Cell type classification method and system based on iterative block matrix completion based on matrix decomposition

The problem of error completion in the completion of gene expression matrix in single-cell RNA sequencing is solved, and the accuracy and accuracy of cell type classification is improved.

CN114927167BActive Publication Date: 2024-11-15NANJING UNIV OF POSTS & TELECOMM
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210658796.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-08
Publication Date
2024-11-15
Estimated Expiration
2042-06-08

AI Technical Summary

Technical Problem

The prior art cannot effectively distinguish biological zero points from technical zero points in the completion process of gene expression matrix in single-cell RNA sequencing, resulting in error completion and affecting the accuracy of cell type classification.

Method used

The iterative block matrix completion method based on matrix decomposition is adopted, and the gene expression matrix is ​​decomposed and iteratively updated through the low-rank matrix decomposition algorithm to distinguish biological zero points and technical zero points, and the biological zero points are retained and technical zero points are replaced during the completion process. Form a complete gene expression matrix after completion.

Benefits of technology

It significantly improves the classification accuracy of the gene expression matrix, ensures the accuracy of the completed data, improves the accuracy of cell type recognition, and does not require a large amount of biological prior knowledge.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114927167B_ABST
    Figure CN114927167B_ABST
Patent Text Reader

Abstract

The present invention provides a cell type classification method and system based on iterative block matrix completion based on matrix decomposition, which obtains single-cell sequencing data, calculates the frequency of occurrence of each gene in a gene expression matrix and sorts it; divides it by columns; uses a low-rank matrix decomposition algorithm for each block matrix after division; compares the iterative matrix obtained by decomposition with the original block matrix, keeps non-zero points unchanged, retains biological zero points without completion, replaces and completes when the zero point is not a biological zero point, and obtains a block gene expression matrix after completion except for the biological zero point; splices and synthesizes a complete matrix; constructs a classification model, and obtains predicted cell types after inputting the complete gene expression matrix data after single-cell completion; the classification performance of the completed gene expression matrix of the present invention is significantly improved, which is conducive to the identification of cell types of sequencing data, and can obtain higher-precision classification results without the need for too much biological-related prior knowledge.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a cell type classification method and system based on iterative block matrix completion of matrix decomposition, belonging to the technical field of machine learning. Background Art

[0002] The mainstream cell identification in the field of single-cell RNA sequencing in current biology is to infer the cell type based on the highly expressed genes of the cell in the gene expression matrix. Currently, it mainly uses two methods: manual annotation and machine learning.

[0003] The first one is that there are many disadvantages in the manual annotation method, such as being limited by high technical requirements, low efficiency, and easy to mislabel. At present, there is still a shortage of scientific research talents in the field of single-cell RNA sequencing, and there is a lack of capable and experienced identification talents; under the influence of high-throughput sequencing technology: sequencing costs have become lower, and the number of cells sequenced at one time has increased (usually hundreds or thousands), and the number of cells that can be measured by manual annotation is small (usually dozens), which is difficult to meet the high efficiency requirements; wrong annotation affects the experimental results of subsequent downstream single-cell RNA sequencing.

[0004] The second method is to use machine learning methods to extract and learn features of the gene expression matrix, directly predict and classify cell types and cell subtypes, and greatly reduce the time and technical costs of manually annotating cell types. However, due to the limitations of single-cell RNA sequencing technology, measuring gene expression in a single cell requires amplification of a truly tiny amount of mRNA, which leads to a phenomenon called "dropout". In this phenomenon, some gene transcripts with low expression levels are not detected, so the expression value is zero. Under this "dropout" phenomenon, an excessive number of zero points will be displayed in the gene expression matrix. At present, the completion of the single-cell RNA gene expression matrix is ​​to complete the entire gene expression matrix, which will cause a large number of biological zero points to be incorrectly completed, thereby affecting the accuracy of subsequent classification results.

