Single cell sequencing data clustering method and system based on similarity constraint

Abstract

The invention discloses a single cell sequencing data clustering method and system based on similarity constraint, and the method comprises the steps: obtaining a Fastq sequencing file, carrying out the quantitative analysis and data preprocessing, and obtaining a preprocessed gene expression matrix; performing similarity priori knowledge and non-similarity priori knowledge extraction processing on the preprocessed gene expression matrix, and constructing a similarity matrix and a non-similarity matrix; constructing a joint non-negative matrix factorization model according to the similarity matrix and the non-similarity matrix, and performing dimension reduction processing to obtain a common coefficient matrix; and carrying out hierarchical clustering and visualization processing to obtain a single cell sequencing data clustering result. According to the method, similarity and non-similarity constraints of a plurality of gene expression matrixes are mined, and the feature extraction capability of the model is improved, so that a better single cell sequencing data clustering effect is realized. The single cell sequencing data clustering method and system based on similarity constraint can be widely applied to the technical field of single cell sequencing data clustering.

Description

Technical Field

[0001] The present invention relates to the technical field of single-cell sequencing data clustering, and in particular to a single-cell sequencing data clustering method and system based on similarity constraints. Background Art

[0002] Single-cell RNA sequencing (scRNA-seq) technology is a key means to reveal gene expression values ​​at the level of individual cells. Unlike batch sequencing technology based on the average value of sequenced cell populations, scRNA-seq technology can accurately display the whole genome gene expression of each single cell. In practical applications, scRNA-seq usually needs to undergo dimensionality reduction before clustering, and there are many quantification methods for gene expression data, such as Kallisto, Salmon, etc. The differences between different methods may lead to inconsistency in gene expression. Data based on a single method may introduce false positives, thereby affecting the accuracy of clustering effects. Therefore, some scholars have proposed using a joint non-negative matrix factorization model to integrate gene expression matrices obtained by different measurement methods. For example, some scholars have proposed applying this model to single-cell transcriptome data, and further using gene expression matrices from different quantitative methods as inputs to the joint non-negative matrix factorization model to improve the accuracy and reliability of clustering analysis. However, related technologies oversimplify the relationship model between cells, ignore the complex interactions and heterogeneity between cells, and result in the loss of correlation and shared structural information between cells during dimensionality reduction, resulting in poor clustering performance of the algorithm. Summary of the invention

[0003] In order to solve the above technical problems, the purpose of the present invention is to provide a single-cell sequencing data clustering method and system based on similarity constraints, which can achieve better single-cell sequencing data clustering effect by mining the joint similarity and dissimilarity constraints of multiple gene expression matrices and improving the model's ability to extract features.

[0004] The first technical solution adopted by the present invention is: a single-cell sequencing data clustering method based on similarity constraints, comprising the following steps:

[0005] Obtain Fastq sequencing files and perform quantitative analysis and data preprocessing to obtain the preprocessed gene expression matrix;

[0006] Extracting similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix;

[0007] A joint non-negative matrix factorization model is constructed based on the similarity matrix and the dissimilarity matrix and dimension reduction is performed to obtain a common coefficient matrix;

[0008] The common coefficient matrix was subjected to hierarchical clustering and visualization to obtain the clustering results of single-cell sequencing data.

[0009] Further, the step of obtaining the Fastq sequencing file and performing quantitative analysis and data preprocessing to obtain a preprocessed gene expression matrix specifically includes:

[0010] Obtain Fastq sequencing files based on scRNA-seq dataset;

[0011] The Fastq sequencing files are compared and quantified by the Salmon quantification method and the Kalisto quantification method, respectively, to obtain a gene expression matrix, wherein the gene expression matrix includes a first gene expression matrix and a second gene expression matrix;

[0012] The gene expression matrix was subjected to redundant gene removal and normalization processing to obtain the preprocessed gene expression matrix.

