Cell classification method based on single cell genome expression, medium and equipment

Abstract

The invention relates to a cell classification method based on single cell genome expression, a medium and equipment, which comprises the following steps: constructing a redivided BED file based on genome sequence data groups and converting the BED file into a file in an agreed format, and constructing a reference between the genome sequence data groups to obtain a T2T genome reference; respectively acting with reference information based on the genome sequence data set and the original real data, and processing to obtain genome sequence data characterization and a corresponding cell type label; processing genomic sequence data representation, inputting the processed genomic sequence data representation into the constructed full-connection network model, and training to a preset condition; extracting to-be-classified genome sequence data representation, processing the representation, and inputting the representation into the trained full-connection network model to obtain a classification result; media and devices are implemented based on the method. According to the invention, genome expression information is fully utilized to predict cell types; and after the genome is partitioned again, more gene information can be extracted, and the cell type identification accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, such as the technical field of information and communication technology particularly suitable for genetic or protein-related data processing in computational molecular biology, and in particular to a cell classification method, medium and device based on single-cell genome expression. Background Art

[0002] Single-cell RNA sequencing (scRNA-seq) is currently one of the main methods for defining cell states at the molecular level. The analysis of scRNA-seq datasets usually starts with dimensionality reduction and clustering. According to the method of annotating cells with cell types based on scRNA-seq data, the data is clustered and the cell types of each cluster are annotated based on the characteristic marker genes (stromal cells) expressed by each clustered cell cluster. However, clusters represent groups of cells with relatively similar gene expression profiles, and the cells clustered together may be of the same or similar cell types. Since different cellular phenomena such as cell transformation may not be fully captured in scRNA-seq datasets, cells may be assigned the wrong identity. At the same time, manual annotation of cell types is usually very time-consuming and has limited reproducibility. During this period, personnel with professional background knowledge are required to annotate cell types for cell clusters based on the expression levels of marker genes in each cell cluster and the markers corresponding to the cell types in the database, and there is a probability of deviation.

[0003] Due to differences in human operation and sequencing depth between different platforms, there will be batch effects in scRNA-seq data, and the genes expressed by each cell in different life cycles will also be slightly different. The traditional method is to manually annotate cell types, which is time-consuming, labor-intensive and extremely costly. In order to overcome these limitations, other computational methods for automatic annotation of cell clusters have emerged. With the rapid development of gene sequencing technology and the rise of artificial intelligence, the use of deep learning technology to predict cell types from the gene expression levels of cells has become a key means to achieve cell type annotation. However, the current automatic annotation methods of cell types are all based on the same gene division, resulting in many gene sequences in the genome sequence not being fully utilized. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a cell classification method, medium and device based on single-cell genome expression.

[0005] The concept of the present invention is that, considering that the genome is a collection of complete genetic information of an organism, composed of DNA, including coding regions and non-coding regions; the current conventional single-cell transcriptome analysis mainly focuses on the expression information of the coding regions in the genome, ignoring the potential functional information of a large number of non-coding regions in the genome. These neglected regions include but are not limited to long non-coding RNA (lncRNA), microRNA (miRNA), enhancers, promoters, pseudogenes, transposons and introns, etc., which may play a key role in gene regulation, epigenetic modification, cell fate determination and the occurrence and development of complex diseases. Integrating and utilizing the information of these non-coding regions in single-cell transcriptome analysis is a good method for the research of single-cell transcriptome analysis. Information can not only provide a more comprehensive gene regulatory network map, but also reveal functional RNA outside the coding region and its role in different biological processes. Therefore, the genome is re-divided into blocks to make full use of the sequence information in the genome. Through blocking, each genome sequence has more gene sequences, which can be better used for cell type identification and promote the annotation of single-cell transcriptomes. Here, "having more gene sequences" means abandoning the currently agreed length of each genome sequence and redefining the length of each genome sequence. To be precise, in order to increase the length of each genome sequence, each redefined genome sequence contains more gene sequences (ATGC sequences) than before.