[0005] For example, Chinese invention patent CN 202010288371.X discloses a cell identification method and device based on machine learning. Based on the machine learning algorithm, the cell screening feature set is input into the pre-trained machine learning model for training to obtain a cell classification machine learning model; the cell data to be identified is identified and the cell category is output. In the above patent, there is also the problem of not distinguishing between biological zero points and technical zero points, and incorrectly completing biological zero points, which affects the classification results.

[0006] The above issues are those that should be considered and addressed during the cell type classification process. Summary of the invention

[0007] The purpose of the present invention is to provide a cell type classification method and system based on iterative block matrix completion based on matrix decomposition to solve the problem in the prior art that in the process of processing cell classification data, the zero points appearing in the entire gene expression matrix are completed, resulting in erroneous completion and the classification accuracy needs to be improved.

[0008] The technical solution of the present invention is:

[0009] A cell type classification method based on iterative block matrix completion based on matrix decomposition, comprising the following steps:

[0010] S1. Obtain single-cell sequencing data, perform preprocessing, obtain a gene expression matrix M of size m×n, calculate the frequency of occurrence of each gene in the gene expression matrix M, and sort them to obtain the sorted gene expression matrix P;

[0011] S2. Divide the sorted gene expression matrix P by columns into k block matrices V of the same size. The columns of the resulting small matrices are Where n is the number of genes, the number of rows in the small matrix is ​​m, and m is the number of cells;

[0012] S3, using a low-rank matrix decomposition algorithm for each partitioned block matrix V, decomposing it into the product of two matrices W and H;

[0013] S4, update matrices W and H, and calculate the error E between the product matrix of W and H and the block matrix V;

[0014] S5. When the error E is not less than the set threshold, return to step S4; when the error E is less than the set threshold, use the product matrix of W and H as the iterative matrix obtained by decomposition, and proceed to the next step S6;

[0015] S6, comparing the iterative matrix obtained by decomposition with the block matrix V, keeping the non-zero points in the block matrix V unchanged, confirming whether the zero point is a biological zero point, retaining the biological zero point without completion, and when the zero point is not a biological zero point, making it a technical zero point, replacing and completing the technical zero point, and obtaining a block gene expression matrix after completion except for the biological zero point;

[0016] S7. The completed block gene expression matrix obtained in step S6 except for the biological zero point is spliced ​​to form a complete matrix, that is, the completed complete gene expression matrix; a classification model is constructed using a supervised learning classification algorithm, and the predicted cell type is obtained after the input of the completed complete gene expression matrix data of a single cell.

[0017] Furthermore, in step S1, single cell sample data is obtained and preprocessed to obtain a gene expression matrix of size m×n. The frequency of occurrence of each gene in the gene expression matrix M is calculated and arranged to obtain a sorted gene expression matrix P, specifically,

[0018] S11, obtaining single-cell sample data, performing preprocessing, and obtaining a gene expression matrix of size m×n; wherein, the preprocessing uses the single-cell transcriptome data analysis and processing tool CellRanger to perform conversion and the synthetic minority oversampling technology SMOTE algorithm is used to oversample the uneven data;

[0019] S12. Calculate the frequency of occurrence of each gene in the gene expression matrix M. Genes with an occurrence frequency higher than the set threshold are considered to be effectively detected in single-cell RNA sequencing and are highly expressed genes. Arrange the gene expression matrix from large to small according to the frequency of the gene to obtain the sorted gene expression matrix P.

[0020] Furthermore, in step S12, the frequency of occurrence of each gene in the gene expression matrix M is calculated as the sum of the expression levels of the corresponding genes in all cells:

[0021]

[0022] Among them, M ij Represents the gene expression of the jth gene in the ith cell of the original gene expression matrix M, where M is the original gene expression matrix, which is a matrix with m rows and n columns.

