Method and system for online integration of multi-source single cell data

Abstract

The invention relates to the technical field of biological information, in particular to a method and system for online integration of multi-source single cell data. The method comprises the following steps: acquiring multi-source single cell transcriptome data; performing data quality evaluation based on gene expression quantity, mitochondrial gene proportion and cell data depth on the single cell transcriptome data to obtain initial quality control data; setting standardized filtering parameters according to the initial quality control data, and performing quality control parameter filtering on the single cell transcriptome data to obtain filtered cell data; performing gene expression characteristic analysis on the filtered cell data to generate high-variation gene expression characteristic data; and performing gene expression spatio-temporal distribution statistics according to the high-variation gene expression feature data to generate gene expression statistical feature data. According to the method, through refined quality control and batch effect correction, data from different sources can be seamlessly integrated on the premise of ensuring data quality, so that noise caused by different batches and technical platforms is eliminated.

Description

Technical Field

[0001] The present invention relates to the field of biological information technology, and in particular to a method and system for online integration of multi-source single-cell data. Background Art

[0002] Single-cell data refers to data obtained through single-cell sequencing technology, reflecting genetic information, gene expression, protein abundance, and other data at the level of individual cells; it includes the following sources: Single-cell transcriptome sequencing (scRNA-seq): This is the most widely used single-cell sequencing technology. By reverse transcribing, amplifying, and high-throughput sequencing the mRNA of individual cells, different cell types can be distinguished based on the similarity of transcriptional profiles, and even new cell types can be revealed. Single-cell genome sequencing: Detect the DNA and RNA of individual cells in the human body to identify new types of cells and their characteristics, as well as the responses of different types of cells to diseases or drugs. Single-cell proteomics: Studies protein expression and function at the single-cell level. Due to the non-amplification characteristics of proteins, a wide variety of types, low abundance, and a wide dynamic distribution range, its research lags behind nucleic acids. The integration of single-cell data from multiple sources can help analyze cell types and states from different levels and deeply explore regulatory mechanisms.

[0003] However, traditional methods for online integration of single-cell data from multiple sources often have the following problems: single-cell data are usually highly sparse, that is, the expression of most genes is zero in most cells, and cells from different data sources may have different sparsities, which will lead to missing data and increased noise during integration. In addition, some platforms may miss some genes or have large errors in the determination of low-expression genes. The different gene expression scales of different data sources may be due to differences in data preprocessing methods (such as normalization methods) or experimental designs. For example, some data sets may use different normalization methods such as TPM, FPKM or Counts, making direct comparison of gene expression difficult. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a method and system for online integration of single-cell data from multiple sources to solve at least one of the above technical problems.

[0005] To achieve the above purpose, a method for online integration of multi-source single-cell data includes the following steps:

[0006] Step S1: Acquire single-cell transcriptome data from multiple sources; perform data quality assessment on the single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; set standardized filtering parameters based on the initial quality control data, and perform quality control parameter filtering on the single-cell transcriptome data to obtain filtered cell data;

[0007] Step S2: performing gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; performing spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; performing differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data;

[0008] Step S3: performing batch effect identification based on the gene expression difference data, and performing batch feature evaluation to generate batch feature evaluation data; performing data dimensionality reduction analysis on the filtered cell data, and performing batch correction processing on the batch feature evaluation data to generate batch feature correction data;

[0009] Step S4: identifying cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classifying cell type feature data into cell subpopulations to obtain cell subpopulation classification data; filling missing values ​​in the cell subpopulation classification data to generate complete cell type annotation data;

[0010] Step S5: construct a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis to generate single-cell integrated data.

[0011] The present invention can increase the representativeness of samples and improve the universality of analysis results by integrating single-cell transcriptome data from multiple sources. This is crucial for capturing changes in different cell types and transcriptome features. Quality control that evaluates gene expression, mitochondrial gene ratio, and cell data depth helps identify low-quality samples. Through this step, the high reliability of the data can be ensured, and low-quality data can be prevented from interfering with subsequent analysis. Data filtering is performed according to the quality control evaluation results to remove low-quality cells, which helps to improve the accuracy of subsequent analysis and the credibility of the results, and ensure that the analysis is based only on high-quality data. By performing gene expression feature analysis on the filtered data, genes with large expression changes are identified, thereby revealing important biological processes in the cell. This helps to find potential biomarkers and key regulatory genes. Statistics based on the spatiotemporal distribution of gene expression can help analyze gene expression patterns at different time points or in different tissue regions, and gain a deeper understanding of the dynamic changes of genes in physiological processes. Differential expression analysis helps to reveal differences in gene expression under different conditions, determine which genes have significant changes under different cell states or treatment conditions, and provide important clues for subsequent biological research. Batch effects often interfere with the comparability of data. By identifying and evaluating batch effects, data from different batches can be ensured to be comparable in subsequent analysis. This step can avoid the interference of technical differences between batches on biological conclusions. Dimensionality reduction techniques (such as PCA or t-SNE) can help simplify the data structure and reveal the main variations in the data. Combining batch effect evaluation data for batch correction can remove technical noise and make biological characteristics more realistic. Correction for batch effects further enhances the consistency of data and improves the accuracy of analysis. Identifying cell type marker genes through gene expression statistical feature data helps to assign clear markers to each cell type, which is crucial for classifying and annotating cells. Accurate marker gene identification provides a solid foundation for subsequent cell population classification and research. Cell subpopulation classification can reveal subpopulation heterogeneity between cells of the same type and further refine the biological characteristics of cells. Filling missing values ​​helps ensure the integrity of the data and avoid missing data affecting the final analysis results, thereby improving the comprehensiveness of the analysis. By integrating batch-corrected data and cell type annotation data, a global single-cell data integration model can be created. This model can effectively integrate data from different sources and provide a unified framework for subsequent comprehensive analysis. This analysis can predict changes in cell types according to different conditions and reveal the dynamic changes of cell populations under different physiological states. This helps to deeply understand the spatiotemporal dynamics of biological processes and provide theoretical support for disease research and treatment. In summary, these steps can effectively improve the quality, accuracy and comparability of single-cell transcriptome data through multi-level data processing, quality control, differential analysis, batch correction and cell type annotation, and promote a deeper understanding of cell heterogeneity, gene expression regulation and disease mechanisms.

[0012] In the analysis of multi-source single-cell data of the present invention, batch effects can introduce systematic biases, making it difficult to compare data from different experimental batches during analysis. Extracting expression difference features and calculating the degree of variation of gene expression between batches helps to identify these batch effects. By constructing a batch effect feature matrix, the impact of batch effects on data can be clearly quantified and displayed. This step provides a data basis for subsequent batch effect correction, which helps to ensure the consistency of data from different batches, thereby avoiding misleading analysis. Calculating the correlation coefficient of gene expression between batches can assess the similarity of gene expression between different batches and identify gene sets significantly affected by batch effects. Identifying gene sets significantly affected by batch effects can focus on the areas with the strongest batch effects and provide target gene sets for correction steps. This enhances the pertinence and efficiency of data analysis, allowing downstream analysis to ignore or reduce deviations affected by batch effects. Principal component analysis (PCA) is a commonly used dimensionality reduction method that converts high-dimensional data into a few principal components by extracting the main sources of variation in the data for easy analysis. PCA dimensionality reduction can increase analysis speed, reduce the impact of noise on data, and help highlight the most valuable features in the data. Selecting principal components whose cumulative contribution rate reaches a preset threshold (such as 80%-90%) ensures that most of the information in the data is retained while reducing the computational complexity. This step provides more simplified and focused feature data for subsequent batch effect correction. Batch effect correction is performed based on batch feature evaluation data so that the reduced-dimensional data performs consistently across different sources, eliminating or reducing the impact of batch effects. Batch effect correction ensures the comparability and consistency of multi-source data. It significantly reduces the bias introduced by batch effects in downstream analysis and improves the accuracy and reliability of data integration and analysis results. This makes gene expression features more realistically reflect biological phenomena, helping to increase confidence in biological conclusions and the accuracy of interpretation.