[0013] Furthermore, the step of extracting similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix specifically includes:

[0014] Performing Euclidean distance calculation on cells in the preprocessed gene expression matrix to construct a Euclidean distance matrix, wherein the dimension of the Euclidean distance matrix is ​​the number of cells;

[0015] The Euclidean distance matrix is ​​transformed, the first n smallest elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced with 0, so as to obtain the Euclidean distance matrix after the first transformation;

[0016] The first transformed Euclidean distance matrix is ​​mapped by a Gaussian kernel function to obtain a first Euclidean distance weight value;

[0017] The linear relationship between cells in the first transformed Euclidean distance matrix is ​​calculated by Pearson coefficient to obtain the first similarity coefficient;

[0018] The first Euclidean distance weight value is multiplied by the first similarity coefficient to obtain a similarity matrix, wherein the similarity matrix includes a first similarity matrix and a second similarity matrix;

[0019] The Euclidean distance matrix is ​​transformed, the largest first n elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced with 0, so as to obtain the second transformed Euclidean distance matrix;

[0020] The second transformed Euclidean distance matrix is ​​mapped by a 1-Gaussian kernel function to obtain a second Euclidean distance weight value;

[0021] The linear relationship of cells in the second transformed Euclidean distance matrix is ​​calculated by 1-Pearson coefficient to obtain the second similarity coefficient;

[0022] The second Euclidean distance weight value is multiplied by the second similarity coefficient to obtain a dissimilarity matrix, wherein the dissimilarity matrix includes a first dissimilarity matrix and a second dissimilarity matrix.

[0023] Furthermore, the step of constructing a joint non-negative matrix decomposition model based on the similarity matrix and the dissimilarity matrix and performing dimensionality reduction processing to obtain a common coefficient matrix specifically includes:

[0024] The first similarity matrix is ​​merged with the second similarity matrix to obtain a similarity constraint;

[0025] The first dissimilarity matrix is ​​merged with the second dissimilarity matrix to obtain a dissimilarity constraint;

[0026] Introducing sparse constraints, combining similarity constraints with non-similarity constraints, and constructing a joint non-negative matrix factorization model;

[0027] Combining the Lagrange multiplier method with the Karush-Kuhn-Tucher condition, the joint non-negative matrix factorization model is optimized and dimensionally reduced to obtain the common coefficient matrix.

[0028] Furthermore, the expression of the objective function of the joint non-negative matrix decomposition model is specifically as follows:

[0029]

[0030] In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the dissimilarity constraint, represents a square matrix consisting of the Euclidean distances between cells, λ1 represents the balance function of the decomposition of two gene expression matrices, and λ2, λ3 and λ4 represent regularization parameters.

[0031] Furthermore, the common coefficient matrix is ​​updated, and the expression of the update criterion of the common coefficient matrix is ​​specifically as follows:

[0032]

[0033] In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the non-similarity constraint, λ1 represents the balance function of the decomposition of the two gene expression matrices, λ3 and λ4 represent the regularization parameters, represents the transposed matrix of the basic matrix obtained by decomposing the first gene expression matrix through the joint non-negative matrix factorization model, represents the transposed matrix of the base matrix obtained by decomposing the second gene expression matrix through the joint non-negative matrix factorization model, H T Represents the transposed matrix of the shared coefficient matrix.

[0034] Furthermore, the step of performing hierarchical clustering and visualization processing on the common coefficient matrix to obtain the clustering result of the single-cell sequencing data specifically includes:

[0035] The common coefficient matrix was hierarchically clustered using a variance-based hierarchical clustering method to obtain preliminary clustering results of single-cell sequencing data;

[0036] The preliminary single-cell sequencing data clustering results were projected into two-dimensional space using the t-SNE visualization method to obtain the single-cell sequencing data clustering results.

[0037] The second technical solution adopted by the present invention is: a single-cell sequencing data clustering system based on similarity constraints, comprising:

[0038] The first module is used to obtain Fastq sequencing files and perform quantitative analysis and data preprocessing to obtain the preprocessed gene expression matrix;

[0039] The second module is used to extract similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix and construct a similarity matrix and a dissimilarity matrix;

[0040] The third module is used to construct a joint non-negative matrix factorization model based on the similarity matrix and the non-similarity matrix and perform dimensionality reduction processing to obtain a common coefficient matrix;

[0041] The fourth module is used to perform hierarchical clustering and visualization on the common coefficient matrix to obtain the clustering results of single-cell sequencing data.