[0006] The technical solution adopted by the present invention is a cell classification method based on single-cell genome expression, the method comprising the following steps:

[0007] S1 constructs the re-divided BED file based on the genome sequence data group and converts it into a agreed format file;

[0008] S2 builds a reference between genome sequence data groups to obtain a reference for the T2T genome; the T2T genome here is a preset (original) gene division, that is, the division of each gene sequence, which is used to guide the software on how to divide the entire gene sequence into individual genes;

[0009] S3 is based on the genome sequence data set and the original real data, respectively interacting with the reference information of S2, using cellranger software and corresponding processing flow to obtain the genome sequence data representation and the corresponding cell type label;

[0010] S4 processes the genome sequence data representation of S3 and inputs it into the constructed fully connected network model, and trains it to the preset conditions;

[0011] S5 extracts the genome sequence data representation to be classified, and after processing, inputs it into the trained fully connected network model to obtain the classification result.

[0012] Preferably, in S1, the BED file based on the genome sequence data set includes the starting position, the ending position and the corresponding gene number of the genome sequence data set.

[0013] Preferably, in S2, the reference data is processed based on a standard data format and CellRanger software.

[0014] Preferably, in S3, 10X matrix data of the genome sequence data set is generated after processing, and the corresponding data expression matrix Z is obtained, where the rows of Z correspond to the block numbers of the genome sequence data set, and the columns correspond to each cell; the i-th row and j-th column of Z are the expression levels of gene i in cell j.

[0015] Preferably, the cell type labels generated after S3 are processed, the categories are marked, the corresponding relationship S between each cell number and category is obtained, and the training set and the test set are delineated from the data expression matrix Z.

[0016] Preferably, in S4, processing the genome sequence data representation of S3 includes scaling the data in the data expression matrix Z to obtain a standardized new matrix Z with m rows and n columns. new ,

[0017]

[0018] Among them, min(Z) is the minimum value in the matrix Z, max(Z) is the maximum value in the matrix, min_val and max_val are the lower limit and upper limit of the preset scaling range.

[0019] Preferably, the new matrix Z is processed new , get the input Z of the fully connected network model in ,

[0020]

[0021] Among them, Z new i,j is the matrix Z new The value of the element in the i-th row and j-th column, where i is an integer from 0 to m-1, j is an integer from 0 to n-1, and t is the column number, that is, an integer from 0 to n.

[0022] Preferably, in S4, the fully connected network model includes a sequentially connected input layer, several groups of linear processing modules and an output layer; the linear processing modules are three groups, and any group of the linear processing modules includes a sequentially connected activation function layer, a dropout layer and a fully connected layer; the number of nodes in the linear layer of the linear processing module decreases successively.

[0023] A computer-readable storage medium stores a cell classification program based on single-cell genome expression, which implements the above-mentioned cell classification method based on single-cell genome expression when executed by a processor.

[0024] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the program, the cell classification method based on single-cell genome expression is implemented.

[0025] The present invention provides a cell classification method, medium and device based on single-cell genome expression, which construct a re-divided BED file based on a genome sequence data group and convert it into a file in a specified format, construct a reference between genome sequence data groups, and obtain a reference for a T2T genome; based on the genome sequence data group and original real data, the reference information is respectively acted on, and the genome sequence data representation and the corresponding cell type label are obtained by processing with cellranger software and a corresponding processing flow; the genome sequence data representation is input into a constructed fully connected network model after processing, and trained to a preset condition; the genome sequence data representation to be classified is extracted, and the processed genome sequence data is input into the trained fully connected network model to obtain a classification result; and a medium and a device are implemented based on the method.

[0026] The beneficial effect of the present invention is that it does not rely on traditional reference genomes to annotate cell types, but instead makes full use of genome expression information to predict cell types; after the genome is re-divided into blocks, each gene block contains a large number of gene sequences, and can make full use of gene blocks that match cell sequences, thereby facilitating the extraction of more gene information to improve the accuracy of cell type identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 is a flow chart of an embodiment of the present invention;

[0029] Figure 3 This is a graph showing the accuracy of peripheral blood mononuclear cell (PBMC) type prediction using a cell classification method based on single-cell genome expression. DETAILED DESCRIPTION

[0030] The present invention is further described in detail below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto.

