High-throughput transcriptome data analysis system and method based on cloud computing

Abstract

The invention discloses a high-throughput transcriptome data analysis system and method based on cloud computing, and relates to the technical field of bioinformatics, and the system comprises a multi-modal data preprocessing module, a dynamic loop screening module, an anchor point iteration and bipartite graph construction module, and a multi-dimensional classification execution module. High-efficiency data access and quality optimization are realized through a cloud computing cluster, and the data analysis efficiency and accuracy are improved by adopting a strategy of dynamically eliminating low-quality data and adaptively adjusting anchor point distribution and combining a local adaptive Gaussian kernel function and a parallelization spectral clustering algorithm. The method can automatically adapt to the quality fluctuation characteristics of different sequencing platforms, continuously optimize the classification result, and improve the intelligent level and biological interpretation ability of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to a high-throughput transcriptome data analysis system and method based on cloud computing. Background Art

[0002] With the rapid development of bioinformatics and high-throughput sequencing technologies, single-cell transcriptome analysis has shown tremendous potential in revealing cellular heterogeneity, gene expression regulatory mechanisms, and disease-related molecular networks. However, faced with the massive amount of single-cell transcriptome data, traditional analysis methods face numerous challenges in terms of data processing efficiency, computing resource requirements, and analytical accuracy. In cloud computing environments, the efficient use of distributed computing resources for data preprocessing, feature extraction, and classification analysis has become a key and challenging area of ​​current research.

[0003] In existing technologies, single-cell transcriptome data analysis usually relies on local computing environments, and its core steps include data preprocessing, dimensionality reduction, clustering, and classification. However, these methods generally have the following shortcomings: First, in the data preprocessing stage, traditional methods often lack the ability to efficiently index and parallelize the raw candidate cell data, resulting in inefficient data loading and screening. Second, in the cell classification process, existing technologies often use fixed thresholds or single indicators to screen and grade cells, which makes it difficult to fully reflect the specific distribution of gene expression, and may lead to the misjudgment of low-quality data or the omission of high-quality data. Third, in terms of anchor selection and optimization, existing methods usually rely on static cluster center determination strategies and lack dynamic iterative optimization mechanisms, which cannot effectively eliminate low-quality data and ensure the representativeness of the final anchor points. In addition, when constructing the cell-anchor similarity model, existing technologies often use a globally fixed Gaussian kernel bandwidth parameter, ignoring the local gene co-expression intensity and the dynamic similarity changes between cells, resulting in limited accuracy of the similarity calculation results. Finally, in the multi-dimensional classification execution stage, traditional methods lack parallel support for solving the Laplacian matrix and implementing the spectral clustering algorithm, which limits the efficient processing capability of large-scale datasets. Based on this, a high-throughput transcriptome data analysis system and method based on cloud computing is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a high-throughput transcriptome data analysis system and method based on cloud computing, which solves the technical problems of low efficiency, inaccurate classification and insufficient computing resource utilization in the existing technology in single-cell transcriptome data analysis.

[0005] The purpose of the present invention can be achieved through the following technical solutions: A high-throughput transcriptome data analysis system based on cloud computing includes: a multimodal data preprocessing module for receiving single-cell transcriptome expression data as raw candidate cell data through parallel computing nodes of a cloud computing cluster, randomly selecting the raw candidate cell data and establishing a candidate cell data index in a distributed memory database; The dynamic cyclic screening module is used to generate a temporary anchor point set based on the candidate cell data index established by the previous module through a hierarchical clustering algorithm, and use a gene expression specificity classification engine to classify the candidate cells into three levels. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation; the three-level classification includes high specificity, medium specificity, and low specificity; The anchor point iteration and bipartite graph construction module is used to randomly select data from the candidate cell dataset after the elimination mechanism for cluster analysis. The data quality indicators are repeatedly checked until no more cells can be eliminated, resulting in the selected raw data and the final anchor point set. A local adaptive Gaussian kernel function is then used to construct a cell-anchor weighted bipartite graph. The Gaussian kernel bandwidth parameter is dynamically adjusted by combining the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells. The multi-dimensional classification execution module is used to perform matrix decomposition on the cell-anchor weighted bipartite graph, extract the core feature vectors and perform dimensionality reduction operations, and finally complete the classification task of cell data through a hierarchical clustering algorithm.