[0013] The present invention also provides a system for online integration of single-cell data from multiple sources, which is used to perform the above-mentioned method for online integration of single-cell data from multiple sources. The system for online integration of single-cell data from multiple sources comprises:

[0014] The data quality control module is used to obtain single-cell transcriptome data from multiple sources; perform data quality assessment on the single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; set standardized filtering parameters based on the initial quality control data, and perform quality control parameter filtering on the single-cell transcriptome data to obtain filtered cell data;

[0015] The feature extraction module is used to perform gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; perform spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; perform differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data;

[0016] The batch correction module is used to identify batch effects based on gene expression difference data, perform batch feature evaluation, and generate batch feature evaluation data; perform data dimensionality reduction analysis on filtered cell data, and perform batch correction processing on batch feature evaluation data to generate batch feature correction data;

[0017] The cell annotation module is used to identify cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classify cell type feature data into cell subpopulations to obtain cell subpopulation classification data; fill missing values ​​in cell subpopulation classification data to generate complete cell type annotation data;

[0018] The integrated analysis module is used to build a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis and generate single-cell integrated data.

[0019] The module of the present invention is used to collect and preliminarily evaluate the quality of multi-source single-cell transcriptome data, and data quality is evaluated by parameters such as gene expression, mitochondrial gene ratio, and cell data depth. Through this initial quality control process, high-quality single-cell data can be screened out to ensure that downstream analysis is based on reliable input data. The filtering of quality control parameters and data standardization help to eliminate potential technical noise and abnormal data points, thereby improving the overall accuracy of the analysis and the repeatability of the results. Deep feature extraction is performed on the filtered cell data, especially analyzing the expression characteristics of high-variable genes and counting their distribution in time and space. The module also includes differential expression analysis to identify important gene expression patterns. High-variable gene expression characteristics help identify important genes that drive biological processes, allowing researchers to focus on the most biologically significant parts. Spatiotemporal distribution statistics provide data support for studying the dynamic behavior of cells under different conditions, and differential expression analysis further explores potential biomarkers and regulatory mechanisms. Batch effects are identified and evaluated through gene expression difference data, and dimensionality reduction analysis is performed on the data at the same time to handle the complexity of large data sets and ultimately achieve batch effect correction. Batch effect correction improves the comparability between different data sources, reduces the impact of technical variation, and ensures the biological interpretability of the analysis results. Dimensionality reduction reduces data dimensions, optimizes computational efficiency, and maintains key biological information. Based on the batch-corrected data, cell type marker genes are identified through gene expression statistical characteristics to achieve cell type annotation and classification. At the same time, missing values ​​are predicted and filled to ensure data integrity. Accurate cell type annotation is key to studying heterogeneity and specific biological processes. Cell subpopulation classification reveals the differentiation and functional characteristics of cell populations. Filling in missing values ​​makes the data more complete, avoids bias in analysis results, and improves the overall quality of the data. A single-cell data integration model based on batch-corrected data and cell type annotation data is constructed, and the model is used for dynamic cell type prediction analysis. The integrated analysis module can realize the integration of single-cell data across experimental conditions or samples, providing a holistic perspective for complex biological problems. Dynamic predictive analysis enables researchers to explore changes in cell states and types under different conditions, revealing cell differentiation trajectories and potential regulatory pathways. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0021] Figure 1 A schematic flow chart of the steps of the method for online integration of multi-source single-cell data of the present invention;

[0022] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0023] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0024] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 technicians in this field without creative work are within the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0026] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0027] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for online integration of multi-source single-cell data, the method comprising the following steps:

[0028] Step S1: Acquire single-cell transcriptome data from multiple sources; perform data quality assessment on the single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; set standardized filtering parameters based on the initial quality control data, and perform quality control parameter filtering on the single-cell transcriptome data to obtain filtered cell data;

[0029] The embodiment of the present invention obtains multiple single-cell transcriptome sequencing data sources, such as data from different experimental samples or public databases (such as GEO databases). The acquired data is preliminarily integrated, and the gene expression of each cell is calculated to generate an initial gene expression matrix. Then, the mitochondrial gene ratio of each cell is calculated based on the gene expression, and the specific method is to calculate the sum of the expression of mitochondrial-related genes and the total expression ratio of the cell. In addition, the number of genes detected in each cell is counted as a cell data depth indicator. Combining these indicators, the cells are quality assessed, and cells that may be technical errors or low quality are eliminated to generate initial quality control data. According to the initial quality control data, quality control filtering parameters are set, such as the minimum number of genes (200), the maximum mitochondrial gene ratio (20%) and the minimum sequencing depth (5000reads), and programming tools (such as Python or R language) are used to filter the data in multiple dimensions, and finally the filtered high-quality cell data is obtained.

[0030] Step S2: performing gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; performing spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; performing differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data;

[0031] The embodiment of the present invention performs gene expression feature analysis on the filtered cell data, first calculates the coefficient of variation (ratio of standard deviation to mean) of each gene in all cells, and selects the top 2000 genes with the highest coefficient of variation in descending order to form high-variability gene expression feature data. Then, by counting the expression levels of these genes in different cell types or time points, the spatiotemporal distribution statistics are generated, for example, the statistical values ​​of each group are calculated using the average expression amount and standard deviation of the grouping, and the results are visualized as a heat map or a bar chart to generate gene expression statistical feature data. In addition, based on the high-variability gene expression feature data, differential expression analysis is carried out between cell subpopulations or experimental groups, and genes with significant differences are identified by analysis tools (such as DESeq2) to generate gene expression difference data.

[0032] Step S3: performing batch effect identification based on the gene expression difference data, and performing batch feature evaluation to generate batch feature evaluation data; performing data dimensionality reduction analysis on the filtered cell data, and performing batch correction processing on the batch feature evaluation data to generate batch feature correction data;

[0033] The embodiment of the present invention identifies batch effects based on the gene expression difference data of step S2. First, cluster analysis is performed on the expression levels of differential genes in different experimental batches to identify whether there is a batch effect. Then, the batch effect is quantified by statistical methods (such as linear models or principal component analysis) to generate batch feature evaluation data. Subsequently, a dimensionality reduction algorithm (such as t-SNE or PCA) is applied to the filtered cell data to reduce the dimension of the data and facilitate visualization. In combination with the batch feature evaluation data, batch correction processing is performed on the reduced dimensionality data, for example, using the Harmony algorithm or the MNN (nearest neighbor matching) method, the differences between batches are adjusted to a unified benchmark, thereby generating corrected batch feature correction data.

[0034] Step S4: identifying cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classifying cell type feature data into cell subpopulations to obtain cell subpopulation classification data; filling missing values ​​in the cell subpopulation classification data to generate complete cell type annotation data;

[0035] The embodiment of the present invention utilizes gene expression statistical feature data to identify cell type marker genes from batch feature correction data. The specific operation is to screen significant marker genes (such as t-test p value <0.05) according to the expression level of genes in different cell types, and generate cell type feature data. Subsequently, according to the cell type feature data, the cells are sub-classified by a clustering algorithm (such as k-means or Louvain clustering) to obtain cell sub-group classification data. In order to complete the missing values ​​that may exist in the classification data, interpolation or a filling algorithm based on similar cells (such as kNN method) is used to fill in the missing values, generate complete cell type annotation data, and save it for downstream analysis.