[0023] Furthermore, in step S4, the matrices W and H are updated, specifically:

[0024] The iterative update formulas for W and H are as follows:

[0025]

[0026]

[0027] Among them, W ij is the value of the i-th row and j-th column of the matrix W, E is the error matrix, T is the transposed symbol, H ij is the value in the i-th row and j-th column of the matrix H.

[0028] Further, in step S6, it is confirmed whether the zero point is a biological zero point, and when the zero point is a technical zero point, replacement and completion are performed; the biological zero point is retained without completion, specifically,

[0029] When the number of non-zero points in the block matrix V is less than the set threshold t, the zero point is determined to be a biological zero point, and the biological zero point is retained without completion; when the number of non-zero points in the matrix is ​​not less than the set threshold t, the zero point is determined to be a technical zero point, and the updated product of W and H obtained in step S4 is used for completion, wherein the value range of the set threshold t is greater than 0 and not greater than the total number of cells m.

[0030] Further, in step S6, the block gene expression matrix after completion except for the biological zero point is obtained, specifically,

[0031]

[0032] Among them, V(i, j) is the value of the i-th row and j-th column of the original block matrix, WH(i, j) is the value of the i-th row and j-th column of the product matrix of the decomposed matrix W and the matrix H, and V′(i, j) is the value of the i-th row and j-th column of the block gene expression matrix.

[0033] Furthermore, in step S7, after the complete gene expression matrix data after single-cell completion is input, a predicted cell type is obtained. Specifically, the complete gene expression matrix data after single-cell completion is input into the constructed classification model. When the match is successful, a predicted cell type is obtained, and the predicted cell type is a known cell type; when the match is unsuccessful, it is a new unknown cell type or cell subtype.

[0034] A system for cell type classification using any of the above-mentioned iterative block matrix completion methods based on matrix decomposition, comprising a data preprocessing module, a block matrix completion module and a classification module,

[0035] The data preprocessing module preprocesses the acquired single-cell sequencing data to obtain a gene expression matrix M of size m×n, calculates the frequency of occurrence of each gene in the gene expression matrix M, and sorts them to obtain the sorted gene expression matrix P;

[0036] The block matrix completion module divides the sorted gene expression matrix P by columns into k block matrices V of the same size, and uses a low-rank matrix decomposition algorithm to decompose each block matrix V into the product of two matrices W and H. The error E between the product matrix of W and H and the block matrix V is calculated, and W and H are continuously updated iteratively until the error E is less than a set threshold. The iterative matrix sum obtained by decomposition is compared with the block matrix V, and the non-zero points in the block matrix V remain unchanged. It is confirmed whether the zero point is a biological zero point, and the biological zero point is retained without completion. If the zero point is not a biological zero point, it is a technical zero point, and the technical zero point is replaced and completed to obtain a block gene expression matrix after completion except for the biological zero point; the obtained block gene expression matrix after completion except for the biological zero point is spliced ​​to synthesize a complete matrix, which is the completed complete gene expression matrix;

[0037] The classification module uses a supervised learning classification algorithm to build a classification model to predict the cell type of the complete gene expression matrix data after input single-cell completion.

[0038] The beneficial effects of the present invention are as follows: the cell type classification method and system based on iterative block matrix completion based on matrix decomposition, in view of the different nature of the generation of biological zero points and technical zero points in the gene expression matrix, distinguishes biological zero points from technical zero points in the completion process, completes the technical zero points, and retains the biological zero points, which can effectively ensure the accuracy of the completion, and can significantly improve the classification accuracy of the gene expression matrix after completion, which is beneficial to improving the recognition accuracy of the cell type of sequencing data, and can obtain higher-precision classification results without the need to rely on too much biological-related prior knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flowchart of a cell type classification method based on iterative block matrix completion of matrix decomposition according to an embodiment of the present invention.