[0006] As a further embodiment of the present invention, the single-cell transcriptome expression data in the multimodal data preprocessing module includes: cell identification: each cell has a unique ID; gene expression matrix: a gene × cell matrix, each element of which represents the expression level of a specific gene in a specific cell, i.e., a gene expression value; metadata: experimental batch, cell type annotation, and sequencing depth; The gene expression matrix is ​​the RNA sequencing data of each cell obtained based on single-cell RNA sequencing technology. After data preprocessing of each RNA sequencing data, the expression value of each gene in each cell is counted, and the gene expression data of all cells are organized into a matrix form, where rows represent genes and columns represent cells. Each value in the matrix represents the expression level of a specific gene in a specific cell.

[0007] As a further solution of the present invention: the random sampling algorithm in the multimodal data preprocessing module further includes: selecting part of the data through the random sampling algorithm to establish the candidate cell data index, and the probability distribution of random sampling is given by the formula Decision, formula The specific form is: ; Where x represents the expression value of the candidate cell data, μ is the mean of the gene expression value, and α is the regulatory factor. is a natural constant.

[0008] As a further solution of the present invention: generating a temporary anchor point set through a hierarchical clustering algorithm includes: adopting a cohesive strategy based on the hierarchical clustering algorithm, initially treating each candidate cell data as an independent cluster, and then gradually merging similar clusters by calculating the inter-cluster distance until a preset number of clusters is reached, and the inter-cluster distance is calculated using the cosine similarity formula; selecting the gene with the highest expression level from each cluster as the anchor point to generate a temporary anchor point set MD; the number of temporary anchor point sets is determined by the data scale and the preset anchor point density parameter.

[0009] As a further solution of the present invention: a gene expression specific classification engine is used to classify candidate cells into three levels, including: the gene expression specific classification engine is defined as follows: for a given candidate cell , whose gene expression specificity score ; Where m represents the total number of preset genes, j is the gene index, Indicates the The specificity score of candidate cells, represents the expression value of the j-th gene in the i-th candidate cell, It represents the baseline expression level of the jth gene in all cells n, and is used to measure the degree of deviation of the gene expression in cell i from the group average. Indicates the The standard deviation of the gene expression value; the gene expression value represents the expression level of a specific gene in a specific candidate cell, and is directly obtained through the gene expression matrix; According to the gene expression specificity score Combined with the preset classification threshold and The candidate cell-specific classification rules of the gene expression specificity classification engine are obtained as follows:

[0010] Among them, when and is the preset classification threshold and .

[0011] As a further solution of the present invention: according to the data quality indicators including gene coverage threshold and expression abundance standard deviation, the low specificity cell data is dynamically eliminated, including: according to the data quality indicators including gene coverage threshold and standard deviation of expression abundance , dynamically eliminate low-specificity candidate cell data; the logic for obtaining the gene coverage threshold is the ratio of the number of effective genes detected in a single candidate cell to the preset total number of genes m; the logic for obtaining the expression abundance standard deviation index is the standard deviation of all gene expression values ​​of a single candidate cell, according to the formula ;in, is the candidate cell index, is the gene index, is the baseline mean of all gene expression values ​​within candidate cell i, which is used to measure the degree of expression fluctuation within the cell; On this basis, combined with the gene coverage threshold and standard deviation of expression abundance Dynamically eliminate low-specificity candidate cell data. The specific elimination mechanism is: when the gene coverage of low-specificity candidate cell data is lower than the gene coverage threshold or its expression abundance standard deviation is higher than the expression abundance standard deviation , then the cell will be marked as invalid candidate cell data and removed from the candidate cell dataset to obtain the candidate cell dataset after the elimination mechanism is executed ; where k represents the number of remaining candidate cells.