[0036] Step S5: construct a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis to generate single-cell integrated data.

[0037] The embodiment of the present invention builds a single-cell data integration model based on batch feature correction data and complete cell type annotation data. The specific method is to use an integrated tool based on a neural network (such as Seurat or Scanpy) to uniformly model the cell types and expression characteristics of all samples. After the model is completed, dynamic predictions are made based on real-time input cell data, and the type and state of the cell are predicted by the classification function of the model to generate single-cell integrated data. These data can be used to explore the developmental trajectories and functional relationships between cells, and further support biological mechanism research.

[0038] The present invention can increase the representativeness of samples and improve the universality of analysis results by integrating single-cell transcriptome data from multiple sources. This is crucial for capturing changes in different cell types and transcriptome features. Quality control that evaluates gene expression, mitochondrial gene ratio, and cell data depth helps identify low-quality samples. Through this step, the high reliability of the data can be ensured, and low-quality data can be prevented from interfering with subsequent analysis. Data filtering is performed according to the quality control evaluation results to remove low-quality cells, which helps to improve the accuracy of subsequent analysis and the credibility of the results, and ensure that the analysis is based only on high-quality data. By performing gene expression feature analysis on the filtered data, genes with large expression changes are identified, thereby revealing important biological processes in the cell. This helps to find potential biomarkers and key regulatory genes. Statistics based on the spatiotemporal distribution of gene expression can help analyze gene expression patterns at different time points or in different tissue regions, and gain a deeper understanding of the dynamic changes of genes in physiological processes. Differential expression analysis helps to reveal differences in gene expression under different conditions, determine which genes have significant changes under different cell states or treatment conditions, and provide important clues for subsequent biological research. Batch effects often interfere with the comparability of data. By identifying and evaluating batch effects, data from different batches can be ensured to be comparable in subsequent analysis. This step can avoid the interference of technical differences between batches on biological conclusions. Dimensionality reduction techniques (such as PCA or t-SNE) can help simplify the data structure and reveal the main variations in the data. Combining batch effect evaluation data for batch correction can remove technical noise and make biological characteristics more realistic. Correction for batch effects further enhances the consistency of data and improves the accuracy of analysis. Identifying cell type marker genes through gene expression statistical feature data helps to assign clear markers to each cell type, which is crucial for classifying and annotating cells. Accurate marker gene identification provides a solid foundation for subsequent cell population classification and research. Cell subpopulation classification can reveal subpopulation heterogeneity between cells of the same type and further refine the biological characteristics of cells. Filling missing values ​​helps ensure the integrity of the data and avoid missing data affecting the final analysis results, thereby improving the comprehensiveness of the analysis. By integrating batch-corrected data and cell type annotation data, a global single-cell data integration model can be created. This model can effectively integrate data from different sources and provide a unified framework for subsequent comprehensive analysis. This analysis can predict changes in cell types according to different conditions and reveal the dynamic changes of cell populations under different physiological states. This helps to deeply understand the spatiotemporal dynamics of biological processes and provide theoretical support for disease research and treatment. In summary, these steps can effectively improve the quality, accuracy and comparability of single-cell transcriptome data through multi-level data processing, quality control, differential analysis, batch correction and cell type annotation, and promote a deeper understanding of cell heterogeneity, gene expression regulation and disease mechanisms.

[0039] Preferably, step S1 comprises the following steps:

[0040] Step S11: Acquire single-cell transcriptome data from multiple sources;

[0041] The embodiment of the present invention collects single-cell transcriptome data from different tissues or samples, and adopts high-throughput single-cell sequencing technology. Taking the 10x Genomics platform as an example, a library is first prepared from a single-cell suspension of the sample, and then the library is sequenced using a high-throughput sequencing device such as Illumina NovaSeq 6000. After sequencing, the raw data is converted into a gene expression matrix format using the supporting basic processing tools to generate expression data tables for each cell on different genes. Each column of the table corresponds to a cell, each row corresponds to a gene, and the value in the cell represents the expression level of a gene in a cell.

[0042] Step S12: Calculate the total expression of each cell and the average expression level of each gene in the expression spectrum for the single-cell transcriptome data, thereby obtaining gene expression data;

[0043] The embodiment of the present invention manually counts the total expression of each cell from the acquired single cell transcriptome data, that is, sums the expression of each cell on all genes. At the same time, the average expression level of each gene in all cells is calculated, and these data are organized into a table for subsequent analysis. In actual operation, the table can be processed by Excel or data analysis software, and a bar graph or line graph can be used to show the total expression distribution of the cell and the average expression level change trend of the gene to evaluate the rationality of the data.

[0044] Step S13: Calculate the mitochondrial gene expression ratio according to the gene expression data to generate mitochondrial gene ratio data;

[0045] The embodiment of the present invention extracts expression data related to mitochondrial genes from gene expression data, and first screens according to a public list of mitochondrial genes. Then, the proportion of mitochondrial gene expression in each cell to its total expression is calculated, and these proportions are sorted into a list and displayed as a percentage. In actual operation, the expression of the screened genes can be summed up manually or using a tool, and the mitochondrial gene expression ratio of each cell can be calculated, and the results can be plotted into a distribution map for intuitive evaluation.

[0046] Step S14: Analyze the number of genes detected in each cell of the single-cell transcriptome data to obtain cell data depth data;

[0047] The embodiment of the present invention performs gene quantity detection and analysis on the single-cell transcriptome data, and counts the number of genes detected in each cell one by one. The specific operation is to check row by row whether the expression of each gene in each cell is greater than zero. If it is greater than zero, it is counted as the gene detected in the cell. Finally, the number of genes in each cell is recorded in a table, and a statistical chart is made to show the distribution of the number of gene detections among cells.

[0048] Step S15: performing cell data quality assessment based on gene expression data, mitochondrial gene ratio data, and cell data depth data, thereby obtaining initial quality control data;

[0049] The embodiment of the present invention combines the results of steps S12-S14 to comprehensively evaluate the quality of cells based on the total cell expression, mitochondrial gene ratio and the number of gene detection. Set quality assessment standards, for example, the number of genes should be between 200 and 5,000, the mitochondrial gene ratio should not exceed 10%, and the total expression should be greater than 20,000. Screen out cells that meet these standards one by one, and record the results after screening as initial quality control data. In actual operation, you can filter step by step through the table screening function or self-compiled code, and organize the screening results in a table for the next step of analysis.

[0050] Step S16: Standardized filtering parameters are set according to the initial quality control data, and the single-cell transcriptome data is filtered in multiple dimensions, and cells that do not meet the gene number threshold, mitochondrial gene ratio threshold and sequencing depth threshold are eliminated in turn to obtain filtered cell data, wherein the standardized filtering parameters include the minimum number of genes, the maximum number of genes, the minimum number of cells, the maximum mitochondrial gene ratio and the minimum sequencing depth.

[0051] According to the initial quality control data, the embodiment of the present invention further sets specific filtering parameters, including standards such as the number of genes, the ratio of mitochondrial genes and the expression level. In actual operation, the single-cell data is first filtered by the number of genes, and cells with a gene number lower than 200 or higher than 5,000 are eliminated; then, more than 10% of the cells are eliminated according to the ratio of mitochondrial genes; finally, cells with a total expression of less than 20,000 are eliminated. In each step of the filtering process, the number and reasons for the elimination of cells are recorded one by one, and finally the cell data that has passed the filtration are sorted out and saved as a separate table for subsequent analysis. The changes in the number of cells before and after filtering can be displayed through statistical charts, and the data processing effect can be intuitively displayed.