[0042] The method and system of the present invention have the following beneficial effects: the present invention obtains a Fastq sequencing file and performs quantitative analysis and data preprocessing to obtain a preprocessed gene expression matrix, further extracts similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix, constructs a similarity matrix and a dissimilarity matrix, and promotes similar cells to be closer and dissimilar cells to be further away by mining the joint similarity and dissimilarity constraints of multiple gene expression matrices, thereby ensuring the accuracy of clustering and enhancing data representation, and then constructs a joint non-negative matrix decomposition model according to the similarity matrix and the dissimilarity matrix and performs dimensionality reduction processing to obtain a common coefficient matrix, and finally performs hierarchical clustering and visualization processing on the common coefficient matrix to obtain a clustering result of single-cell sequencing data, applies coefficient constraints to the coefficient matrix of the joint non-negative matrix decomposition model, avoids redundant features from affecting the output result, improves the ability of the model to extract features, and constructs a single-cell clustering visualization diagram, which can realize intuitive observation of the distribution and clustering pattern of cell clusters, and in the two-dimensional visualization diagram, cells of the same type can be clustered together and different cells can be separated from each other, thereby improving the clustering effect of single-cell sequencing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a flowchart of the steps of a single-cell sequencing data clustering method based on similarity constraints of the present invention;

[0044] Figure 2 It is a structural block diagram of a single-cell sequencing data clustering system based on similarity constraints of the present invention;

[0045] Figure 3 It is a schematic diagram of the visualization results of clustering of single-cell sequencing data provided by a specific embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only provided for the convenience of explanation and description, and the order between the steps is not limited in any way. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0047] First of all, it should be noted that dimensionality reduction is usually required before clustering. Dimensionality reduction plays a key role in this process, which helps to analyze the complex single-cell data structure and changes. Its main purpose is to convert high-dimensional data, which usually contains thousands of gene expression features, into a lower-dimensional representation while retaining the key information in the data as much as possible. This approach helps to visualize and interpret the data, discover potential cell subpopulations, and understand the state changes of cells under different conditions. Non-negative matrix factorization (NMF) is a commonly used unsupervised machine learning dimensionality reduction algorithm, which is particularly suitable for extracting meaningful features from non-negative data matrices. The core idea of ​​NMF is to decompose a non-negative data matrix into the product of two non-negative matrices, one of which represents the feature space and the other represents the sample weight. In the field of bioinformatics, NMF is widely used to process single-cell transcriptome data.

[0048] The traditional joint non-negative matrix factorization (JNMF) model currently applied to single-cell sequencing data ignores the potential prior knowledge between multi-omics data, oversimplifies the relationship model between cells, and ignores the complex interactions and heterogeneity between cells. In addition, none of the current JNMF improved models considers mining the prior knowledge of similarity and dissimilarity between cells, which may cluster similar cells into different categories or cluster dissimilar cells in the same category, resulting in inaccurate clustering results. To this end, an algorithm based on JNMF that combines the prior knowledge of cell similarity and dissimilarity is proposed, called the joint non-negative matrix factorization model with similarity and dissimilarity constraints (SDJNMF).

[0049] Based on this, refer to Figure 1 The present invention provides a single-cell sequencing data clustering method based on similarity constraints, the method comprising the following steps:

[0050] S100, obtaining Fastq sequencing files and performing quantitative analysis and data preprocessing to obtain a preprocessed gene expression matrix;

[0051] S110, obtain Fastq sequencing files based on scRNA-seq dataset;

[0052] In this embodiment, the Fastq sequencing files of the required scRNA-seq data set are downloaded from the GEO database. These Fastq files contain the base information of the DNA sequence and the quality score of each base in the sequencing process. These data can be used for various bioinformatics analyses, such as genome assembly, gene expression analysis, SNP (single nucleotide polymorphism) detection, etc.