[0031] The present invention relates to a cell classification method based on single-cell genome expression. The method is described below using peripheral blood mononuclear cell (PBMC) data as an example.

[0032] (1) Construct a re-divided BED file based on the genome sequence data set and convert it into a agreed format file;

[0033] First, a BED file for partitioning the T2T human genome sequence is constructed. The file is divided into four columns, which represent the genome sequence data group number (a total of 23 entries), the starting position of the bin block gene (genome sequence data group, the same below), the ending position of the bin block gene, and the bin block gene number. The starting position of the bin block gene and the bin block gene number of each chromosome start from 1. The size of each bin block is 80,000, which is much larger than the existing partition size. When the number of base pairs is not enough to support the partition size, multiple genes are combined, that is, a bin block contains multiple previous genes, thereby obtaining more and richer information. After the BED file is constructed, the bedtogenepred and genepredtogtf tools are used to convert the BED file into a gtf file.

[0034] In the specific implementation process, in fact, the experiment is also divided into 5000, 10000, 20000 and 40000 lengths of each gene sequence. Different division schemes have an impact on the subsequent cell type prediction accuracy. In this embodiment, the prediction accuracy after taking 80000 is the highest. In fact, when dividing different tissues, this data is not unique.

[0035] (2) construct references between genome sequence data sets to obtain references for T2T genomes;

[0036] Use the cellranger mkref pipeline of cellranger to build a bin gene block reference. The parameters used are "-genome T2T-bin-fasta GCF_009914755.1_T2T-CHM13v2.0_genomic.fna-nthreads 10-genes genomic_chr.gtf", where "-genome T2T-bin" indicates the directory where the cellranger run results are saved, "-fasta GCF_009914755.1_T2T-CHM13v2.0_genomic.fna" indicates the fasta file used when building the reference. This file is the fasta file that comes with the T2T genome, "-nthreads 10" indicates the number of CPU threads used when running the cellranger mkref pipeline, and "-genes genomic_chr.gtf" indicates the gtf file (1) used when building the reference;

[0037] Use cellranger's cellranger mkref pipeline to build the T2T genome reference. The parameters used are "-genome T2T-fasta GCF_009914755.1_T2T-CHM13v2.0_genomic.fna-nthreads 10-genes genomic.gtf", where "-genome T2T" indicates the directory where the cellranger run results are saved, "-fasta GCF_009914755.1_T2T-CHM13v2.0_genomic.fna" indicates the fasta file used when building the reference, which is the fasta file that comes with the T2T genome, "-nthreads10" indicates the number of CPU threads used when running the cellranger mkref pipeline, and "-genes genomic.gtf" indicates the gtf file that comes with the T2T genome data;

[0038] (3) Based on the genome sequence data set and the original real data, respectively interacting with the reference information of (2), using cellranger software and corresponding processing flow to obtain genome sequence data representation and corresponding cell type labels;

[0039] Use the cellranger count pipeline of cellranger to generate bin 10X matrix data. The parameters used are "-id PBMC_20K_bin-localcores 24-localmem 200-expect-cells24000-transcriptome T2T-bin-fastqs PBMC_20K_fastqs", where "-id PBMC_20K_bin" indicates the directory where the results are saved, "-localcores 24" indicates the number of CPU cores used when running the cellranger count pipeline, "-localmem 200" indicates the memory size used, "-expect-cells 24000" indicates the estimated number of cells in the sample, "-transcriptome T2T-bin" indicates the directory name of the reference generated in step 2), and "-fastqs PBMC_20K_fastqs" indicates the directory name of the fastq file of the sample;