[0052] The present invention collects single-cell transcriptome data from different sources or experimental conditions, which can improve the diversity and extensiveness of the data, ensure that the samples are more representative, and can cover different cell types, tissues, conditions or disease states. This is crucial for capturing the comprehensiveness of biological systems, revealing potential change characteristics, and improving the universality of analysis results. The total expression and average expression level of each gene in the single-cell transcriptome data are calculated, and the expression intensity of each gene in different cells can be provided. This helps to reveal the expression pattern and changes of genes in cells, and then understand the gene activity and regulatory mechanism in different cell types or states. In this way, key genes or regulatory factors with high expression levels can be identified, providing in-depth information about cell function, phenotypic differences and their role in biological processes. The high expression ratio of mitochondrial genes usually indicates changes in cell health. By calculating the ratio of mitochondrial genes in each cell, it can be identified whether the cell is in a state of stress, aging or death. A high proportion of mitochondrial gene expression may reflect changes in the energy demand and metabolic state of the cell, especially when analyzing cell quality and health. As an important indicator of cell quality control, the mitochondrial gene ratio can be used to eliminate cells with poor health status and ensure that the cell quality in the data set is high, thereby avoiding the interference of low-quality cells on the analysis results. Calculating the number of genes detected in each cell can help evaluate the sequencing depth and data integrity. Low gene counts may reflect insufficient sequencing depth or poor cell status (such as death or inactivity). This analysis helps to evaluate whether each cell can provide enough information to ensure high quality and comparability of the data. By analyzing the number of genes in each cell, cells with low expression or more missing data can be found and eliminated in subsequent steps to ensure the integrity of the data set and reduce the impact of unreliable data on the analysis results. Based on the comprehensive evaluation of gene expression, mitochondrial gene ratio and cell data depth, the quality of cell data can be comprehensively judged. This step can effectively eliminate low-quality, insufficiently sequenced or damaged cells to ensure that subsequent analysis relies only on high-quality data. By generating initial quality control data, the data quality status can be grasped from a global perspective, providing a basis for subsequent standardized filtering and more refined quality control steps. This comprehensive quality assessment is the basis for ensuring the success of single-cell data analysis. By defining standardized filtering parameters such as the number of genes, the proportion of mitochondrial genes, and sequencing depth, a systematic approach can be provided for data quality control. These standardized parameters help filter out abnormal cells, such as cells with too few or too many genes, cells with too high or too low a proportion of mitochondrial genes, and cells with unqualified sequencing depth. The use of multi-dimensional filtering methods can ensure a more refined filtering process and perform comprehensive quality control for different data features (such as the number of genes, gene expression ratio, and depth).Through these multi-dimensional filtering steps, the interference of low-quality cells on data analysis can be greatly reduced, and the accuracy and reliability of the final analysis results can be improved. The final data will only contain high-quality and representative cells, which will better reflect the true biological characteristics and cell heterogeneity, and ensure the scientificity and reliability of subsequent analysis results. These steps ensure the high quality and reliability of single-cell transcriptome data through multi-level quality control, data analysis and standardized filtering. By acquiring multi-source data, calculating gene expression, analyzing the proportion of mitochondrial genes, evaluating data depth, and performing meticulous quality control, a solid foundation can be provided for subsequent biological analysis and cell type identification. This processing method not only improves the accuracy of data analysis, but also ensures the biological significance and repeatability of the analysis results.

[0053] Preferably, step S2 comprises the following steps:

[0054] Step S21: constructing the original count matrix of each cell according to the filtered cell data, and performing logarithmic normalization transformation on the original count matrix to obtain logarithmic expression matrix data;

[0055] The embodiment of the present invention extracts the original count matrix from the filtered cell data, wherein each cell is used as the column of the matrix, the gene is used as the row of the matrix, and the value in the cell is the original gene expression. Then, in order to reduce the range difference of the expression, the logarithmic standardization method is used for conversion. The specific operation is to add a small constant (such as 1) to the original expression of each gene to avoid zero value when the logarithm is calculated, and then take the natural logarithm to form a logarithmic expression matrix data. Taking real application as an example, the log1p() function can be used in R language to quickly realize logarithmic conversion, and the result is saved as a matrix file for subsequent analysis.

[0056] Step S22: Calculating the coefficient of variation of genes based on the average expression amount and the standard deviation according to the logarithmic expression matrix data, thereby obtaining coefficient of variation data;

[0057] The embodiment of the present invention calculates the coefficient of variation of genes for the obtained logarithmic expression matrix data. First, the average expression of each gene in all cells is calculated; then, the standard deviation of the gene in all cells is calculated; finally, the coefficient of variation of each gene is obtained by the formula "coefficient of variation = standard deviation / mean value". In actual operation, Excel or programming tools (such as Python's pandas library) can be used to complete the calculations in sequence, and the results can be organized into a table, including gene name, average expression, standard deviation, and coefficient of variation columns.

[0058] Step S23: Arrange the genes in descending order according to the coefficient of variation data, and select the top 2000 genes with the highest coefficient of variation as high-variability genes, thereby obtaining the expression characteristic data of high-variability genes;

[0059] The embodiment of the present invention arranges the coefficient of variation data calculated in step S22 in descending order, and selects the top 2000 genes with the highest coefficient of variation as high-variability genes. In specific operations, the table can be sorted by the coefficient of variation column first, and then the top 2000 genes can be extracted manually or through a screening function. The extracted gene names and corresponding expression data are saved as a new table as high-variability gene expression feature data. This step helps to focus on the most biologically variable gene set in subsequent analysis.

[0060] Step S24: performing spatiotemporal distribution statistical analysis on the highly variable gene expression feature data, thereby generating gene expression statistical feature data;

[0061] The embodiment of the present invention uses the high-variability gene expression feature data to perform spatiotemporal distribution statistical analysis on the expression of genes in different cell types or sample time points. The specific method includes first calculating the average expression of each gene in different cell subpopulations or time points, and visually displaying the distribution change trend of gene expression in the form of a bar graph or heat map. The pheatmap package of the R language or the seaborn library of Python can be used to generate a visualization chart of the statistical results, and save the numerical statistical results as a feature table for subsequent differential analysis.

[0062] Step S25: using the gene expression statistical feature data to perform differential expression analysis on the highly variable gene expression feature data, thereby generating gene expression differential data.

[0063] The embodiment of the present invention performs differential expression analysis on highly variable genes based on gene expression statistical feature data. The specific steps include selecting two groups of biologically significant cell types or sample conditions, using commonly used differential analysis tools (such as DESeq2 or EdgeR) for calculation, and identifying genes with significant differences under the two groups of conditions. Statistical standards (such as p value <0.05, Fold Change>2) need to be set during the analysis process, and the differential genes and their corresponding statistical results are saved as a table as the output result of the gene expression difference data for subsequent biological interpretation and functional analysis.