[0012] As a further solution of the present invention, the dynamic adjustment mechanism of the local adaptive Gaussian kernel function includes: re-executing the clustering analysis step in the multimodal data preprocessing module and the data quality index evaluation step in the dynamic loop screening module for iterative analysis, and terminating the loop when no candidate cells are eliminated after t consecutive iterations; wherein the logic for obtaining t is: ;in, and are the gene expression-specific classification thresholds in the dynamic screening module, and Represent the preset regulatory factors and the mean expression abundance, represents the average gene coverage of the candidate cell dataset, represents the standard deviation of the mean expression abundance, is the preset maximum standard deviation; Get a selected candidate cell dataset ;in, Indicates the number of selected candidate cell data, for the selected candidate cell dataset Perform full hierarchical clustering, and each cluster obtained That is, the centroid of each candidate cell , which is obtained by calculating the average value of all cells in the cluster and determined as the final anchor point; Indicates the index of the final anchor point; Generate a set of cluster centers as the final anchor point set ; Where v represents the number of final anchor points; Based on the obtained selected candidate cell dataset and the final anchor point set The local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity ;in, Indicates selected candidate cell data With the final anchor point The similarity between and Select candidate cell data and the final anchor point The eigenvector of , σ is the bandwidth parameter of the Gaussian kernel; Introducing a preset gene co-expression intensity threshold θ and a preset inter-cell dynamic similarity attenuation coefficient , and define the update rule as ; Realize the dynamic adjustment of Gaussian kernel bandwidth parameters, where is the initial Gaussian kernel bandwidth parameter.

[0013] As a further solution of the present invention, the hierarchical clustering algorithm of the multi-dimensional classification execution module includes: using singular value decomposition (SVD) to decompose the cell-anchor weighted bipartite graph to extract the core feature vector; the cell-anchor weighted bipartite graph is L, and its decomposition form is: , where U and Y are the left singular vector matrix and the right singular vector matrix respectively, Σ is the singular value diagonal matrix, and T is the transpose operator; through this decomposition, the eigenvector after dimensionality reduction is obtained, and the clusters with the minimum distance are merged layer by layer until the final classification result of the selected candidate cell data is formed.

[0014] A high-throughput transcriptome data analysis method based on cloud computing includes the following steps: Step 1: receiving single-cell transcriptome expression data through parallel computing nodes of a cloud computing cluster, randomly selecting original candidate cell data and establishing candidate cell data indexes in a distributed memory database; Step 2: Based on the candidate cell data index established in step 1, a temporary anchor point set is generated through a hierarchical clustering algorithm. The candidate cells are classified into three levels using a gene expression specificity classification engine. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation. Step 3: Randomly select data from the candidate cell data after the data is eliminated in step 2 for cluster analysis, and recheck the data quality indicators until no original candidate cell data can be eliminated, thereby obtaining selected candidate cell data; the cluster center obtained by clustering the selected candidate cell data is determined as the final anchor point; Step 4: Based on the final anchor points determined in step 3, a local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity and construct a cell-anchor weighted bipartite graph. This is combined with the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells to implement a dynamic adjustment mechanism for the bandwidth parameter of the local calibration Gaussian kernel. Step 5: By performing matrix decomposition on the cell-anchor weighted bipartite graph, the core feature vectors are extracted and dimensionality reduction is performed, and finally the classification task of the cell data is completed through the hierarchical clustering algorithm.

[0015] Beneficial effects of the present invention: (1) This invention achieves efficient preprocessing of massive single-cell transcriptome data and dynamic elimination of low-quality data through parallel data index construction and dynamic loop screening mechanisms in cloud computing clusters. Compared with traditional static screening methods, this system can adapt to the data quality fluctuation characteristics of different sequencing platforms, significantly improving data cleaning efficiency and classification accuracy, while reducing the need for manual intervention and enhancing the automation and adaptability of the system. (2) This paper addresses the limitations of traditional methods, such as rigid anchor point selection and fixed kernel bandwidth parameters, by dynamically iteratively optimizing the anchor point set and adjusting the parameters of the local adaptive Gaussian kernel function, combined with matrix decomposition and hierarchical clustering algorithms. This system can accurately capture dynamic similarities between cells in complex transcriptome data, enhance the biological interpretation of feature extraction and classification tasks, and balance computational complexity with model generalization performance, providing high-quality structured representations for downstream analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 Schematic diagram of the system framework structure of the cloud computing-based high-throughput transcriptome data analysis system of the present invention; Figure 2 is an illustrative diagram of a multimodal data preprocessing module of the present invention; Figure 3 It is a schematic diagram of the implementation logic of the anchor point iteration and bipartite graph construction module of the present invention; Figure 4 It is a system flow diagram of the high-throughput transcriptome data analysis method based on cloud computing of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] Example 1 See also Figure 1As shown, the present invention is a high-throughput transcriptome data analysis system based on cloud computing, including a multimodal data preprocessing module, a dynamic loop screening module, an anchor point iterative optimization module, a bipartite graph construction engine module and a multidimensional classification execution module.