[0053] S120, comparing and quantifying the Fastq sequencing file by using the Salmon quantitative method and the Kalisto quantitative method, respectively, to obtain a gene expression matrix, wherein the gene expression matrix includes a first gene expression matrix and a second gene expression matrix;

[0054] In the present embodiment, according to the species to which the downloaded single-cell sequencing data belongs, the corresponding reference transcriptome file is downloaded. In Salmon and Kalisto software, the reference transcriptome file is used to construct the index file required for Salmon. This step only needs to be performed once, and the same index file can be reused in subsequent analysis. Subsequently, Salmon and Kalisto software are run to specify the path of the reference transcriptome file, the path of the original sequencing data file and other related parameters, such as sample information files, technical repetitions, etc., and then Salmon and Kalisto software are run to compare and quantify, and the expression amount estimation results of the gene are generated. These results will be stored in text format, and for this reason, the text data needs to be converted into tabular data, so that subsequent cluster analysis, at this time, the gene expression matrix with the same but different gene expression of two cells has been obtained, i.e., the first gene expression matrix and the second gene expression matrix.

[0055] S130, performing redundant gene removal and normalization processing on the gene expression matrix to obtain a preprocessed gene expression matrix.

[0056] In this embodiment, with reference to the method of SC3, the redundant genes present in the gene expression matrix are removed to reduce the influence of redundant information on the clustering results, and then the processing method of log2(x+1) is adopted to convert the data. Finally, the data is normalized to reduce the load of the operation. In order to meet the basic requirements of non-negative matrix decomposition, it is necessary to ensure that each value of the gene expression matrix is ​​non-negative.

[0057] S200, extracting similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix;

[0058] Specifically, the Euclidean distance of the cells in the preprocessed gene expression matrix is ​​calculated to construct a Euclidean distance matrix, wherein the dimension of the Euclidean distance matrix is ​​the number of cells.

[0059] The Euclidean distance matrix is ​​transformed, the first n smallest elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced by 0 to obtain a first transformed Euclidean distance matrix; the first transformed Euclidean distance matrix is ​​mapped by a Gaussian kernel function to obtain a first Euclidean distance weight value; the linear relationship of the cells in the first transformed Euclidean distance matrix is ​​calculated by the Pearson coefficient to obtain a first similarity coefficient; the first Euclidean distance weight value is multiplied by the first similarity coefficient to obtain a similarity matrix, wherein the similarity matrix includes a first similarity matrix and a second similarity matrix.

[0060] The Euclidean distance matrix is ​​transformed, the largest first n elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced by 0 to obtain a second transformed Euclidean distance matrix; the second transformed Euclidean distance matrix is ​​mapped by a 1-Gaussian kernel function to obtain a second Euclidean distance weight value; the linear relationship of the cells in the second transformed Euclidean distance matrix is ​​calculated by a 1-Pearson coefficient to obtain a second similarity coefficient; the second Euclidean distance weight value is multiplied by the second similarity coefficient to obtain a dissimilarity matrix, wherein the dissimilarity matrix includes a first dissimilarity matrix and a second dissimilarity matrix.

[0061] In this embodiment, the Euclidean distance between each cell in each gene expression matrix is ​​calculated separately. The larger the Euclidean distance, the smaller the similarity between cells, so as to form a Euclidean distance matrix with a dimension of cell number. When calculating similarity, the present invention introduces the Pearson correlation coefficient and the Gaussian kernel function. Specifically, the three smallest values ​​in each row of the Euclidean distance matrix are used as variables of the Pearson correlation coefficient and the Gaussian kernel function for calculation, and the remaining values ​​are converted to 0 to obtain the Euclidean distance matrix after the first conversion, and then the Gaussian kernel function is multiplied with the Pearson correlation coefficient to obtain the product. The Gaussian kernel function maps the gene expression of the cell to a high-dimensional space, making it linearly separable, and can represent the weight of the Euclidean distance between cells. The Pearson coefficient can measure the linear relationship between two variables, so introducing the Pearson coefficient into the similarity matrix can better dig the correlation between cells. In summary, putting the Gaussian kernel function and the Pearson correlation coefficient together for weighting can better calculate the similarity between cells. So far, the similarity matrix will be constructed. Similarly, when calculating dissimilarity, the Pearson correlation coefficient and Gaussian kernel function are also introduced. Specifically, the three largest values ​​in each row of the Euclidean distance matrix are used as variables for calculating the Pearson correlation coefficient and the Gaussian kernel function, and the remaining values ​​are converted to 0 to obtain the second converted Euclidean distance matrix. Since the calculated Pearson correlation coefficient and Gaussian kernel function are both positive numbers less than 1, (1-Gaussian kernel function) and (1-Pearson correlation coefficient) are then multiplied to obtain the product to form a dissimilarity matrix.