[0040] Figure 2 It is a schematic diagram for explaining the low-rank matrix decomposition after the sorted gene expression matrix P is split by columns in the embodiment. DETAILED DESCRIPTION

[0041] The preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0042] Example

[0043] A cell type classification method based on iterative block matrix completion based on matrix decomposition, such as Figure 1 , including the following steps,

[0044] S1. Obtain single-cell sequencing data, perform preprocessing, obtain a gene expression matrix M of size m×n, calculate the frequency of occurrence of each gene in the gene expression matrix M, and sort them to obtain the sorted gene expression matrix P;

[0045] S11, obtaining single-cell sample data, performing preprocessing, and obtaining a gene expression matrix of size m×n; wherein, the preprocessing uses the single-cell transcriptome data analysis and processing tool CellRanger to perform conversion and the synthetic minority oversampling technology SMOTE algorithm is used to oversample the uneven data;

[0046] S12. Calculate the frequency of each gene in the gene expression matrix M. Genes with a frequency higher than the set threshold are considered to be effectively detected in single-cell RNA sequencing and are highly expressed genes. Arrange the gene expression matrix from large to small according to the frequency of the gene to obtain the sorted gene expression matrix P. This facilitates the subsequent matrix block operation and ensures that the gene frequencies of the same block of the matrix are similar after block division.

[0047] In step S12, the essence of the expression amount of genes in the gene expression matrix is ​​determined by the number of occurrences of different genes in each cell in the sequencing. Therefore, the frequency of occurrence of each gene in the gene expression matrix M is calculated by adding the expression amounts of the corresponding genes of all cells:

[0048]

[0049] Among them, M ij Represents the gene expression of the jth gene in the ith cell of the original gene expression matrix M, where M is the original gene expression matrix, which is a matrix with m rows and n columns.

[0050] G obtained by formula (1) j Used to approximately determine the frequency of genes, i.e. G j The larger the value, the greater the proportion of the gene in this sequencing, and the higher the importance of the gene in the sequenced cells. When the gene in this column is missing in a cell, it may be a technical zero point caused by insufficient sequencing technology, and it should be supplemented. j A smaller value indicates that the gene is expressed at a very low level in the original gene expression matrix. j Then, the original gene expression matrix needs to be sorted by gene column, that is, according to the G calculated by gene j The gene expression matrix V is rearranged from large to small, so that genes with more frequent expression appear at the front in the sorted matrix, while genes with less expression or even almost no expression appear at the back.

[0051] S2. Divide the sorted gene expression matrix P by columns into k block matrices V of the same size. The columns of the resulting small matrices are Where n is the number of genes, the number of rows in the small matrix is ​​m, and m is the number of cells, such as Figure 2 ;

[0052] After the gene expression matrix is ​​sorted from high to low according to gene frequency, in order to complete the zero points in high-frequency and highly expressed genes and retain the zero points in low-frequency and low-expressed genes, it is necessary to divide the gene expression matrix that has been sorted from high to low according to gene frequency into blocks. The diagram of matrix block completion is as follows: Figure 2 As shown in the figure, the original matrix of size m×n is divided into k small matrices of the same size by column, and then each small matrix is ​​completed using an iterative matrix decomposition algorithm.

[0053] One of the difficulties in matrix partitioning is to determine the number of blocks to be partitioned into using parameter k. When parameter k is too large, the small matrix after partitioning will be too small, and the effect of the matrix decomposition completion algorithm will not be ideal. If parameter k is too small, the granularity after partitioning will be too large, and the meaning of partitioning using differences in gene frequency will be lost. Therefore, choosing a suitable parameter k, with a range of [0,n], will have a huge impact on the performance of the matrix completion model. Since the total number of genes in the gene expression matrix is ​​n, we should test when k is 1, When n (n is the number of columns of the gene expression matrix, i.e. the number of all genes), the cell classification performance is achieved by partitioning the gene expression matrix and then completing it.

[0054] S3. Use a low-rank matrix decomposition algorithm to decompose each partitioned matrix V into the product of two matrices, W and H, such as Figure 2 .