[0020] These modules work together to complete the entire analysis process, from raw data input to final cell classification. In practical applications, the system is deployed on a cloud computing platform, leveraging its powerful computing resources and storage capabilities to efficiently process large-scale single-cell transcriptome data.

[0021] The multimodal data preprocessing module is used to receive single-cell transcriptome expression data as raw candidate cell data through the parallel computing nodes of the cloud computing cluster, randomly select the raw candidate cell data and establish the candidate cell data index in the distributed memory database; The multimodal data preprocessing module receives the single-cell transcriptome expression data obtained using single-cell RNA sequencing (scRNA-seq) technology and divides it into multiple subsets according to batches to form an initial dataset; among them, a part of the data in the initial dataset is randomly selected as the original candidate cell dataset ,in represents the i-th cell, n represents the number of original candidate cell data, and a candidate cell data index is established in the distributed memory database for subsequent efficient retrieval and processing; the distributed memory database uses a key-value pair storage method to map the gene expression data of each cell to a unique index. A portion of the data is selected through a random sampling algorithm to establish a candidate cell data index, where i is the candidate cell index; The process of establishing a candidate cell data index in a distributed memory database is as follows: The cloud computing cluster consists of multiple computing nodes, each equipped with independent CPU and GPU resources. The single-cell transcriptome expression data are distributed to each computing node through the task scheduler. After receiving the data, the computing node uses a hash algorithm to generate a unique identifier for each data and stores the identifier in the distributed memory database. A portion of the data is selected through a random sampling algorithm to establish a candidate cell data index. The probability distribution of random sampling is given by the formula Decision, formula The specific form is: ; Where x represents the expression abundance of candidate cell data, μ is the mean expression abundance, and α is the regulatory factor. is a natural constant; the purpose of this formula is to take into account both high-expression abundance and low-expression abundance data, thereby improving the representativeness of the data. For example, when α = 0.5 and μ = 100, the probability of a cell with an expression abundance of 80 being selected is approximately 0.4, while the probability of a cell with an expression abundance of 120 being selected is approximately 0.6. This mechanism ensures the balance of the sampling results and the randomness and representativeness of the selection process. When establishing the candidate data index, an inverted index structure is used to group and store the data by gene name or cell ID for subsequent rapid retrieval. After the index is established, all computing nodes synchronize the index information through the message queue to ensure data consistency between nodes. It should be noted that the single-cell transcriptome expression data may include: gene expression profile data characterizing a single cell, including cell identification: each cell has a unique ID (such as Cell_001), a gene expression matrix: a gene × cell matrix, each element of which represents the expression level of a specific gene in a specific cell, i.e., a gene expression value, and metadata: information such as experimental batch, cell type annotation, and sequencing depth; the gene expression matrix is ​​based on the RNA sequencing data of each cell obtained by single-cell RNA sequencing (scRNA-seq) technology. After data preprocessing of each RNA sequencing data, the expression level of each gene in each cell is counted, for example, by counting UMIs (unique molecular identifiers) to eliminate bias caused by PCR amplification, and the gene expression data of all cells are organized into a matrix form, in which rows represent genes, columns represent cells, and each value in the matrix represents the expression level of a specific gene in a specific cell; See also Figure 2 For example, if the input single-cell transcriptome data contains 1 million cells, each with expression values ​​for 20,000 genes, the parallel computing nodes will split this data into multiple small blocks, each containing data for 10,000 cells, and randomly distribute it to different memory nodes. This approach not only significantly improves data loading speed but also effectively avoids performance bottlenecks caused by excessive data volume on a single node, laying the foundation for the efficient operation of subsequent modules.