[0062] S300, constructing a joint non-negative matrix decomposition model according to the similarity matrix and the non-similarity matrix and performing dimensionality reduction processing to obtain a common coefficient matrix;

[0063] S310, fusing the first similarity matrix with the second similarity matrix to obtain a similarity constraint;

[0064] S320, fusing the first dissimilarity matrix with the second dissimilarity matrix to obtain a dissimilarity constraint;

[0065] In this embodiment, the similarity and dissimilarity matrices of two gene expression matrices (i.e., the first gene expression matrix and the second gene expression matrix) are fused to construct similarity constraints and dissimilarity constraints. Specifically, if the values ​​of the same position in the two similarity matrices are not 0, the sum of the two values ​​is taken as the value of the joint similarity matrix at that position, otherwise the value of the joint similarity matrix at that position is 0. Similarly, if the values ​​of the same position in the two dissimilarity matrices are not 0, the sum of the two values ​​is taken as the value of the joint similarity matrix at that position, otherwise the value of the joint similarity matrix at that position is 0.

[0066] S330, introducing sparse constraints, combining similarity constraints and non-similarity constraints, and constructing a joint non-negative matrix factorization model;

[0067] In this embodiment, based on the above-mentioned construction of the joint similarity and dissimilarity matrix, the present invention imposes a sparse constraint on the common coefficient matrix of the joint non-negative matrix decomposition model to ensure that the cell features are extracted in a sparse environment, and obtains the objective function of the joint non-negative matrix decomposition model, which is specifically shown in the following formula:

[0068]

[0069] In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the dissimilarity constraint, represents a square matrix consisting of the Euclidean distances between cells, λ1 represents the balance function of the decomposition of two gene expression matrices, and λ2, λ3 and λ4 represent regularization parameters.

[0070] The balance function of the decomposition of the two gene expression matrices is set to λ1 = |number of genes in D1| / |number of genes in D2|.

[0071] S340. Combining the Lagrange multiplier method with the Karush-Kuhn-Tucher condition, the joint non-negative matrix factorization model is optimized and reduced in dimension to obtain a common coefficient matrix.

[0072] In this embodiment, the objective function is optimized by using the Lagrange multiplier method combined with the Karush-Kuhn-Tucher condition to obtain the update criteria of the common coefficient matrix and the basic matrix, which are specifically as follows:

[0073]

[0074]

[0075] In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the non-similarity constraint, λ1 represents the balance function of the decomposition of the two gene expression matrices, λ3 and λ4 represent the regularization parameters, represents the transposed matrix of the basic matrix obtained by decomposing the first gene expression matrix through the joint non-negative matrix factorization model, represents the transposed matrix of the base matrix obtained by decomposing the second gene expression matrix through the joint non-negative matrix factorization model, H T Represents the transposed matrix of the shared coefficient matrix.

[0076] S400, performing hierarchical clustering and visualization processing on the common coefficient matrix to obtain clustering results of single-cell sequencing data.

[0077] S410, performing hierarchical clustering on the common coefficient matrix by using a variance-based hierarchical clustering method to obtain preliminary clustering results of the single-cell sequencing data;

[0078] In this embodiment, hierarchical clustering is performed using the common coefficient matrix obtained by dimensionality reduction, and the hierarchical clustering method used in the embodiment of the present invention is the Ward method. The Ward method is a variance-based hierarchical clustering algorithm, and its idea is to select clusters at each step that can minimize the increase in total internal variance after the merger of two clusters. Specifically, Ward's method calculates the increase in variance between each pair of clusters, and selects two clusters that minimize the increase in internal variance after the merger to merge. The advantage of this method is that it tends to form clusters of similar size, so it can effectively capture the structure and pattern in the data. In the process of performing hierarchical clustering, the Ward method gradually merges samples or features until all samples or features are aggregated into one cluster. In this way, a hierarchical clustering structure is formed, which can be visualized by a dendrogram (tree diagram). In the dendrogram, branches at lower positions represent more similar samples or features, while branches at higher positions represent greater differences.