[0040] Use the cellranger count pipeline of cellranger to generate gene 10X matrix data. The parameters used are "-id PBMC_20K_gene-localcores 24-localmem 200-expect-cells 24000-transcriptome T2T-fastqs PBMC_20K_fastqs", where "-id PBMC_20K_gene" indicates the directory where the results are saved, "-localcores 24" indicates the number of CPU cores used when running the cellranger count pipeline, "-localmem 200" indicates the memory size used, "-expect-cells 24000" indicates the estimated number of cells in the sample, "-transcriptome T2T" indicates the directory name of the reference generated in step 3), and "-fastqs PBMC_20K_fastqs" indicates the directory name of the fastq file of the sample;

[0041] The bin 10X matrix data generated by the Seurat standard process is processed. In the processing process, the function GetAssayData(object=seuart_object, assay='RNA') is used to obtain the bin block gene expression matrix Z of the bin 10X matrix data, where the rows of Z are the bin block numbers and the columns are the cell numbers. The gene 10X matrix data generated by the Seurat standard process is processed. According to the manual cell types in the processing results, the correspondence S between the cell numbers and the cell types is obtained. 20% of the cells are selected from the bin block gene expression matrix Z as the test set, and the remaining 80% of the cells are used as the training set.

[0042] In the present invention, cellranger software and corresponding processing flow are used to obtain genome sequence data representation, which is the data required for subsequent Seurat processing, i.e., matrix Z, which represents the gene expression level of cells, with row numbers representing genes and column numbers representing cells; from a single row, it represents the expression level of the gene represented by this row in each cell, and from a single column, it represents the gene expression level of the cell represented by this column, and the specific number of row i and column j represents the expression level of gene i in cell j; the "expression level" here is equivalent to the obvious degree of a gene in sequencing, and different genes have different expression levels in different cells. Seurat's cell clustering and classification are based on the gene expression level of cells;

[0043] Based on this matrix Z, we can identify specific cells based on the column index. For example, the 5th column is named "AAACCCATCGGTAACT", which represents a unique cell. Similarly, we can identify a unique gene based on the row.

[0044] The correspondence S refers to the type of each cell obtained after analysis by the Seurat process. For example, the cell "AAACCCATCGGTAACT" is annotated as a T cell type after analysis. Then the corresponding label of the cell "AAACCCATCGGTAACT" in the correspondence S is "T cell", that is, the cell number is "AAACCCATCGGTAACT", and the cell type is "T cell". "Cell type label, marker category" all refer to the same concept, namely "T cell".

[0045] (4) processing the genome sequence data representation of (3) and inputting it into the constructed fully connected network model, and training it to the preset conditions;

[0046] Scale the data in the bin block gene expression matrix Z of the bin10X matrix data to obtain a new matrix Z with m rows and n columns new , do the following transformation:

[0047]

[0048] Among them, min(Z) is the minimum value in the matrix Z, max(Z) is the maximum value in the matrix, min_val=1 and max_va=l000 are the manually specified scaling ranges;

[0049] For the new matrix Z new Perform data processing to obtain the input Z of the neural network in , the formula is as follows:

[0050]

[0051] Among them, Z new i,j is the matrix Z new The value of the element in the i-th row and j-th column, 0≤i<m, 0≤j<n;

[0052] Build a fully connected network model:

[0053] The fully connected network consists of an input layer, three fully connected layers, and an output layer; the input layer accepts an input of X nodes, the number of nodes in the three fully connected layers are 2000, 1000, and 25 respectively, and a ReLU nonlinear activation function and a dropout layer are used between each fully connected layer, where the dropout probability is set to 0.9. The output layer consists of a softmax nonlinear function, and the number of output nodes is the total number of cell types; Xavier uniform distribution is used for initialization, and the Adam Optimizer method is used to optimize the network weights. The initial learning rate is set to 0.0001. During the training process, the learning rate is gradually reduced with the number of iterations, and the learning rate of each training batch of data is attenuated by 0.95. Multivariate cross entropy CrossEntropyLoss is used as the loss function;

[0054] Training model parameters, bin block gene expression matrix Z of processed bin 10X matrix data in The features are input into the fully connected network, and after feature extraction and back propagation, the training model parameters are obtained after 100 training times.