[0022] The dynamic cyclic screening module is used to generate a temporary anchor point set based on the candidate cell data index established by the previous module through a hierarchical clustering algorithm, and use a gene expression specificity classification engine to classify the candidate cells into three levels. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation; the three-level classification includes high specificity, medium specificity, and low specificity; Specifically, based on the candidate cell data index established by the previous module, the process of generating a temporary anchor point set through the hierarchical clustering algorithm is as follows: Based on the hierarchical clustering algorithm, an agglomerative strategy is adopted. In the initial stage, each candidate cell data is regarded as an independent cluster. Then, similar clusters are gradually merged by calculating the inter-cluster distance until the preset number of clusters is reached. The inter-cluster distance is calculated using the cosine similarity formula, which is as follows: , where A and B represent the feature vectors of the two clusters respectively; when generating the temporary anchor point set MD; the gene with the highest expression level is selected from each cluster as the anchor point; the number of temporary anchor point sets is determined by the data scale and the preset anchor point density parameter; the specific process of using the gene expression specific classification engine to classify the candidate cells into three levels is as follows: the gene expression specific classification engine is defined as follows: for a given candidate cell , whose gene expression specificity score ; Where m represents the total number of preset genes, j is the gene index, Indicates the The specificity score of candidate cells, represents the expression value of the j-th gene in the i-th candidate cell, It represents the baseline expression level of the jth gene in all cells n, and is used to measure the degree of deviation of the gene expression in cell i from the group average. Indicates the The standard deviation of the gene expression value; the gene expression value represents the expression level of a specific gene in a specific candidate cell, and is directly obtained through the gene expression matrix; According to the gene expression specificity score Combined with the preset classification threshold and The candidate cell-specific classification rules of the gene expression specificity classification engine are obtained as follows:

[0023] in, and is the preset classification threshold and ,The specific value can be determined based on experimental data or professionals in related fields; It should be noted that when the candidate cell specificity is classified as high specificity, it means that the gene has a highly specific expression pattern in a particular candidate cell type; When the candidate cell specificity is classified as medium specificity, it means that the gene shows moderate specific expression in multiple relevant candidate cell types; When the candidate cell specificity is classified as low specificity, it means that the gene is commonly expressed in a wide range of candidate cell types; According to data quality indicators including gene coverage threshold and standard deviation of expression abundance , dynamically eliminate low-specificity candidate cell data; the logic for obtaining the gene coverage threshold is the ratio of the number of effective genes detected in a single candidate cell to the preset total number of genes m; the logic for obtaining the expression abundance standard deviation index is the standard deviation of all gene expression values ​​of a single candidate cell, according to the formula ;in, is the candidate cell index, is the gene index, is the baseline mean of all gene expression values ​​within candidate cell i, which is used to measure the degree of expression fluctuation within the cell; On this basis, combined with the gene coverage threshold and standard deviation of expression abundance Dynamically eliminate low-specificity candidate cell data to ensure the data quality of subsequent analysis; the specific elimination mechanism is: when the gene coverage of a low-specificity candidate cell data is lower than the gene coverage threshold or its expression abundance standard deviation is higher than the expression abundance standard deviation , then the cell will be marked as invalid candidate cell data and removed from the candidate cell dataset to obtain the candidate cell dataset after the elimination mechanism is executed ; Where k represents the number of remaining candidate cells; through the above elimination mechanism, the system can significantly improve the data quality and provide a reliable data basis for subsequent analysis.

[0024] See also Figure 3 The anchor point iteration and bipartite graph construction module is used to randomly select data from the candidate cell dataset after the elimination mechanism for cluster analysis. The data quality indicators are repeatedly checked until no more cells can be eliminated, resulting in the selected raw data and the final anchor point set. A local adaptive Gaussian kernel function is then used to construct a cell-anchor weighted bipartite graph. The Gaussian kernel bandwidth parameter is dynamically adjusted by combining the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells. Candidate cell dataset processed from the Dynamic Cycle Screening module Randomly select a subset H from the random sampling, and the probability distribution of random sampling is still determined by the formula The decision is made, and the cluster analysis step in the multimodal data preprocessing module and the data quality index evaluation step in the dynamic cycle screening module are executed again for iterative analysis. When no cells are eliminated after t consecutive iterations, the cycle is terminated. The logic for obtaining t is: ;in, and are the gene expression-specific classification thresholds in the dynamic screening module, and Represent the preset regulatory factors and the mean expression abundance, represents the average gene coverage of the candidate cell dataset, represents the standard deviation of the mean expression abundance, is the preset maximum standard deviation; Get a selected candidate cell dataset ;in, Indicates the number of selected candidate cell data, for the selected candidate cell dataset Perform full hierarchical clustering, and each cluster obtained That is, the centroid of each candidate cell , which is obtained by calculating the average value of all cells in the cluster and determined as the final anchor point; Indicates the index of the final anchor point; Generate a set of cluster centers as the final anchor point set ; Where v represents the number of final anchor points; Based on the obtained selected candidate cell dataset and the final anchor point set The local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity ;in, Indicates selected candidate cell data With the final anchor point The similarity between and Select candidate cell data and the final anchor point The eigenvector of , σ is the bandwidth parameter of the Gaussian kernel; In order to achieve dynamic adjustment of Gaussian kernel bandwidth parameters, a preset gene co-expression intensity threshold θ and a preset inter-cell dynamic similarity attenuation coefficient are introduced. , and define the update rule as ;in, is the initial Gaussian kernel bandwidth parameter; for example, when the initial bandwidth parameter =1.0, gene co-expression intensity threshold θ=0.5, attenuation coefficient = 0.2, if the distance between a selected candidate cell and the final one is 0.6, then the calibrated bandwidth parameter is Through the above mechanism, the system can effectively improve the accuracy of cell-anchor similarity calculation, thereby constructing a high-quality cell-anchor weighted bipartite graph; the cell-anchor weighted bipartite graph is used to select candidate cell datasets and the final anchor point set is a weighted bipartite graph vertex, with cell-anchor similarity is the edge weight; In this example, the anchor point iteration and bipartite graph construction module achieves efficient cleaning and structured representation of cellular data through iterative screening, anchor point generation, and dynamic bipartite graph construction, providing a high-quality input foundation for multimodal integration and downstream analysis. Its core innovation lies in combining a local adaptive kernel function with a dynamic parameter mechanism, balancing computational complexity and model generalization capabilities.

