Method and computer system for analyzing single-cell transcriptome data pseudo-time trajectories

Abstract

The application discloses a single-cell transcriptome data pseudo-time trajectory analysis method and a computer system, which comprises the following steps: 1) calculating a gene explicit comparison advantage matrix; 2) obtaining a gene similarity matrix by similarity and constructing a gene network; 3) taking an initial node in the gene network, starting random walking from the initial node, recording each gene walked through to form a gene text composed of gene sequences; 4) converting the gene text into a gene word vector; 5) adding all single-cell expressed gene vectors with expression as weight to form a sum vector as a word vector representation of the single cell in the gene space; and 6) visualizing all cell vector representations to obtain an embryo cell development pseudo-time trajectory result. The application provides an analysis basis for identifying different rare cell subtypes in tissues and variant genes of different cell subtypes and has a wide and important application prospect in the fields of tumors, developmental biology and life science.

Description

TECHNICAL FIELD

[0001] The present application relates to a method for analyzing single-cell transcriptome data pseudo-time trajectory and a computer system. BACKGROUND

[0002] Single-cell transcriptome sequencing technology was invented in 2009, which is a rapidly developing frontier technology in life sciences in the past decade. Single-cell transcriptome sequencing refers to the technology of high-throughput sequencing of mRNA reverse transcription and amplification of single cells, which reveals the cell gene arrangement structure and gene expression state at the single cell level, providing a unique perspective for life science research. This technology allows simultaneous transcriptome analysis of thousands of single cells, providing cell type identification at the single-cell transcriptome resolution level and enabling detailed analysis of cell heterogeneity and complexity in tissues at the transcriptome level, providing a powerful new method for scientists to study single-cell behavior and mechanisms. Single-cell sequencing technology plays an increasingly important role in the fields of tumor, developmental biology, microbiology, neuroscience, and has become a highly potential research field in life sciences.

[0003] According to single-cell transcriptome sequencing data, we can analyze cell change trajectory, discuss the single or multiple developmental trajectory differentiation and fate change of progenitor cells along a certain process due to gene mutation and other factors during cell development and evolution. When the cell changes from one gene sequence state to another sequence state, the size of the gene variation can be measured by pseudo-time index, so that the progress and change of single cell in the cell differentiation process can be measured by pseudo-time.

[0004] In the past few years, various different pseudo-time trajectory analysis methods have been developed, such as PAGA, DTFLOW, VIA, and Palantir. PAGA is a partition-based graph abstraction method that constructs an interpretable graph mapping of manifold data based on the estimation of manifold partition connectivity, which preserves the global topology of the data, allows analysis of data at different resolutions, and improves the computational efficiency of data analysis. PAGA is an easily interpretable topological data analysis method, in order to analyze single-cell gene dynamics, PAGA extends the distance measure based on random walk to consider the case of disconnected graphs, by tracking high-confidence paths in the abstract graph and sorting cells in the path from the progenitor cell, to analyze gene change size and construct pseudo-time trajectory.

[0005] The DTFLOW method first calculates the k nearest neighbors of each cell based on the Euclidean distance to obtain a nearest neighbor graph. Then, a Gaussian kernel function matrix is constructed to convert the Euclidean distance between cells into Gaussian kernel weights to represent the similarity between cells. By using the restart random walk algorithm, each data point is converted into a discrete distribution, and a diffusion matrix is obtained. Then, a Bhattacharyya kernel matrix is constructed based on the Bhattacharyya coefficient between different distributions, and a low-dimensional dataset is used to visualize the underlying structure of the original high-dimensional single-cell dataset. The pseudo-time trajectory calculation is performed based on the matrix corresponding to the root cell.

[0006] The VIA method is an innovative method that adds a lazy-teleporting behavior to the previous random walk model. The lazy-teleporting random walk in the VIA model allows for a predefined pause state (staying at the original node / state) and a jump state (jumping to any other node / state). This method considers the lazy behavior of random walking, i.e., each node has a (1-) probability of staying at the original position, and then performs scalable pseudo-time calculation on the dataset by considering random walking on the cluster graph.