[0055] (5) extracting the genome sequence data representation to be classified, and inputting the processed data into the trained fully connected network model to obtain the classification result;

[0056] Extract the gene expression feature information of the test set and input it into the trained model to obtain the probability distribution of the cell type of each cell, which is used to predict the cell type;

[0057] Taking peripheral blood mononuclear cell (PBMC) data as an example, the above method is used to obtain the accuracy of the cell type label prediction result of the cell data set, such as Figure 3 As shown, the training accuracy reaches 98.95% and the test accuracy reaches 98.36%.

[0058] The present invention also relates to a computer-readable storage medium in its application, on which is stored a cell classification program based on single-cell genome expression, which implements the above-mentioned cell classification method based on single-cell genome expression when executed by a processor.

[0059] The present invention also relates to a computer device in its application, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the above-mentioned cell classification method based on single-cell genome expression is implemented.

[0060] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0062] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0063] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0064] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0065] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A cell classification method based on single-cell genome expression, characterized in that: The method comprises the following steps: S1 constructs the re-divided BED file based on the genome sequence data group and converts it into a agreed format file; S2 builds a reference between genome sequence data sets to obtain a reference for the T2T genome; S3 is based on the genome sequence data set and the original real data, respectively interacting with the reference information of S2, using cellranger software and corresponding processing flow to obtain the genome sequence data representation and the corresponding cell type label; S4 processes the genome sequence data representation of S3 and inputs it into the constructed fully connected network model, and trains it to the preset conditions; S5 extracts the genome sequence data representation to be classified, and after processing, inputs it into the trained fully connected network model to obtain the classification result.

2. A cell classification method based on single-cell genome expression according to claim 1, characterized in that: In S1, the BED file based on the genome sequence data set includes the starting position, the ending position and the corresponding gene number of the genome sequence data set.

3. The cell classification method based on single-cell genome expression according to claim 1, characterized in that: In S2, reference data processing was completed based on the standard data format and CellRanger software.

4. The cell classification method based on single-cell genome expression according to claim 1, characterized in that: In S3, after processing, 10X matrix data of the genome sequence data set is generated, and the corresponding data expression matrix Z is obtained, where the rows of Z correspond to the block numbers of the genome sequence data set and the columns correspond to each cell.

5. The cell classification method based on single-cell genome expression according to claim 4, characterized in that: After processing S3, the cell type labels generated are marked, and the corresponding relationship S between each cell number and category is obtained, and the training set and test set are delineated from the data expression matrix Z.

6. A cell classification method based on single cell genome expression according to claim 4, characterized in that: In S4, processing the genome sequence data representation of S3 includes scaling the data in the data expression matrix Z to obtain a standardized new matrix Z with m rows and n columns. new , Among them, min(Z) is the minimum value in the matrix Z, max(Z) is the maximum value in the matrix, min_val and max_val are the lower limit and upper limit of the preset scaling range.

7. A cell classification method based on single cell genome expression according to claim 6, characterized in that: Processing the new matrix Z new , get the input Z of the fully connected network model in , Among them, Z newi,j is the matrix Z new The value of the element in the i-th row and j-th column, where i is an integer from 0 to m-1, j is an integer from 0 to n-1, and t is the column index.

8. The cell classification method based on single-cell genome expression according to claim 1, characterized in that: In S4, the fully connected network model includes a sequentially connected input layer, several groups of linear processing modules and an output layer; the linear processing modules are three groups, and any group of the linear processing modules includes a sequentially connected activation function layer, a dropout layer and a fully connected layer; the number of nodes in the linear layer of the linear processing module decreases successively.

9. A computer-readable storage medium, characterized in that: A cell classification program based on single-cell genome expression is stored thereon, and when the program is executed by the processor, the above-mentioned cell classification method based on single-cell genome expression is implemented.

10. A computer 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 above-mentioned cell classification method based on single-cell genome expression is implemented.

Images