[0025] The multi-dimensional classification execution module is used to perform matrix decomposition on the cell-anchor weighted bipartite graph, extract the core feature vectors and perform dimensionality reduction operations, and finally complete the classification task of the selected candidate cell data through a hierarchical clustering algorithm. The implementation process is as follows: Specifically, the multi-dimensional classification execution module uses singular value decomposition (SVD) to decompose the cell-anchor weighted bipartite graph and extract the core feature vector; assuming that the cell-anchor weighted bipartite graph is L, its decomposition form is: , where U and Y are the left singular vector matrix and the right singular vector matrix respectively, Σ is the singular value diagonal matrix, and T is the transpose operator. Through this decomposition, the core eigenvectors of the cell-anchor weighted bipartite graph can be extracted, thereby reducing the data dimension. Subsequently, the multi-dimensional classification execution module uses a hierarchical clustering algorithm to classify the eigenvectors after dimensionality reduction. The core idea of ​​the hierarchical clustering algorithm is to gradually merge the nearest samples or clusters by calculating the distance matrix between samples until the final classification result is formed. Let the distance matrix between samples be JL, then the merging rule of hierarchical clustering can be expressed as: argmin( ),in is the distance between sample a and sample b. Through this algorithm, efficient classification of selected candidate cell data can be achieved while ensuring that the classification results have high accuracy and reliability. It should be noted that the above-mentioned singular value decomposition and hierarchical clustering algorithms are existing mature technologies, and the specific implementation process will not be described in detail. In this embodiment, the present invention realizes efficient processing and accurate classification of high-throughput transcriptome data through the collaborative design of multiple modules, combined with the parallel computing capabilities of cloud computing and a variety of innovative algorithms; in practical applications, for example in the field of cancer research, researchers can use the system and method provided by the present invention to analyze single-cell transcriptome data in tumor tissue, thereby revealing the gene expression characteristics of different cell subpopulations and their potential biological functions; through the above-mentioned detailed implementation steps, the present invention not only has significant technical advantages, but can also be widely used in biomedical research, drug development and other fields, showing high practicality and promotion value.

[0026] Example 2 Reference Figure 4, the present application also includes a high-throughput transcriptome data analysis method based on cloud computing, comprising the following steps: Step 1: receiving single-cell transcriptome expression data through parallel computing nodes of a cloud computing cluster, randomly selecting original candidate cell data and establishing a candidate cell data index in a distributed memory database; Step 2: Based on the candidate cell data index established in step 1, a temporary anchor point set is generated through a hierarchical clustering algorithm. The candidate cells are classified into three levels using a gene expression specificity classification engine. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation. Step 3: Randomly select data from the candidate cell data after the data is eliminated in step 2 for cluster analysis, and recheck the data quality indicators until no original candidate cell data can be eliminated, thereby obtaining selected candidate cell data; the cluster center obtained by clustering the selected candidate cell data is determined as the final anchor point; Step 4: Based on the final anchor points determined in step 3, a local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity and construct a cell-anchor weighted bipartite graph. This is combined with the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells to implement a dynamic adjustment mechanism for the bandwidth parameter of the local calibration Gaussian kernel. Step 5: By performing matrix decomposition on the cell-anchor weighted bipartite graph, the core feature vectors are extracted and dimensionality reduction is performed, and finally the classification task of the cell data is completed through the hierarchical clustering algorithm.