[0064] The present invention can record the expression of each gene in each cell in a quantitative manner by arranging the gene expression of each cell into a raw count matrix. The raw count matrix is ​​the basic data format for single-cell transcriptome analysis and provides a starting point for subsequent analysis. There are often large differences in the expression in single-cell data. Logarithmic transformation helps to compress extreme values ​​and reduce the dominant effect of highly expressed genes on the results, making the data distribution more uniform. Logarithmic standardization transformation makes the data suitable for subsequent statistical analysis and model training, reduces the influence of sequencing depth and other technical errors, and thus improves the comparability of the data. The coefficient of variation (CV) is the ratio of the standard deviation to the mean value, which can measure the degree of change in gene expression between different cells. A high coefficient of variation of a gene indicates that its expression difference between cells is large, and these genes usually have high dynamic variability under cell types or biological conditions. By calculating the coefficient of variation, genes with large expression changes in cell populations can be identified, which may be involved in key biological processes, such as cell differentiation, stress response, etc. This step helps to screen out genes with potential biological significance in different cell populations from the perspective of overall expression level. The top 2000 genes with the highest coefficient of variation were screened out, focusing on those genes that vary most in the cell population. These genes are often important in biological processes such as cell differentiation, functional differences, or stress responses. After screening out highly variable genes, the sensitivity of the analysis can be improved, and more significant biological differences can be captured. By limiting the number of genes analyzed to 2000, the computational burden of subsequent analysis can be effectively reduced while maintaining the biological significance of the analysis. This makes subsequent analysis steps more efficient and targeted. Spatiotemporal distribution statistics can reveal the expression patterns of highly variable genes under different conditions or time points. By analyzing the expression of these genes in different cell populations, the trend of these genes changing over time or space in biological processes can be found, and genes that may be involved in key biological events can be identified. This step helps to identify the dynamic changes of genes under different environments or treatments, and gain a deeper understanding of the functions of genes under different conditions. For example, some genes may be highly expressed under certain conditions, while their expression decreases under other conditions, which can provide clues about the regulatory mechanisms of these genes. Based on the gene expression statistical feature data generated by spatiotemporal distribution statistics, differential expression analysis of highly variable genes can identify genes that show significant differences in different cell types, different conditions or different time points. These genes may be related to important changes in biological processes or specific physiological functions. Differential expression analysis helps reveal the differences in gene expression under different conditions and provides important clues for subsequent biological interpretations and research. For example, differentially expressed genes may be markers of a disease, therapeutic targets or driving factors for changes in cell function.Through differential expression analysis, genes with consistent expression under all conditions can be removed, and genes with significant expression differences under specific conditions can be analyzed intensively. This makes the analysis results more specific and relevant, helping to focus on key genes in biological research. These steps, from raw data processing to screening of highly variable genes, to spatiotemporal distribution and differential expression analysis, are constantly extracting and strengthening the most biologically meaningful information in the data. Through standardization of gene expression, variability analysis, spatiotemporal statistics, and differential expression analysis, researchers can discover potential cell type characteristics, dynamic changes, and regulatory factors in key biological processes. The effects of these steps are reflected in data quality control, improved analysis accuracy, and in-depth expansion of biological interpretation, which helps to understand cell heterogeneity, gene regulation, and disease mechanisms.

[0065] Preferably, step S21 includes the following steps:

[0066] Step S211: establishing a cell-gene correspondence relationship based on the filtered cell data, and constructing an original expression matrix including cells and genes;

[0067] The embodiment of the present invention extracts the gene information detected in each cell based on the filtered cell data, and constructs the original expression matrix based on the cell-gene correspondence. The specific operation is to organize the data in a way that cells are rows and genes are columns, and the initial expression value of each gene in the cell (e.g., original reads count) is filled into the matrix. Data processing tools (such as Pandas or Excel) can be used to implement this operation, for example, reading a filtered cell data file, matching the gene detection results to the corresponding cells one by one and constructing a matrix to ensure that each cell contains only filtered gene data.

[0068] Step S212: performing gene expression count statistics on the original expression matrix in the cell, and filling the count values ​​in the original expression matrix, thereby obtaining an original count matrix;

[0069] The embodiment of the present invention statistically completes the original expression matrix and calculates the expression counts of each gene in different cells. For example, missing expression values ​​are filled as zero counts; for data with expression values ​​less than a specific threshold (such as 1), a simple weighted smoothing algorithm can be used to adjust the original count matrix. Tools (such as Scanpy or Seurat) can be used for automated statistics, or custom scripts can be used to traverse the matrix row by row and fill and correct missing values ​​or deviation data to ensure that the count matrix is ​​complete and correct.

[0070] Step S213: calculating the total expression level of each cell in the original count matrix, thereby obtaining cell size factor data;

[0071] The embodiment of the present invention counts the total expression of each cell in the original count matrix, and the specific operation is to sum the gene expression values ​​in each row (cell) to generate the total expression data of each cell, which is called the cell size factor. After the calculation is completed, the result is saved as an independent size factor list and associated with the corresponding cells one by one. For example, a new column can be added to the data processing tool to store the cell size factor, and this can be used as a benchmark for subsequent data standardization processing. Usually, the total expression of the cell will be controlled within a reasonable range (such as 10,000 to 50,000) in the experiment to eliminate abnormal cells.

[0072] Step S214: Perform natural logarithm transformation on each expression value in the original count matrix, and standardize the logarithm-transformed values ​​according to the cell size factor data, thereby obtaining logarithmic expression matrix data.

[0073] The embodiment of the present invention performs natural logarithm transformation (log(x+1) to avoid log(0) calculation problems) on the expression values ​​in the original count matrix, and uses the cell size factor generated in step S213 to standardize the logarithm-transformed expression values. The specific operation is to divide each value in the matrix by its corresponding size factor, and then perform logarithmic transformation to eliminate systematic deviations caused by differences in sequencing depth or experimental conditions. After standardization is completed, the logarithmic expression matrix data is output and saved as a format file required for analysis (such as .csv or .h5ad) to ensure that the data is suitable for subsequent gene variation analysis or dimensionality reduction operations.

[0074] The present invention organizes the expression data of each gene in each cell into a matrix form by establishing a corresponding relationship between cells and genes to form an original expression matrix. This is a basic step in data analysis, which helps to organize complex single-cell data into a structured form, which is convenient for subsequent statistical analysis and processing. This matrix structure can intuitively display the expression of each gene in each cell, facilitates efficient data storage and rapid retrieval, and improves the operability and efficiency of the subsequent analysis process. The expression of genes in each cell is counted and counted, and the quantitative relationship of the expression situation can be clearly defined, thereby constructing an original count matrix. The count data is the expression value closest to the original sequencing result, which retains the original biological information and is the basis for subsequent analysis. By filling the count value, the integrity of the data is ensured, so that the original count matrix can fully reflect the expression level of each gene in the cell. The complete original count matrix provides rich biological information, which helps subsequent data normalization and noise reduction processing. The total expression of each cell can reflect the sequencing depth and cell activity, which is very important for data standardization and normalization processing. By calculating the total expression of cells, the difference in expression between different cells can be measured and a reference can be provided for standardization. The size factor is a key parameter for standardized expression data. By determining the size factor of each cell, the difference in total expression caused by sequencing depth or technical differences can be corrected, making the data between cells more comparable and reducing the impact of technical noise. By performing a natural logarithmic transformation on the raw count values, the expression range can be compressed and the dominant role of highly expressed genes in the overall analysis can be reduced. This helps to make the data distribution more even and reduce the impact of extreme values ​​on downstream analysis. The logarithmic transformed data is standardized using the cell size factor so that the expression data between different cells can be compared at the same level. Standardization helps to eliminate bias caused by sequencing depth or technical differences between samples, making the analysis results more reliable. The standardized logarithmic expression matrix provides high-quality data input for subsequent analysis. It not only retains biological variability, but also reduces the interference of technical noise on the analysis results, promoting more accurate gene difference analysis and cell clustering.

[0075] Preferably, step S24 comprises the following steps:

[0076] Step S241: grouping cells by time points according to the highly variable gene expression feature data, thereby generating time series grouping data;

[0077] The embodiment of the present invention groups cells by time point according to the characteristic data of highly variable gene expression. The specific operation is to first extract the cell time point information provided in the experimental design, for example, to obtain the time point corresponding to each cell through the sample number or marking label. Then, the cells are grouped by time point, time series grouping data is generated, and a unique identifier is assigned to each group. For example, assuming that the time points are T1, T2, and T3, the grouping results can be recorded as a grouping table, listing the time point grouping to which each cell belongs. The Pandas library in Python can be used to group the highly variable gene expression data by time point and store it as a CSV file for subsequent analysis.