[0055] S4, update matrices W and H, and calculate the error E between the product matrix of W and H and the block matrix V;

[0056] In step S4, the matrices W and H are updated, specifically:

[0057] The iterative update formulas for W and H are as follows:

[0058]

[0059]

[0060] Among them, W ij is the value of the i-th row and j-th column of the matrix W, E is the error matrix, T is the transposed symbol, H ij is the value in the i-th row and j-th column of the matrix H.

[0061] The core idea of ​​matrix completion using matrix decomposition is to decompose the original block matrix V into the product of two matrices W and H, calculate the product of the two matrices W and H and the error E of the original block matrix V, and then iteratively update W and H using the error E until the error E is less than the set threshold. The product matrix of W and H after the final iteration can be used to complete the elements at the corresponding position in the matrix.

[0062] S5. When the error E is not less than the set threshold, return to step S4; when the error E is less than the set threshold, use the product matrix of W and H as the iterative matrix obtained by decomposition, and proceed to the next step S6;

[0063] S6, comparing the iterative matrix obtained by decomposition with the block matrix V, keeping the non-zero points in the block matrix V unchanged, confirming whether the zero point is a biological zero point, retaining the biological zero point without completion, and when the zero point is not a biological zero point, making it a technical zero point, replacing and completing the technical zero point, and obtaining a block gene expression matrix after completion except for the biological zero point;

[0064] In step S6, it is confirmed whether the zero point is the biological zero point, and if the zero point is the technical zero point, it is replaced and completed; the biological zero point is retained without completion, specifically,

[0065] When the number of non-zero points in the block matrix V is less than the set threshold t, the zero point is determined to be a biological zero point, and the biological zero point is retained without completion; when the number of non-zero points in the matrix is ​​not less than the set threshold t, the zero point is determined to be a technical zero point, and the updated product of W and H obtained in step S4 is used for completion, wherein the value range of the set threshold t is greater than 0 and not greater than the total number of cells m.

[0066] If the element at the corresponding position in the original block matrix V is not zero, the value on the original matrix can be retained; for biological zero points, the biological zero points are retained without filling; for technical zero points, the corresponding position in the original block matrix V is zero, and it is filled by the product of W and H. The specific formula is as follows:

[0067]

[0068] Among them, V(i, j) is the value of the i-th row and j-th column of the original block matrix, WH(i, j) is the value of the i-th row and j-th column of the product matrix of the decomposed matrix W and the matrix H, and V′(i, j) is the value of the i-th row and j-th column of the block gene expression matrix.

[0069] S7. The completed block gene expression matrix obtained in step S6 except for the biological zero point is spliced ​​to form a complete matrix, that is, the completed complete gene expression matrix; a classification model is constructed using a supervised learning classification algorithm, and the predicted cell type is obtained after the input of the completed complete gene expression matrix data of a single cell.

[0070] In step S7, after the input single-cell completed complete gene expression matrix data, the predicted cell type is obtained. Specifically, the complete gene expression matrix data after single-cell completion is input into the constructed classification model. When the match is successful, the predicted cell type is obtained, and the predicted cell type is a known cell type; when the match is unsuccessful, it is a new unknown cell type or cell subtype. This method also helps researchers discover new cell types and cell subtypes, providing a realistic method for the discovery of new cell types.

[0071] This cell type classification method based on iterative block matrix completion of matrix decomposition distinguishes biological zero points from technical zero points according to the different causes of their generation, and performs corresponding retention and completion processing. Specifically, the biological zero points that were originally zero points are retained, and the technical zero points are replaced and completed, thereby improving the accuracy of completion, effectively avoiding the problem of incorrect completion of biological zero points, and providing accurate data for subsequent classification processing.