[0027] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0028] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A high-throughput transcriptome data analysis system based on cloud computing, characterized by: include: The multimodal data preprocessing module is used to receive single-cell transcriptome expression data as raw candidate cell data through the parallel computing nodes of the cloud computing cluster, randomly select the raw candidate cell data and establish the candidate cell data index in the distributed memory database; The dynamic cyclic screening module is used to generate a temporary anchor point set based on the candidate cell data index established by the previous module through a hierarchical clustering algorithm, and use a gene expression specificity classification engine to classify the candidate cells into three levels. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation; the three-level classification includes high specificity, medium specificity, and low specificity; The anchor point iteration and bipartite graph construction module is used to randomly select data from the candidate cell dataset after the elimination mechanism for cluster analysis. The data quality indicators are repeatedly checked until no more cells can be eliminated, resulting in the selected raw data and the final anchor point set. A local adaptive Gaussian kernel function is then used to construct a cell-anchor weighted bipartite graph. The Gaussian kernel bandwidth parameter is dynamically adjusted by combining the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells. The multi-dimensional classification execution module is used to perform matrix decomposition on the cell-anchor weighted bipartite graph, extract the core feature vectors and perform dimensionality reduction operations, and finally complete the classification task of cell data through a hierarchical clustering algorithm.

2. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that The single-cell transcriptome expression data in the multimodal data preprocessing module includes: cell identification: each cell has a unique ID, gene expression matrix: a gene × cell matrix, each element represents the expression level of a specific gene in a specific cell, i.e., gene expression value, metadata: experimental batch, cell type annotation, sequencing depth; The gene expression matrix is ​​the RNA sequencing data of each cell obtained based on single-cell RNA sequencing technology. After data preprocessing of each RNA sequencing data, the expression value of each gene in each cell is counted, and the gene expression data of all cells are organized into a matrix form, where rows represent genes and columns represent cells. Each value in the matrix represents the expression level of a specific gene in a specific cell.

3. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that: The random sampling algorithm in the multimodal data preprocessing module further includes: selecting part of the data through the random sampling algorithm to establish the candidate cell data index, and the probability distribution of random sampling is given by the formula Decision, formula The specific form is: ; Where x represents the expression value of the candidate cell data, μ is the mean of the gene expression value, and α is the preset regulatory factor. is a natural constant.

4. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that Generating a temporary anchor point set through a hierarchical clustering algorithm includes: using an agglomerative strategy based on the hierarchical clustering algorithm, initially treating each candidate cell data as an independent cluster, and then gradually merging similar clusters by calculating the inter-cluster distance until the preset number of clusters is reached. The inter-cluster distance is calculated using the cosine similarity formula; selecting the gene with the highest expression level from each cluster as the anchor point to generate a temporary anchor point set MD; the number of temporary anchor point sets is determined by the data scale and the preset anchor point density parameter.

5. The cloud computing-based high-throughput transcriptome data analysis system according to claim 1, characterized in that: A gene expression specific classification engine is used to classify candidate cells into three levels, including: the gene expression specific classification engine is defined as follows: for a given candidate cell , whose gene expression specificity score ; Where m represents the total number of preset genes, j is the gene index, Indicates the The specificity score of candidate cells, represents the expression value of the j-th gene in the i-th candidate cell, It represents the baseline expression level of the jth gene in all cells n, and is used to measure the degree of deviation of the gene expression in cell i from the group average. Indicates the The standard deviation of the gene expression value; the gene expression value represents the expression level of a specific gene in a specific candidate cell, and is directly obtained through the gene expression matrix; According to the gene expression specificity score Combined with the preset classification threshold and The candidate cell-specific classification rules of the gene expression specificity classification engine are obtained as follows: in, and is the preset classification threshold and .

6. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that: Dynamically remove low-specificity cell data based on data quality indicators including gene coverage threshold and expression abundance standard deviation, including: and standard deviation of expression abundance , dynamically eliminate the low-specificity candidate cell data; the logic for obtaining the gene coverage threshold is the ratio of the number of effective genes detected in a single candidate cell to the preset total number of genes m; the logic for obtaining the expression abundance standard deviation index is the standard deviation of all gene expression values ​​of a single candidate cell, according to the formula ;in, is the candidate cell index, is the gene index, is the baseline mean of all gene expression values ​​within candidate cell i, which is used to measure the degree of expression fluctuation within the cell; Combined gene coverage threshold and standard deviation of expression abundance Dynamically eliminate low-specificity candidate cell data. The specific elimination mechanism is: when the gene coverage of low-specificity candidate cell data is lower than the gene coverage threshold or its expression abundance standard deviation is higher than the expression abundance standard deviation , then the cell will be marked as invalid candidate cell data and removed from the candidate cell dataset to obtain the candidate cell dataset after the elimination mechanism is executed ; where k represents the number of remaining candidate cells.

7. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that: The dynamic adjustment mechanism of the local adaptive Gaussian kernel function includes: when the iterative analysis is performed again, if no candidate cells are eliminated after t consecutive iterations, the loop is terminated; wherein the acquisition logic of t is: ;in, and are the gene expression-specific classification thresholds in the dynamic screening module, and Represent the mean of preset regulatory factors and gene expression values, represents the average gene coverage of the candidate cell dataset, represents the standard deviation of the mean expression abundance, is the preset maximum standard deviation; Get a selected candidate cell dataset ;in, Indicates the number of selected candidate cell data, for the selected candidate cell dataset Perform full hierarchical clustering, and each cluster obtained That is, the centroid of each candidate cell , which is obtained by calculating the average value of all cells in the cluster and determined as the final anchor point; Indicates the index of the final anchor point; Generate a set of cluster centers as the final anchor point set ; Where v represents the number of final anchor points; Based on the obtained selected candidate cell dataset and the final anchor point set The local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity ;in, Indicates selected candidate cell data With the final anchor point The similarity between and Select candidate cell data and the final anchor point The eigenvector of , σ is the bandwidth parameter of the Gaussian kernel; Introducing a preset gene co-expression intensity threshold θ and a preset inter-cell dynamic similarity attenuation coefficient , and define the update rule as ; Realize the dynamic adjustment of Gaussian kernel bandwidth parameters, where is the initial Gaussian kernel bandwidth parameter.

8. The high-throughput transcriptome data analysis system based on cloud computing according to claim 1, characterized in that: The hierarchical clustering algorithm of the multi-dimensional classification execution module includes: using singular value decomposition to decompose the cell-anchor weighted bipartite graph and extract the core feature vector; the cell-anchor weighted bipartite graph is L, and its decomposition form is: , where U and Y are the left singular vector matrix and the right singular vector matrix respectively, Σ is the singular value diagonal matrix, and T is the transpose operator; through this decomposition, the eigenvector after dimensionality reduction is obtained, and the clusters with the minimum distance are merged layer by layer until the final classification result of the selected candidate cell data is formed.

9. A cloud computing-based high-throughput transcriptome data analysis method for implementing the cloud computing-based high-throughput transcriptome data analysis system according to any one of claims 1 to 8, characterized in that: Step 1: Receive single-cell transcriptome expression data through the parallel computing nodes of the cloud computing cluster, randomly select the original candidate cell data and establish the candidate cell data index in the distributed memory database; Step 2: Based on the candidate cell data index established in step 1, a temporary anchor point set is generated through a hierarchical clustering algorithm. The candidate cells are classified into three levels using a gene expression specificity classification engine. Low-specificity cell data are dynamically eliminated based on data quality indicators including gene coverage threshold and expression abundance standard deviation. Step 3: Randomly select data from the candidate cell data after the data is eliminated in step 2 for cluster analysis, and recheck the data quality indicators until no original candidate cell data can be eliminated, thereby obtaining selected candidate cell data; the cluster center obtained by clustering the selected candidate cell data is determined as the final anchor point; Step 4: Based on the final anchor points determined in step 3, a local adaptive Gaussian kernel function is used to calculate the cell-anchor similarity and construct a cell-anchor weighted bipartite graph. This is combined with the gene co-expression intensity threshold parameter and the dynamic similarity attenuation coefficient between cells to implement a dynamic adjustment mechanism for the bandwidth parameter of the local calibration Gaussian kernel. Step 5: By performing matrix decomposition on the cell-anchor weighted bipartite graph, the core feature vectors are extracted and dimensionality reduction is performed, and finally the classification task of the cell data is completed through the hierarchical clustering algorithm.

Images