[0007] The Palantir method uses an adjacency graph and pseudo-time to construct a Markov chain, which considers cell differentiation as a random process where a cell reaches one or more terminal states through a series of steps in the manifold. Pseudo-time provides directionality to orient the edges in the adjacency graph in a way that is consistent with the order. For each directed edge, the transition probability of reaching a neighboring cell in one step is calculated. The probability of reaching a distant cell is calculated through multiple steps, and if there are multiple paths connecting them, the probability will be high, i.e., the observed intermediate cell state density is high. In this way, the manifold graph structure gives the pseudo-time trajectory over longer distances through each step of random walking.

[0008] Chinese patent CN112768001A discloses a single-cell trajectory inference method based on manifold learning and principal curve. The invention mainly includes: (1) collecting known single-cell RNA seq related data; (2) performing feature extraction and selecting variable genes as features; (3) performing data dimension reduction to alleviate the curse of dimensionality; (4) defining a principal curve locally and proposing an initialization process; (5) applying a mean shift algorithm with piecewise subspace constraints to establish a final principal curve model; (6) performing pseudo-time analysis on the model and establishing a straight tree topology graph; (7) detecting differential expression genes on the straight tree topology graph. The invention provides a single-cell trajectory inference method to reconstruct and capture the lineage relationship within an organism, detect differential expression genes, and obtain the influence based on cell differentiation.

[0009] Chinese patent CN113808670A discloses a method for predicting cell differentiation from single-cell transcriptome data, including data input, data preprocessing, data dimensionality reduction / visualization, differentiation trajectory inference, false time prediction, and output of predicted cell differentiation trajectory. The patent ensures that any topological structure can be constructed by finding transition cells in cell clusters and connecting corresponding cell clusters.

[0010] Now these methods can be used for single-cell transcriptome data for single-cell pseudo-time trajectory analysis, but these methods mainly use some traditional manifold learning-based methods for data dimensionality reduction and subsequent analysis, and the analysis speed is not fast enough, and the accuracy of the algorithm needs to be further improved. In single-cell transcriptome sequencing analysis, there is no method to define the distance between cells based on the revealed comparative advantage index, and there is no natural language text processing technology to analyze the pseudo-time trajectory of cells. SUMMARY

[0011] In view of the deficiencies in the prior art, the purpose of the present application is to provide a method for analyzing the pseudo-time trajectory of single-cell transcriptome data and a computer system.

[0012] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0013] A method for analyzing the pseudo-time trajectory of single-cell transcriptome data, comprising the following steps:

[0014] 1) Calculate the gene explicit comparative advantage matrix using the single-cell transcriptome sequencing matrix;

[0015] 2) After obtaining the gene explicit comparative advantage matrix, calculate the similarity between genes and obtain the gene similarity matrix, and construct a gene network according to the similarity;

[0016] 3) Take an initial node in the gene network and start random walking from the initial node, record each gene walked through to form a gene text composed of gene sequences;

[0017] 4) Convert the gene text into a gene word vector using a word embedding algorithm;

[0018] 5) Add the gene vectors expressed by the cells with the expression amount as the weight to form a sum vector as the word vector representation of the cell in the gene space;

[0019] 6) Visualize the vector representation of all cells to obtain the pseudo-time trajectory result of embryonic cell development.

[0020] In step 1), the single-cell transcriptome data is normalized:

[0021] X RNAcg =log p(1+X) cg )

[0022] Among them, X RNAcg For the standardized transcript profile, X cg This represents the expression level of the g-th gene in the c-th cell of the transcription profile.

[0023] In step 1), the standardized single-cell transcriptome data is processed using... The explicit comparative advantage of each gene expressed in each cell was calculated, and the single-cell transcriptome sequencing matrix was converted into a gene explicit comparative advantage matrix, where x cg RCA represents the expression level of the c-th cell on the g-th gene. cg The explicit comparative advantage of cell c on gene g.

[0024] In step 2), through

[0025]

[0026] Calculate the similarity between genes, where P represents the conditional probability, φ ij This indicates the similarity between gene i and gene j.

[0027] In step 2), each gene is considered as a node in the network. A set similarity threshold is set. When the similarity between two genes is greater than the set similarity threshold, it is considered that there is an edge connecting the two genes. Based on this, the gene network is constructed.