[0072] There are many biological zeros and technical zeros in the obtained gene expression matrix. The biological zero means that the gene is not expressed in the cell. This biological zero is not caused by the insufficiency of sequencing technology, but should be zero. At this time, these biological zeros should not be completed, but should be retained in the final matrix. The technical zero means that the expressed gene is not detected due to the "loss" phenomenon. The reason for the technical zero is due to sequencing technology.

[0073] In this cell type classification method based on iterative block matrix completion of matrix decomposition, the principle of distinguishing biological zero points from technical zero points is: since the expression level of a gene is actually determined by the frequency of occurrence of this gene, the sequenced cells are generally from the same tissue or organ. Therefore, when a gene is missing in a cell, and other cells have high expression of this gene, it is very likely that this is a technical zero point caused by sequencing technology. On the contrary, if a gene is missing in a cell, and the expression level of this gene is low or not expressed at all in other cells, then the zero point at this time is very likely to be a biological zero point.

[0074] This cell type classification method based on iterative block matrix completion of matrix decomposition analyzes the difference between the above biological zero points and technical zero points, determines whether the zero point is a technical zero point or a biological zero point, and completes the technical zero point; there is no need to complete the biological zero point, and the zero point can be retained in the completed matrix.

[0075] Since steps S1-S4 have sorted and divided the gene expression matrix by gene frequency, genes with similar gene frequencies can be divided into one matrix as much as possible. Therefore, when performing the biological zero point retention operation, it is only necessary to consider the gene frequency of each small matrix after the block. The genes with lower frequencies in the small matrix after the block are retained at zero points, without using the matrix decomposition algorithm for completion. Let the threshold be t. When the number of genes expressed in the entire matrix is ​​less than t, it is considered that the zero point in the small matrix after the block is most likely a biological zero point and does not need to be completed. Since the size of the entire small matrix is (m,n are the number of rows and columns of the gene expression matrix), and the test t is The performance of the matrix classification algorithm at time 0. The best performance is retained as the zero point.

[0076] S6. The completed block gene expression matrix obtained in step S5 except for the biological zero point is spliced ​​into a complete matrix, that is, the completed complete gene expression matrix; a classification model is constructed using a supervised learning classification algorithm to predict the cell type of the input single-cell gene expression matrix data.

[0077] After steps S1-S5, the completed small matrix is ​​obtained, and the large matrix obtained by splicing is the final result of matrix completion. Then, various supervised learning methods such as support vector machine (SVM), random forest (RF) and logistic regression (LR) classification algorithms are used to train the classifier, and the classifier with superior performance is selected as the single cell type classifier to obtain the classification result of the cell type.

[0078] A system for cell type classification using any of the above-mentioned iterative block matrix completion methods based on matrix decomposition, comprising a data preprocessing module, a block matrix completion module and a classification module,

[0079] The data preprocessing module preprocesses the acquired single-cell sequencing data to obtain a gene expression matrix M of size m×n, calculates the frequency of occurrence of each gene in the gene expression matrix M, and sorts them to obtain the sorted gene expression matrix P;

[0080] The block matrix completion module divides the sorted gene expression matrix P by columns into k block matrices V of the same size, and uses a low-rank matrix decomposition algorithm to decompose each block matrix V into the product of two matrices W and H. The error E between the product matrix of W and H and the block matrix V is calculated, and W and H are continuously updated iteratively until the error E is less than a set threshold. The iterative matrix sum obtained by decomposition is compared with the block matrix V, and the non-zero points in the block matrix V remain unchanged. It is confirmed whether the zero point is a biological zero point, and the biological zero point is retained without completion. If the zero point is not a biological zero point, it is a technical zero point, and the technical zero point is replaced and completed to obtain a block gene expression matrix after completion except for the biological zero point; the obtained block gene expression matrix after completion except for the biological zero point is spliced ​​to synthesize a complete matrix, which is the completed complete gene expression matrix;

[0081] The classification module uses a supervised learning classification algorithm to build a classification model to predict the cell type of the complete gene expression matrix data after input single-cell completion.