[0078] Step S242: Calculate the expression level distribution of each highly variable gene in the time series grouping data, so as to obtain the expression distribution data at the time point;

[0079] The embodiment of the present invention calculates the expression level distribution of each high-variability gene in the time series grouping data. The specific method is to extract the expression value of the specified gene from the cell data in each time point grouping, calculate the distribution characteristics of the expression value, such as the mean, median and standard deviation, and obtain the time point expression distribution data. The operation can be performed by looping through each time grouping, performing statistical calculations on the expression data of each gene in the grouping, and storing the results in a matrix form, where the rows represent time points, the columns represent genes, and the matrix values ​​are the calculated expression distribution parameters. For practical applications, the statistical functions in SciPy can be used to directly perform batch statistics on the data set.

[0080] Step S243: fitting the expression distribution data at each time point using a kernel density estimation method to generate expression distribution curve data, wherein the fitting process specifically includes using a Gaussian kernel function as a kernel function for density estimation, calculating a bandwidth parameter using Scott's law, performing kernel density estimation on the expression data at each time point, and calculating a 95% confidence interval to evaluate the uncertainty of the estimate;

[0081] The embodiment of the present invention uses a kernel density estimation method to fit the expression distribution data at the time point. The specific method includes: selecting a Gaussian kernel function as the kernel function of the kernel density estimation, and using Scott's law to automatically calculate the bandwidth parameter to optimize the degree of smoothness. For the expression data at each time point, its distribution curve is calculated by kernel density estimation to generate expression distribution curve data. In order to evaluate the uncertainty of the fit, it is also necessary to calculate the 95% confidence interval by kernel density estimation. The operation can be implemented with the help of the statsmodels library in Python, using the KDEUnivariate module. The fitted distribution curve data can be stored in JSON format for visualization.

[0082] Step S244: analyzing the distribution differences between adjacent time points in the expression distribution curve data, thereby obtaining time dynamic characteristic data;

[0083] The embodiment of the present invention analyzes the distribution difference between adjacent time points in the expression distribution curve data. The specific operation is to calculate the difference index of the distribution curve data of adjacent time points, for example, using the earth moving distance (EMD) or Kullback-Leibler (KL) divergence of the distribution difference, and record the result as the time dynamic feature data. The operation can be implemented by a custom script, first comparing the integral values ​​of the distribution curves of adjacent time points, then calculating the difference index, and saving the result matrix, in which each cell in the matrix represents the difference value of adjacent time points.

[0084] Step S245: constructing a spatial autocorrelation matrix of gene expression based on the highly variable gene expression feature data, thereby obtaining spatial co-expression data;

[0085] The embodiment of the present invention constructs a spatial autocorrelation matrix of gene expression based on high-variability gene expression feature data. The specific method is to first extract the expression value of each gene and the spatial position information of the corresponding cell, and use a spatial autocorrelation analysis method (such as Moran's I or Geary's C) to evaluate the co-expression trend of genes in space. By traversing all genes, a spatial autocorrelation matrix with rows and columns as genes is constructed, and the matrix values ​​represent the spatial co-expression relationship between any two genes. This matrix is ​​usually saved in HDF5 format for efficient storage and fast reading.

[0086] Step S246: Perform module recognition analysis on the spatial co-expression data to identify gene modules with spatial co-expression patterns, thereby generating gene expression statistical feature data.

[0087] The embodiment of the present invention carries out module identification analysis to spatial co-expression data, identifies the gene module with spatial co-expression pattern. The specific operation is to utilize graph clustering method (such as Louvain algorithm or spectral clustering) to carry out module division to spatial co-expression matrix, and identifies gene module. Each module represents a gene set, and these genes show similar co-expression pattern in space. After completing identification, module information is recorded as a list structure, wherein each module comprises corresponding gene ID set, and it is saved as TXT file. The gene expression statistical characteristic data generated in this way can be directly used for subsequent analysis and visualization.

[0088] The present invention groups cells according to time points, which is helpful for studying the changing trend of gene expression at different time points and revealing the development or reaction trajectory of cells. Time series grouping data provides a framework so that subsequent expression analysis can be carried out in a dynamic manner, revealing the expression law of genes in time. This grouping helps to study the spatiotemporal regulation mechanism of genes and helps scientists understand the time-dependent characteristics of developmental processes or cell responses to environmental stimuli. By performing distribution calculations on the expression levels of each highly variable gene in the time series grouping, the expression differences of genes at different time points can be found. This is very useful for identifying the high expression peak or low expression period of genes. The expression level distribution data provides a quantitative basis for understanding the dynamic changes of genes and helps to identify potential regulatory patterns and biological events. The kernel density estimation (KDE) method generates a continuous expression distribution curve by smoothing the data, making the trend of gene expression over time clearer. The Gaussian kernel function provides an appropriate smoothing effect, making the distribution curve smooth and natural. Using Scott's rule to calculate the bandwidth parameter can ensure the stability and adaptability of the estimate. By calculating the 95% confidence interval, the uncertainty of the estimate can be evaluated, providing a confidence range for data analysis, and making the results more reliable. The generated expression distribution curve data provides the expression trend of genes over the entire time series, which can be used to compare the expression levels between different time points, thereby revealing the occurrence and development of biological events. Analyzing the differences in expression distribution between adjacent time points helps to identify significant change points in gene expression, revealing the transition of cell states or the occurrence of key biological processes. This analysis can capture rapid or gradual changes in gene expression patterns, helping researchers better understand the regulatory mechanisms of genes in the temporal dimension. By constructing a spatial autocorrelation matrix of gene expression, the co-expression pattern of genes in spatial distribution can be revealed. Spatial co-expression data reflects how genes are co-expressed in a cell population and is an important marker of intercellular communication and functional relationships. This data provides information about whether genes are spatially associated, helps understand the possible co-regulatory relationships between genes, and further reveals biological interactions within and between cells. Through module analysis, co-expressed genes can be clustered into modules, revealing potential gene regulatory networks and functional pathways. This helps to simplify complex data into easy-to-understand modular structures. The identified gene modules may represent gene populations with common biological functions or acting in the same pathway, thereby revealing the functional regions or cell type characteristics of cells. Module identification provides a new dimension for the study of gene function, helps to identify core genes and potential regulatory mechanisms in biological processes, and provides clues for further experimental verification.

[0089] Preferably, step S25 comprises the following steps:

[0090] Step S251: constructing a zero-inflated negative binomial distribution model of gene expression based on the gene expression statistical feature data, wherein the model construction specifically includes establishing a negative binomial distribution part for fitting non-zero expression values, establishing a Bernoulli distribution part for fitting zero expression probability, and constructing a complete zero-inflated negative binomial distribution model based on the above two parts;

[0091] The embodiment of the present invention constructs a zero-inflated negative binomial distribution model of gene expression based on the statistical feature data of gene expression. The specific operation is to first extract the expression value of each gene and count the proportion of zero expression values. For genes with a significantly high proportion of zero values, it is assumed that their expression data conform to the zero-inflated negative binomial distribution (ZINB). In model construction, the negative binomial distribution is first used to partially fit the distribution characteristics of non-zero expression values, and its mean and discrete parameters are calculated; the Bernoulli distribution is then used to partially fit the probability of zero expression, and the corresponding zero value probability parameters are calculated. Finally, these two parts are combined to construct a complete zero-inflated negative binomial distribution model, and the statsmodels library in Python is used to implement the initialization and fitting of the distribution model. For actual data, for example, the zero value ratio of a gene is 40%, and the initial zero inflation probability can be set to 0.4 to optimize the convergence of the model fitting.