[0028] In step 2), the Pearson correlation coefficient similarity measure or cosine similarity and Jaccard coefficient are used to construct the gene network.

[0029] In step 3), the initial node in the gene network can be selected by traversing the network or by randomly selecting nodes.

[0030] In step 4), the word embedding algorithm uses the word embedding methods Word2vec, GloVe, or Deeplearning4j.

[0031] In step 6), the UMAP method or t-SNE method is used to visualize the vector representation of all cells.

[0032] A computer system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0033] The beneficial effects of this invention are:

[0034] A single-cell pseudo-time trajectory analysis model based on word vectors is first proposed, which can improve the analysis speed of single-cell pseudo-time trajectory.

[0035] A cell distance representation method based on revealed comparative advantage index is proposed, which can improve the accuracy of single-cell pseudo-time trajectory.

[0036] The method proposed in the application can further comprehensively and systematically depict the state and fate of cells, including cell cycle, cell differentiation and cell activation, etc., and provides an analysis basis for identifying different rare cell subtypes and variant genes of different cell subtypes in tissues, and has wide application prospects in the fields of tumor, developmental biology, life science, etc. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 It is a workflow schematic diagram of the application.

[0038] Figure 2 It is a gene text sample schematic diagram generated by random walk of embryonic somatic cell data.

[0039] Figure 3 It is a word vector representation schematic diagram of genes.

[0040] Figure 4 It is a pseudo-time trajectory result schematic diagram of embryonic cell development. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0042] The application provides a single-cell transcriptome data pseudo-time trajectory analysis method, which first calculates a gene revealed comparative advantage matrix by using a single-cell transcriptome sequencing matrix; calculates the proximity between genes by using the revealed comparative advantage matrix; generates a gene network by taking a specific proximity threshold; generates a gene text by random walking on the gene network; encodes the gene text by word vectors to finally generate an embedded vector representation of the gene. Based on the embedded vector representation of the gene, each cell can be regarded as a vector in the gene space. After vectorization of the cell, a pseudo-time trajectory can be inferred, which is used for single-cell analysis downstream tasks such as cell type identification, cell clustering, visualization, cell development trajectory inference, etc.

[0043] It includes a complete set of analysis process, including the following seven modules: 1) data standardization preprocessing module, 2) gene explicit comparative advantage analysis module, 3) network construction based on gene similarity module, 4) random walk gene text generation module, 5) gene word vector representation module, 6) cell word vector representation module, 7) cell pseudo-time trajectory generation module.

[0044] Among them, the data standardization preprocessing module normalizes the single cell transcriptome data, standardizes the data, the gene explicit comparative advantage analysis module is used to obtain the explicit comparative advantage matrix, the gene network construction based on gene similarity module is used for gene network construction, the random walk gene text generation module is used to form the gene text composed of gene sequences; The gene word vector representation module is used to convert the gene text composed of gene sequences into gene word vectors; The cell word vector representation module is used to obtain the word vector representation of the cell in the gene space; The cell pseudo-time trajectory generation module is used to generate the embryonic cell development pseudo-time trajectory result.

[0045] It specifically includes the following steps:

[0046] 1) Calculate the gene explicit comparative advantage matrix using the single cell transcriptome sequencing matrix, which needs to be normalized before calculation:

[0047] X RNAcg =log p(1+X cg )

[0048] Among them, X RNAcg is the normalized transcriptome, X cg is the expression amount of the gth gene of the cth cell in the transcriptome;

[0049] Using to calculate the explicit comparative advantage of each gene expressed by each cell, and convert the single cell transcriptome sequencing matrix into a gene display comparative advantage matrix, wherein x cg is the expression amount of the cth cell on the gth gene, and RCA cg is the explicit comparative advantage of the cth cell on the gth gene.

[0050] Among them, the explicit comparative advantage index RCA is an index used to quantitatively measure the competitiveness or importance, that is, the comparative advantage in the population. The explicit comparative advantage is introduced into the single cell transcriptome data analysis, and its index can represent the importance of the gene to the function of the cell.

[0051] 2) After obtaining the gene display comparative advantage matrix, calculate the similarity between genes and obtain the gene similarity matrix, and construct a gene network according to the similarity.