[0082] This cell type classification method and system based on iterative block matrix completion based on matrix decomposition, in view of the different nature of the generation of biological zero points and technical zero points in the gene expression matrix, distinguishes biological zero points from technical zero points during the completion process, completes the technical zero points, and retains the biological zero points, which can effectively ensure the accuracy of the completion and can significantly improve the classification accuracy of the completed gene expression matrix, which is beneficial to improving the recognition accuracy of cell types in sequencing data, and can obtain higher-precision classification results without the need for too much biological-related prior knowledge.

[0083] This cell type classification method and system based on iterative block matrix completion of matrix decomposition can effectively complete and retain the excessive zero points in the gene expression matrix under the "loss" phenomenon, and can ensure that the zero points that should be zero points are retained, and the zero points that should not be zero points are completed, so as to obtain a more accurate gene expression matrix, thereby improving the classification accuracy of cell types. This method and system can effectively solve the problem that most of the existing single-cell RNA gene expression matrix completion does not distinguish between technical zero points and biological zero points, and completes the entire gene expression matrix, resulting in a large number of biological zero points being incorrectly completed, affecting the accuracy of the classification results.

[0084] The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiment. Any equivalent modifications or changes made by ordinary technicians in this field based on the contents disclosed by the present invention should be included in the protection scope recorded in the claims.

Claims

1. A cell type classification method based on iterative block matrix completion based on matrix decomposition, characterized by: The following steps are included: S1. Obtain single-cell sequencing data, perform preprocessing, obtain a gene expression matrix M of size m×n, calculate the frequency of occurrence of each gene in the gene expression matrix M, and sort them to obtain the sorted gene expression matrix P; S2. Divide the sorted gene expression matrix P by columns into k block matrices V of the same size. The columns of the resulting small matrices are Where n is the number of genes, the number of rows in the small matrix is ​​m, and m is the number of cells; S3, using a low-rank matrix decomposition algorithm for each partitioned block matrix V, decomposing it into the product of two matrices W and H; S4, update matrices W and H, and calculate the error E between the product matrix of W and H and the block matrix V; S5. When the error E is not less than the set threshold, return to step S4; when the error E is less than the set threshold, use the product matrix of W and H as the iterative matrix obtained by decomposition, and proceed to the next step S6; S6, comparing the iterative matrix obtained by decomposition with the block matrix V, keeping the non-zero points in the block matrix V unchanged, confirming whether the zero point is a biological zero point, retaining the biological zero point without completion, and when the zero point is not a biological zero point, making it a technical zero point, replacing and completing the technical zero point, and obtaining a block gene expression matrix after completion except for the biological zero point; S7. The completed block gene expression matrix obtained in step S6 except for the biological zero point is spliced ​​to form a complete matrix, that is, the completed complete gene expression matrix; a classification model is constructed using a supervised learning classification algorithm, and the predicted cell type is obtained after the input of the completed complete gene expression matrix data of a single cell.

2. The cell type classification method based on iterative block matrix completion based on matrix decomposition as claimed in claim 1, characterized in that: In step S1, single cell sample data is obtained and preprocessed to obtain a gene expression matrix of size m×n. The frequency of occurrence of each gene in the gene expression matrix M is calculated and arranged to obtain a sorted gene expression matrix P, which is specifically, S11, obtaining single-cell sample data, performing preprocessing, and obtaining a gene expression matrix of size m×n; wherein, the preprocessing uses the single-cell transcriptome data analysis and processing tool CellRanger to perform conversion and the synthetic minority oversampling technology SMOTE algorithm is used to oversample the uneven data; S12. Calculate the frequency of occurrence of each gene in the gene expression matrix M. Genes with an occurrence frequency higher than the set threshold are considered to be effectively detected in the single-cell RNA sequencing. For highly expressed genes, the gene expression matrix is ​​arranged from large to small according to the frequency of the gene to obtain the sorted gene expression matrix P.