[0079] S420. Project the preliminary single-cell sequencing data clustering results into two-dimensional space through the t-SNE visualization method to obtain the single-cell sequencing data clustering results.

[0080] In this embodiment, after the scRNA-seq data is reduced in dimension, a common coefficient matrix representing the expression pattern of cells in the latent space will be obtained. This matrix captures the key structures in the data and projects the cells into a low-dimensional space. In order to better understand these reduced-dimensional data, the embodiment of the present invention uses the t-SNE algorithm for visualization. t-SNE can map high-dimensional data to two-dimensional or three-dimensional space, maintaining the local structural relationship between data points. In this reduced-dimensional space, the similarities between cells are more intuitively represented, so that potential cell subtypes or gene expression patterns can be more clearly observed.

[0081] like Figure 3 As shown in Figure 1, by drawing a scatter plot of the t-SNE visualization results and coloring them according to cell type or sample attributes, the distribution and clustering patterns of cell clusters can be intuitively observed. This visualization not only helps to discover structures and relationships in the data, but also provides important clues for subsequent biological interpretation and analysis.

[0082] Reference Figure 2 , a single-cell sequencing data clustering system based on similarity constraints, comprising:

[0083] The first module 201 is used to obtain Fastq sequencing files and perform quantitative analysis and data preprocessing to obtain a preprocessed gene expression matrix;

[0084] The second module 202 is used to extract similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix;

[0085] The third module 203 is used to construct a joint non-negative matrix factorization model according to the similarity matrix and the non-similarity matrix and perform dimensionality reduction processing to obtain a common coefficient matrix;

[0086] The fourth module 204 is used to perform hierarchical clustering and visualization processing on the common coefficient matrix to obtain the clustering results of the single-cell sequencing data.

[0087] The contents of the above method embodiments are all applicable to the present system embodiments. The functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0088] The above is a specific description of the preferred implementation of the present invention, but the invention is not limited to the embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.

Claims

1. A single-cell sequencing data clustering method based on similarity constraints, characterized in that: The following steps are involved: Obtain Fastq sequencing files and perform quantitative analysis and data preprocessing to obtain the preprocessed gene expression matrix; Extracting similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix; A joint non-negative matrix factorization model is constructed based on the similarity matrix and the dissimilarity matrix and dimension reduction is performed to obtain a common coefficient matrix; The common coefficient matrix was subjected to hierarchical clustering and visualization to obtain the clustering results of single-cell sequencing data.

2. According to claim 1, a single-cell sequencing data clustering method based on similarity constraints is characterized in that: The step of obtaining the Fastq sequencing file and performing quantitative analysis and data preprocessing to obtain a preprocessed gene expression matrix specifically includes: Obtain Fastq sequencing files based on scRNA-seq dataset; The Fastq sequencing files are compared and quantified by the Salmon quantification method and the Kalisto quantification method, respectively, to obtain a gene expression matrix, wherein the gene expression matrix includes a first gene expression matrix and a second gene expression matrix; The gene expression matrix was subjected to redundant gene removal and normalization processing to obtain the preprocessed gene expression matrix.

3. According to claim 2, a single-cell sequencing data clustering method based on similarity constraints is characterized in that: The step of extracting similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix to construct a similarity matrix and a dissimilarity matrix specifically includes: Performing Euclidean distance calculation on cells in the preprocessed gene expression matrix to construct a Euclidean distance matrix, wherein the dimension of the Euclidean distance matrix is ​​the number of cells; The Euclidean distance matrix is ​​transformed, the first n smallest elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced with 0, so as to obtain the Euclidean distance matrix after the first transformation; The first transformed Euclidean distance matrix is ​​mapped by a Gaussian kernel function to obtain a first Euclidean distance weight value; The linear relationship between cells in the first transformed Euclidean distance matrix is ​​calculated by Pearson coefficient to obtain the first similarity coefficient; The first Euclidean distance weight value is multiplied by the first similarity coefficient to obtain a similarity matrix, wherein the similarity matrix includes a first similarity matrix and a second similarity matrix; The Euclidean distance matrix is ​​transformed, the largest first n elements in each row of the Euclidean distance matrix are retained, and the remaining elements are replaced with 0, so as to obtain the second transformed Euclidean distance matrix; The second transformed Euclidean distance matrix is ​​mapped by a 1-Gaussian kernel function to obtain a second Euclidean distance weight value; The linear relationship of cells in the second transformed Euclidean distance matrix is ​​calculated by 1-Pearson coefficient to obtain the second similarity coefficient; The second Euclidean distance weight value is multiplied by the second similarity coefficient to obtain a dissimilarity matrix, wherein the dissimilarity matrix includes a first dissimilarity matrix and a second dissimilarity matrix.