[0052] That is, through

[0053]

[0054] Calculate the similarity between genes, where P represents the conditional probability, φ ij This indicates the similarity between gene i and gene j.

[0055] After obtaining the gene similarity matrix, each gene is regarded as a node in the network. A suitable similarity threshold is selected. When the similarity between two genes is greater than the threshold, it is considered that there is an edge connecting the two genes. Based on this, the gene network can be constructed.

[0056] The gene similarity construction network module uses the Pearson Correlation Coefficient similarity measure, which can be replaced with other similarity measures, such as Cosine Similarity or Jaccard Coefficient.

[0057] 3) Take an initial node in the gene network and start walking randomly from the initial node, recording each gene visited to form a gene text composed of gene sequences; take an initial node in the network (you can traverse the network or take a random node), start walking randomly from the initial node, that is, walk from one node to the next neighbor node with the edge weight as the probability, and record each gene visited to form a gene text composed of gene sequences.

[0058] 4) A word embedding algorithm is used to convert the gene text composed of gene sequences into word vectors of genes. The word embedding algorithm uses Word2vec or GloVe, Deeplearning4j.

[0059] 5) After obtaining the vector representation of the gene, since each single cell expresses a set of genes, the gene vectors expressed by the cell are summed with the expression level as the weight, and the resulting vector can be used as the word vector representation of the cell in the gene space.

[0060] 6) Visualize the vector representations of all cells to obtain the pseudo-time trajectory results of embryonic cell development. The UMAP method or t-SNE method is used to visualize the vector representations of all cells.

[0061] All cells are visualized using the UMAP (Uniform Manifold Approximation and Projection) method, which can better preserve the global structure, has fast running speed, and has better scalability compared to other visualization algorithms. In addition, UMAP has no computational limit on the dimension of embedding, and is a general visualization method. The UMAP visualization method is used to generate cell pseudo-time trajectories. The basic hyperparameters of UMAP include n_neighbors, k-nn clustering k, min_dist minimum distance, then n_components is the target dimension of dimension reduction, and metric can set the distance calculation method here.

[0062] Embodiment

[0063] Based on the patent of the application, the RNA-seq data set of embryonic body cells (EBs) from the human body in a 27-day time process is processed and analyzed. The embryonic body cell data set contains 16825 cells, and each cell detects 17580 genes.

[0064] The embryonic body cell data set is a medium-sized single-cell transcriptome data set, which contains 16,825 human embryonic body cells, each cell is regarded as a piece of data, which contains the expression amount of all genes in the cell, that is, the transcriptome data. First, the data is quality controlled, and dead cells and cells with too high gene expression are screened out, and finally 14528 cells are left. Secondly, the first 15000 genes with the largest variance are selected. Finally, the transcriptome data is subjected to a standardized operation of log (1+X

[0065] X RNAij = log p(1+X ij )

[0066] In the above formula, X RNAij represents the standardized preprocessed transcriptome data, and X ij represents the expression amount of the jth gene of the ith cell. In this case, the maximum value of i is the number of cells, that is, 14528, and the maximum value of j is the number of gene species, that is, 15000.

[0067] After the standardized data, the explicit comparative advantage (RCA) of each gene expressed by each cell is first calculated to obtain the gene explicit comparative advantage matrix.

[0068] After obtaining the gene explicit comparison advantage matrix, the similarity between genes and genes is calculated to obtain the gene similarity matrix. After obtaining the gene similarity matrix, each gene is regarded as a node of the network, a suitable similarity threshold is taken, 0.65 in this case, when the similarity between two genes is greater than the threshold, it is considered that there is an edge between the two genes, and accordingly the gene network can be constructed.

[0069] After the gene network is constructed, an initial node is randomly selected in the network, and a random walk is started from the initial node, that is, from a node to the next neighbor node with the edge weight as the probability. Each gene walked through is recorded, that is, a text composed of gene sequences is formed, as shown in Figure 2 .

[0070] After generating the text composed of gene sequences, the Word2vec algorithm is used, and the model includes an input layer, a hidden layer and an output layer. The model framework mainly includes the continuous bag-of-words model (CBOW) and the skip-gram model (SKIM) according to the input and output. The text is input into the CBOW or SKIM model to generate the embedded vector representation of the gene. The word vector representation result of the gene is shown in Figure 3 .