3. The cell type classification method based on iterative block matrix completion based on matrix decomposition as claimed in claim 2, characterized in that: In step S12, the frequency of occurrence of each gene in the gene expression matrix M is calculated as the sum of the expression levels of the corresponding genes in all cells: Among them, M ij Represents the gene expression of the jth gene in the ith cell of the original gene expression matrix M, where M is the original gene expression matrix, which is a matrix with m rows and n columns.

4. The cell type classification method based on iterative block matrix completion based on matrix decomposition according to any one of claims 1 to 3, characterized in that: In step S4, W and H are updated, specifically: The iterative update formulas for W and H are as follows: Among them, W ij is the value of the i-th row and j-th column of the matrix W, E is the error matrix, T is the transposed symbol, H ij is the value in the i-th row and j-th column of the matrix H.

5. The cell type classification method based on iterative block matrix completion based on matrix decomposition according to any one of claims 1 to 3, characterized in that: In step S6, it is confirmed whether the zero point is the biological zero point, and if the zero point is the technical zero point, it is replaced and completed; the biological zero point is retained without completion, specifically, When the number of non-zero points in the block matrix V is less than the set threshold t, the zero point is determined to be a biological zero point, and the biological zero point is retained without completion; when the number of non-zero points in the matrix is ​​not less than the set threshold t, the zero point is determined to be a technical zero point, and the updated product of W and H obtained in step S4 is used for completion, wherein the value range of the set threshold t is greater than 0 and not greater than the total number of cells m.

6. The cell type classification method based on iterative block matrix completion based on matrix decomposition according to any one of claims 1 to 3, characterized in that: In step S6, the completed block gene expression matrix except for the biological zero point is obtained, specifically, Among them, V(i, j) is the value of the i-th row and j-th column of the original block matrix, WH(i, j) is the value of the i-th row and j-th column of the product matrix of the decomposed matrix W and the matrix H, and V′(i, j) is the value of the i-th row and j-th column of the block gene expression matrix.

7. The cell type classification method based on iterative block matrix completion based on matrix decomposition according to any one of claims 1 to 3, characterized in that: In step S7, after the complete gene expression matrix data after single-cell completion is input, the predicted cell type is obtained. Specifically, the complete gene expression matrix data after single-cell completion is input into the constructed classification model. When the match is successful, the predicted cell type is obtained, and the predicted cell type is a known cell type; when the match is unsuccessful, it is a new unknown cell type or cell subtype.

8. A system for cell type classification using the iterative block matrix completion method based on matrix decomposition according to any one of claims 1 to 7, characterized in that: It includes data preprocessing module, block matrix completion module and classification module. The data preprocessing module preprocesses the acquired single-cell sequencing data to obtain a gene expression matrix M of size m×n, calculates the frequency of occurrence of each gene in the gene expression matrix M, and sorts them to obtain the sorted gene expression matrix P; The block matrix completion module divides the sorted gene expression matrix P by columns into k block matrices V of the same size, and uses a low-rank matrix decomposition algorithm to decompose each block matrix V into the product of two matrices W and H. The error E between the product matrix of W and H and the block matrix V is calculated, and W and H are continuously updated iteratively until the error E is less than a set threshold. The iterative matrix sum obtained by decomposition is compared with the block matrix V, and the non-zero points in the block matrix V remain unchanged. It is confirmed whether the zero point is a biological zero point, and the biological zero point is retained without completion. If the zero point is not a biological zero point, it is a technical zero point, and the technical zero point is replaced and completed to obtain a block gene expression matrix after completion except for the biological zero point; the obtained block gene expression matrix after completion except for the biological zero point is spliced ​​to synthesize a complete matrix, which is the completed complete gene expression matrix; The classification module uses a supervised learning classification algorithm to build a classification model to predict the cell type of the complete gene expression matrix data after input single-cell completion.