4. A single-cell sequencing data clustering method based on similarity constraints according to claim 3, characterized in that: The step of constructing a joint non-negative matrix decomposition model based on the similarity matrix and the dissimilarity matrix and performing dimensionality reduction processing to obtain a common coefficient matrix specifically includes: The first similarity matrix is ​​merged with the second similarity matrix to obtain a similarity constraint; The first dissimilarity matrix is ​​merged with the second dissimilarity matrix to obtain a dissimilarity constraint; Introducing sparse constraints, combining similarity constraints with non-similarity constraints, and constructing a joint non-negative matrix factorization model; Combining the Lagrange multiplier method with the Karush-Kuhn-Tucher condition, the joint non-negative matrix factorization model is optimized and dimensionally reduced to obtain the common coefficient matrix.

5. A single-cell sequencing data clustering method based on similarity constraints according to claim 4, characterized in that: The expression of the objective function of the joint non-negative matrix decomposition model is specifically as follows: In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the dissimilarity constraint, represents a square matrix consisting of the Euclidean distances between cells, λ1 represents the balance function of the decomposition of two gene expression matrices, and λ2, λ3 and λ4 represent regularization parameters.

6. A single-cell sequencing data clustering method based on similarity constraints according to claim 5, characterized in that: It also includes updating the common coefficient matrix. The expression of the update criterion of the common coefficient matrix is ​​as follows: In the above formula, D1 represents the first gene expression matrix, D2 represents the second gene expression matrix, W1 and W2 represent the basic matrices obtained by decomposing D1 and D2 through the joint non-negative matrix decomposition model, H represents the shared coefficient matrix extracted from D1 and D2 through the joint non-negative matrix decomposition model, S represents the similarity constraint, V represents the non-similarity constraint, λ1 represents the balance function of the decomposition of the two gene expression matrices, λ3 and λ4 represent the regularization parameters, represents the transposed matrix of the basic matrix obtained by decomposing the first gene expression matrix through the joint non-negative matrix factorization model, represents the transposed matrix of the base matrix obtained by decomposing the second gene expression matrix through the joint non-negative matrix factorization model, H T Represents the transposed matrix of the shared coefficient matrix.

7. A single-cell sequencing data clustering method based on similarity constraints according to claim 6, characterized in that: The step of performing hierarchical clustering and visualization processing on the common coefficient matrix to obtain the clustering result of the single-cell sequencing data specifically includes: The common coefficient matrix was hierarchically clustered using a variance-based hierarchical clustering method to obtain preliminary clustering results of single-cell sequencing data; The preliminary single-cell sequencing data clustering results were projected into two-dimensional space using the t-SNE visualization method to obtain the single-cell sequencing data clustering results.

8. A single-cell sequencing data clustering system based on similarity constraints, characterized in that: Includes the following modules: The first module is used to obtain Fastq sequencing files and perform quantitative analysis and data preprocessing to obtain the preprocessed gene expression matrix; The second module is used to extract similarity prior knowledge and dissimilarity prior knowledge from the preprocessed gene expression matrix and construct a similarity matrix and a dissimilarity matrix; The third module is used to construct a joint non-negative matrix factorization model based on the similarity matrix and the non-similarity matrix and perform dimensionality reduction processing to obtain a common coefficient matrix; The fourth module is used to perform hierarchical clustering and visualization on the common coefficient matrix to obtain the clustering results of single-cell sequencing data.

Images