[0092] Step S252: Calculate the initial parameter value of the zero-inflated negative binomial distribution model based on the moment estimation method, and use the EM algorithm to perform iterative optimization until the relative change of the parameter estimate is less than a preset threshold and the iteration is stopped, thereby generating model parameter data, wherein the EM algorithm includes an E step for calculating the posterior probability that each observation value comes from the zero-inflated part, and an M step for updating the negative binomial distribution parameters and the zero-inflated probability;

[0093] The embodiment of the present invention performs parameter estimation and optimization on the zero-inflated negative binomial distribution model. The specific operation includes: first, using the moment estimation method to calculate the initial parameter values ​​of the model, for example, by calculating the mean and variance of the non-zero expression value to initialize the negative binomial distribution parameters. Then use the EM algorithm for iterative optimization. In the E step, the posterior probability that each observation value comes from the zero-inflated part is calculated; in the M step, the negative binomial distribution parameters (such as mean and dispersion) and the zero-inflated probability are updated. Repeat the EM step until the relative change in the parameter is less than 0.001, and record the final parameter estimate as the model parameter data. This process can be implemented through the custom iteration method of the scipy.optimize library, and the iteration results are stored in JSON format for subsequent verification.

[0094] Step S253: performing a likelihood ratio test on the zero-inflated negative binomial distribution model according to the zero-inflated negative binomial distribution model, and evaluating the significance of differential expression, thereby generating significance test data;

[0095] The embodiment of the present invention performs a likelihood ratio test based on a zero-inflated negative binomial distribution model and evaluates the significance of differential expression. The specific operation is to compare the model fitting result with the corresponding simple negative binomial distribution model, and evaluate the significance of differential expression in the model by calculating the likelihood ratio values ​​of the two models and their corresponding p-values. The likelihood ratio values ​​are calculated one by one based on the expression data of each gene, and significance test data are generated. For actual scenarios with large amounts of data, a parallel computing framework (such as Python's multiprocessing module) can be used to accelerate the test process, and the results are stored as a CSV file.

[0096] Step S254: performing multiple test correction based on the Benjamini-Hochberg method on the significance test data to generate corrected significance data;

[0097] The embodiment of the present invention performs multiple test correction based on the Benjamini-Hochberg method on the significance test data. The specific operation is to first extract the original p-value list of all genes, arrange them in order from small to large, apply the Benjamini-Hochberg formula to each p-value to calculate its correction value (q value), and determine a significance threshold for controlling the false discovery rate (FDR), such as 0.05. All genes with q values ​​less than this threshold are retained to generate corrected significance data. This process can be completed by the p.adjust function in the R language or the statsmodels library of Python, and the corrected results are saved as Excel files for further analysis.

[0098] Step S255: screening differentially expressed genes according to the corrected significance data and the preset effect size, thereby obtaining a differentially expressed gene set;

[0099] The embodiment of the present invention performs differential expression gene screening according to the corrected significance data and the preset effect size. The specific operation is to screen genes that meet the q value less than 0.05 from the corrected significance data, and further filter out gene sets that meet the conditions of significance and biological effect in combination with the preset effect size threshold (such as log2 fold change greater than 1), and generate differential expression gene sets. Screening can be automatically completed by Excel's screening function or Python script, and the results are saved in the form of gene ID list.

[0100] Step S256: performing functional enrichment analysis on the differentially expressed gene set to obtain functional enrichment result data;

[0101] The embodiment of the present invention performs functional enrichment analysis on the differentially expressed gene set. The specific operation is to input the screened genes into a gene enrichment analysis tool (such as DAVID, GSEA or ClusterProfiler), select the corresponding biological database (such as GO or KEGG), calculate the significant enrichment degree of the differentially expressed genes in a specific functional module or pathway, and generate functional enrichment result data. The output results include the name of the significantly enriched functional module, the enrichment p value, and a specific list of participating genes. The functional analysis results are usually exported in the form of charts for intuitive presentation.

[0102] Step S257: The differentially expressed gene set, the corrected significance data, and the functional enrichment result data are processed to generate gene expression difference data.

[0103] The embodiment of the present invention integrates differentially expressed gene sets, correction significance data and functional enrichment result data. The specific operation is to set up a unified analysis report, and the significance information (q value and effect size) of the differentially expressed genes is associated with its functional enrichment module to form a comprehensive table, which is convenient for downstream research and use. The report content includes gene ID, correction significance q value, effect size, enrichment function module and its descriptive information. The final gene expression difference data is saved as an Excel file containing multi-layer information to support subsequent data mining and visual analysis.

[0104] The zero-inflated negative binomial distribution model of the present invention combines negative binomial distribution and Bernoulli distribution, and is applicable to the situation of too many zero values ​​(such as no expression) that are prevalent in single-cell transcriptome data. By fitting zero values ​​and non-zero expression values ​​separately, this model improves the accuracy of data fitting. The model can better describe the discreteness and variability of gene expression data, help reduce the deviation caused by too many zero values, and improve the reliability of differential expression analysis. The parameter values ​​are preliminarily obtained by the moment estimation method, and the EM algorithm is further iteratively optimized so that the model parameters gradually converge to the optimal value. The E step calculates the posterior probability of each observed value from the zero-inflated part, and the M step updates the model parameters to adapt the model to the data. This iterative optimization ensures the accuracy of parameter estimation, thereby improving the adaptability of the model to complex gene expression patterns. It is used to compare the goodness of fit of the full model and the simplified model (such as a model that does not consider differential expression), and evaluate the statistical support of gene expression significance in the model. The significance test data helps researchers identify genes with significant expression differences and provides statistical evidence to support the determination of differentially expressed genes. When a large number of significance tests are performed in genomic data, multiple comparisons can lead to an increase in false positive rates. The Benjamini-Hochberg method is a correction method for controlling the false discovery rate (FDR). Through correction, it ensures that the differentially expressed genes identified in high-throughput analysis have a low false positive rate, thereby improving the reliability of the research conclusions. In addition to statistical significance, the biological significance of expression changes must also be considered. Screening genes with large effect sizes can ensure that the screened genes are biologically meaningful. The screened differentially expressed gene sets are more accurate and focus on genes that have a potential significant impact on cell function. Through enrichment analysis, researchers can identify the enrichment trends of differentially expressed gene sets in biological pathways, gene ontology (GO) terms, KEGG pathways, etc. This helps to reveal the functional characteristics of gene sets, infer gene regulatory networks and their roles in biological processes, and support a deep understanding of cell behavior and function. The differentially expressed gene sets are integrated with their significance data and functional enrichment results to form a comprehensive gene expression difference data set. This comprehensive analysis helps researchers fully understand the statistical significance and biological functions of differentially expressed genes, supports further research on gene regulatory mechanisms, and provides a solid foundation for subsequent experimental verification.

[0105] Preferably, step S256 includes the following steps:

[0106] The differentially expressed gene sets were mapped to GO terms and KEGG pathways, and the enrichment significance was evaluated using the hypergeometric test. The enrichment score of each functional category was calculated, and the functional categories were sorted according to the enrichment score, and the top 50 functional categories with the highest scores were selected to obtain the functional enrichment result data. The formula for calculating the enrichment score of each functional category was:

[0107]

[0108] Among them, S is the enrichment score of the functional category, p is the p-value of the hypergeometric test, Os is the high-variability gene set, and Og is the gene set in the functional category.