[0071] After obtaining the embedded vector representation of the gene, since each single cell expresses a group of genes, the gene vectors expressed by the single cell are added with the expression amount as the weight to form a combined vector which can be used as the vector representation of the cell in the gene space.

[0072] The vector representations of all cells are visualized using the UMAP (Uniform Manifold Approximation and Projection) method. Here we use the Euclidean distance, and actually the two variables n_neighbors and min_dist have a great influence on clustering. Here n_neighbor is 20. min_dist is a variable between 0 and 1. The min_dist is optimized by Bayesian optimization to generate the embryonic cell development pseudo-time trajectory result Figure 4 ).

[0073] The method provided by the application can further comprehensively and systematically depict the state and fate of cells, including cell cycle, cell differentiation and cell activation, etc., and provides an analysis basis for identifying different rare cell subtypes and variant genes of different cell subtypes in tissues, and has a wide important application prospect in the fields of tumor, developmental biology, life science, etc.

[0074] The application further provides a computer system comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the steps of the single-cell transcriptome data pseudo-time trajectory analysis method when executing the computer program.

[0075] The application further provides a computer readable storage medium, wherein computer readable storage medium stores computer program instructions, wherein the computer program instructions are executed by a processor to make the processor execute the steps of the single-cell transcriptome data pseudo-time trajectory analysis method.

[0076] The embodiments should not be considered as limiting the application, but any improvement based on the spirit of the application should be within the protection scope of the application.

Claims

1. A method of analyzing single-cell transcriptomic data pseudo-time trajectories, the method comprising: It comprises the following steps: ​ 1) Calculate the gene explicit comparative advantage matrix using single cell transcriptome sequencing matrix; 2) After obtaining the gene explicit comparative advantage matrix, calculate the similarity between genes and obtain the gene similarity matrix, and construct a gene network according to the similarity; 3) Take an initial node in the gene network, and start random walking from the initial node, record each gene walked through to form a gene text composed of gene sequences; 4) Convert the gene text into gene word vectors using a word embedding algorithm; 5) Add the gene vectors expressed by the cells with expression as the weight to form a sum vector as the word vector representation of the cell in the gene space; 6) Visualize the vector representation of all cells to obtain the embryonic cell development pseudo-time trajectory result, In step 1), the single cell transcriptome data is normalized. wherein, is the normalized transcript profile, is the expression level of the c gene of the g cell in the transcript profile, In step 1), for the normalized single-cell transcriptome data, the explicit comparative advantage of each gene expressed by each cell is calculated, and the single-cell transcriptome sequencing matrix is converted into a gene explicit comparative advantage matrix, wherein is the explicit comparative advantage of the cth cell on the gth gene. is the explicit comparative advantage of the cth cell on the gth gene. In step 2), by , The similarity between genes and genes is calculated, where P represents conditional probability, represents the similarity between genes i and genes j .

2. The method of analysis of single-cell transcriptomic data pseudo-time trajectories according to claim 1, wherein: In step 2), each gene is regarded as a node of the network, and a set similarity threshold is taken. When the similarity between two genes is greater than the set similarity threshold, it is considered that there is an edge between the two genes, and accordingly the gene network is constructed.

3. The method of analysis of single-cell transcriptomic data pseudo-time trajectories according to claim 1, characterized in that: In step 2), the Pearson correlation coefficient similarity measurement method or the cosine similarity, Jaccard coefficient is used to construct the gene network.

4. The method of claim 1, wherein: In step 3), the initial node in the gene network can be taken by traversal or random point.

5. The method of claim 1, wherein: In step 4), the word embedding algorithm uses the word embedding method Word2vec or GloVe, Deeplearning4j.

6. The method of single-cell transcriptomic data pseudo-time trajectory analysis of claim 1, wherein: In step 6), the UMAP method or t-SNE method is used to visualize the vector representation of all cells.

7. A computer system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that: The processor executes the computer program to realize the steps of the single cell transcriptome data pseudo-time trajectory analysis method in any one of claims 1 to 6.

Images