecting open chromatin regions, we can identify different cell types, subtypes, and states, revealing the trajectory of cell development and differentiation.

[0003] However, because changes in chromatin accessibility are rare and difficult to capture, scATAC-seq is often extremely sparse and high-dimensional, making it difficult to directly identify cell types. In contrast, scRNA-seq (single-cell RNA sequencing) often provides richer gene expression information, which helps to more accurately annotate and identify cell types. By transferring cell type labels from scRNA-seq to scATAC-seq, researchers can integrate these two omics datasets to obtain a more comprehensive perspective.

[0004] For large-scale and extremely sparse multi-omics datasets, a deep learning method is currently needed to integrate scATAC-seq and scRNA-seq data. Summary of the invention

[0005] The purpose of the present invention is to solve the above problems and provide a single-cell multi-omics data integration method.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows:

[0007] The present invention provides a single-cell multi-omics data integration method, comprising the following steps:

[0008] S1: Preprocessing of single-cell RNA sequencing data and single-cell ATAC sequencing data, including data screening and gene selection;

[0009] S2: Construct a low-dimensional shared space, using autoencoders to construct a low-dimensional shared space for two omics data;

[0010] S3: Training deep pyramid convolutional neural networks using single-cell RNA sequencing data in a low-dimensional shared space;

[0011] S4: Use the trained deep pyramid convolutional neural network to classify the single-cell ATAC sequencing data, obtain cell type labels, and realize data integration of single-cell RNA sequencing data and single-cell ATAC sequencing data.

[0012] The present invention is further configured as follows: Step S1 specifically comprises:

[0013] S1-1. Data collection: Human peripheral blood mononuclear cell dataset was used;

[0014] S1-2. Data preprocessing: screening out high-quality data from single-cell RNA sequencing data and single-cell ATAC sequencing data, and selecting the common gene features of the two.

[0015] The present invention is further configured as follows: the specific operation of step S2 is: using a single-cell RNA sequencing dataset with label information and a single-cell ATAC sequencing dataset without label information to jointly train an autoencoder, learn a shared low-dimensional embedding space, and maximize the similarity between the two omics data.

[0016] The present invention is further configured as follows: the specific operation of step S3 is: learning the distribution of different cell types, and finally using the trained deep pyramid convolutional neural network to classify the unlabeled single-cell ATAC sequencing data to determine its cell type.

[0017] Compared with the prior art, the present invention has the following beneficial effects: the present invention constructs a low-dimensional shared space by using an autoencoder network, maximizes the similarity between different omics data, learns the distribution between different cell type data through a text convolutional network, and finally achieves the purpose of integrating multi-omics data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a method model diagram in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of the present invention will be further described in detail below in conjunction with the embodiments of the present invention and the accompanying drawings. 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 should fall within the scope of protection of the present invention.

[0020] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will be described in detail below in conjunction with the embodiments.

[0021] Example:

[0022] 1. Data preprocessing

[0023] 1.1 Data Collection

[0024] This experiment uses a human peripheral blood mononuclear cell (PBMC) dataset, which is a paired scATAC-seq and scRNA-seq data at the cell level, with the GEO access number GSE126074. We used Signic for calculations and ultimately obtained a gene activity matrix containing 79,190 cells.

[0025] 1.2 Data Preprocessing

[0026] For scATAC-seq and scRNA-seq data, we screened out the same gene features for subsequent use.

[0027] 2. Autoencoder builds a low-dimensional shared space

[0028] Given a scRNA-seq dataset X and a scATAC-seq dataset Y, we first screen out the features common to the two omics data sets, where σ represents the number of features and represents the scRNA-seq gene expression matrix, and is represented by represents the label of each cell, where is a vector of dimension σ, x∈(1,2,...,X). Similarly, represents the scATAC-seq active gene expression matrix, where y∈(1,2,...,Y).

[0029] We use an autoencoder to construct a low-dimensional shared space by sharing weights, where the weights of the autoencoder are uniformly represented by θ. represents the mapping function of the autoencoder. And a suitable loss function is applied to the output of the fully connected layer to minimize the embedding loss of the two omics data.

[0030] NearestNeighborDistance Ratio (NNDR) loss

[0031] Similar to the PCA algorithm, NNDR loss (nearest neighbor distance ratio loss) aims to capture the orthogonal features of two omics data in a low-dimensional embedding space. For each batch of data where h x ,h y is from and Sample out the minimum batch for training.

[0032]

[0033] Cosine Similarity Loss

[0034] The cosine similarity loss is mainly applied to the embedding layer output to maximize the similarity between the aligned scRNA-seq and scATAC-seq data.

[0035]

[0036] Through training, we find the i that minimizes the loss so that 3. Text convolutional network for cell label transfer

[0037] The text convolutional network is a classification model based on convolutional neural networks. It is simple, efficient and powerful in feature extraction, but it is easy to cause the problem of excessive computation. Therefore, we use Deep Pyramid Convolutional Neural Network (DPCNN). The core of DPCNN is the deep pyramid structure, which gradually reduces the size of the feature map through multiple pooling layers. This structure helps to capture features at different levels, from local to global, so that the model has better expression ability. At the same time, using a fixed-size local receptive field (1*3 convolution kernel) for convolution operations helps to capture information between local features, making the model more sensitive to local information in cells. In order to solve the problem of excessive computation, DPCNN gradually reduces the size of the feature map by stacking pooling layers, thereby effectively extracting the global information of the cell and improving training efficiency.

[0038] The above specific embodiments are merely explanations of the present invention and are not limitations of the present invention. After reading this specification, those skilled in the art may make modifications to the embodiments without any creative contribution as needed. However, such modifications are protected by the patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for integrating single-cell multi-omics data, characterized in that: The following steps are involved: S1: Preprocessing of single-cell RNA sequencing data and single-cell ATAC sequencing data, including data screening and gene selection; S2: Construct a low-dimensional shared space, using autoencoders to construct a low-dimensional shared space for two omics data; S3: Training deep pyramid convolutional neural networks using single-cell RNA sequencing data in a low-dimensional shared space; S4: Use the trained deep pyramid convolutional neural network to classify the single-cell ATAC sequencing data, obtain cell type labels, and realize data integration of single-cell RNA sequencing data and single-cell ATAC sequencing data.

2. A single-cell multi-omics data integration method as claimed in claim 1, characterized in that: The step S1 is specifically as follows: S1-1. Data collection: Human peripheral blood mononuclear cell dataset was used; S1-2. Data preprocessing: screening out high-quality data from single-cell RNA sequencing data and single-cell ATAC sequencing data, and selecting the common gene features of the two.

3. A single-cell multi-omics data integration method as claimed in claim 1, characterized in that: The specific operation of step S2 is: using the single-cell RNA sequencing dataset with label information and the single-cell ATAC sequencing dataset without label information to jointly train an autoencoder, learn a shared low-dimensional embedding space, and maximize the similarity between the two omics data.

4. A single-cell multi-omics data integration method as claimed in claim 1, characterized in that: The specific operation of step S3 is: learning the distribution of different cell types, and finally using the trained deep pyramid convolutional neural network to classify the unlabeled single-cell ATAC sequencing data to determine its cell type.

Images