[0109] The embodiment of the present invention maps genes to GO terms and KEGG pathways based on the differentially expressed gene set. The specific operation is to use a bioinformatics database (such as a GO database or a KEGG database) and its annotation file to associate the gene ID with the functional category, for example, by matching the functional module through the Ensembl gene ID. During the mapping process, it is necessary to select the GO level (such as biological process or molecular function) and KEGG pathway related to the research objective. For actual scenarios, for example, the differential gene set contains 500 genes, and the mapping can be completed through the clusterProfiler package of the R language or the gseapy library of Python. After the mapping is completed, a hypergeometric test is applied to each functional category to evaluate the significance of enrichment. The specific operation is to calculate the number of intersection genes between the high-variability gene set and the functional category gene set, and calculate the corresponding p-value based on the total gene set size, the number of genes in the functional category and the number of intersection genes. The hypergeometric test formula can be implemented by the phyper function of the R language or the scipy.stats.hypergeom module of Python. After obtaining the p-value for each functional category, the enrichment score of the functional category is calculated. According to the formula Where Os is the high-variability gene set, Og is the functional category gene set, |Os∩Og| represents the number of intersection genes, and |Og| represents the total number of functional category gene sets. For example, if the hypergeometric test p-value of a GO term is 0.0001, it contains 200 genes, and the number of intersection genes is 20, then the enrichment score is The formula can be calculated in batches through Python's numpy module, and the enrichment scores of all functional categories can be stored in the DataFrame structure. Next, all functional categories are sorted from high to low according to the enrichment score, and the top 50 functional categories with the highest scores are selected. Sorting and filtering can be done through the sort_values ​​and head methods of the Pandas library to ensure that the results can intuitively reflect the most significant functional categories. The screened functional categories include their names, corresponding p-values, intersection genes, and enrichment scores, and the results are saved as Excel files or JSON files. In application scenarios, this method can help researchers quickly identify biological processes or pathways that are significantly associated with high-variability gene expression. For example, in single-cell transcriptome data analysis, we can focus on whether the top-ranked functional categories are closely related to cell type differentiation or disease occurrence, thereby providing a reference for downstream experimental verification.

[0110] The present invention maps gene sets to GO terms (Gene Ontology) and KEGG pathways (Biological Pathway Database), which helps to reveal the specific roles of differentially expressed genes in cellular processes, molecular functions, and biological pathways. This mapping provides a systematic way to understand how genes work together in cellular functions and biological processes, helping to discover potential regulatory networks and biological mechanisms. The hypergeometric test is a statistical method used to determine whether a given gene set is significantly enriched in a specific functional category. It evaluates the difference between the observed number of genes and the randomly expected number. Through the hypergeometric test, researchers can quantitatively evaluate whether there is a higher functional enrichment in the gene set than the random distribution, helping to identify functional categories associated with biological significance. The calculation formula for the enrichment score S is: The p-value part: -log transformation p-value can convert significance into enrichment score, and the higher the score, the stronger the significance. Represents the proportion of genes that match the functional category in the highly variable gene set. Considering this proportion helps to identify functional categories that are not only significant but also highly enriched. This enrichment score can more comprehensively reflect the enrichment and significance of a functional category in the gene set, making the score more biologically meaningful. Sort the functional categories and select the top 50 with the highest scores to ensure that researchers can focus on the most significant and potentially biologically important functional categories. Narrow the scope of analysis and focus on the most relevant functional categories to facilitate verification in subsequent experiments; high-scoring functional categories usually involve key biological pathways or functions, which can suggest possible regulatory mechanisms and disease-related biological processes.

[0111] Preferably, step S3 comprises the following steps:

[0112] Step S31: extracting expression difference features between different data sources based on gene expression difference data, and calculating the degree of variation of gene expression between batches, thereby obtaining a batch effect feature matrix;

[0113] The embodiment of the present invention extracts the expression difference characteristics between different data sources according to the gene expression difference data. The core of this step is to analyze the gene expression data from different batches or experimental conditions and identify the difference of expression pattern. In specific operation, the gene expression between different batches of data can be compared, and the expression difference of each gene between different batches can be calculated. Based on these differences, the degree of variation of each gene in different batches is calculated, and the degree of variation of gene expression is quantified using analysis of variance (ANOVA) or coefficient of variation (CV), so as to obtain a batch effect feature matrix. For example, the numpy.var function in Python is used to calculate the expression variance of each gene in different batches, and then the variability of expression is evaluated by calculating the standard deviation or coefficient of variation.

[0114] Step S32: Calculate the correlation coefficient of gene expression between batches for the batch effect identification matrix, and identify the gene set significantly affected by the batch effect, so as to obtain batch feature evaluation data;

[0115] The embodiment of the present invention calculates the correlation coefficient of gene expression between batches to the batch effect identification matrix. Here, the Pearson correlation coefficient is mainly used to calculate the correlation of gene expression between each pair of batches, reflecting whether there is a difference in systematic expression pattern between different batches. For example, the gene expression correlation coefficient between different batches of data sets is calculated to obtain a matrix in which each element represents the correlation coefficient between two batches. Then, the gene sets significantly affected by the batch effect are identified according to the significance level of the correlation coefficient. The expression changes of these gene sets in different batches are large, which may be the source of the batch effect. Generally, a correlation coefficient less than a certain threshold value (such as 0.7 or 0.8) may represent a gene set significantly affected by the batch effect.

[0116] Step S33: performing data dimensionality reduction based on principal component analysis on the filtered cell data, calculating the contribution rate of the principal component, and selecting the principal component whose cumulative contribution rate reaches a preset threshold, thereby generating dimensionality reduction feature data;

[0117] The embodiment of the present invention performs data dimensionality reduction based on principal component analysis (PCA) on the filtered cell data. In this step, high-dimensional gene expression data is reduced to a low-dimensional space through PCA to better reveal the main change patterns between samples. First, the PCA algorithm is used to reduce the dimensionality of the filtered cell data, and the contribution rate of each principal component is calculated to reflect the contribution of each principal component to the overall data variation. PCA analysis can be performed through the sklearn.decomposition.PCA module, and the prcomp function can be used in R. Specifically, for each principal component, the proportion of its variance to the total variance is calculated, which is the contribution rate. In practical applications, the principal components with a cumulative contribution rate of 80% or 90% are selected, and the principal components that can represent most of the data variation are retained.

[0118] Step S34: performing batch effect correction processing on the dimension-reduced feature data based on the batch feature evaluation data, thereby generating batch feature correction data.

[0119] The embodiment of the present invention performs batch effect correction processing on the reduced dimension feature data based on batch feature evaluation data. Here, the ComBat algorithm or other batch effect correction methods can be used to correct the data after PCA dimension reduction. The ComBat algorithm estimates the effect of each batch through a model and makes adjustments to eliminate inconsistencies between batches. The combat function (for example, the sva package in R) can be used to correct the reduced dimension feature data. The corrected data can eliminate the influence of the batch effect, so that the data from different batches are more consistent in the feature space after dimension reduction, and generate batch feature correction data.

[0120] The present invention generates a batch effect feature matrix by extracting expression difference features and calculating the degree of variation between batches, which helps to identify and quantify the existence of batch effects. In this way, specific gene-level information can be provided in data analysis and subsequent correction to avoid misleading biological interpretations. By calculating the correlation coefficient of the batch effect feature matrix, the gene sets that are significantly affected by the batch effect are identified, which helps to understand which genes show consistent or inconsistent expression patterns between different batches. This analysis step helps to screen out genes that may be interfered by the batch effect, ensuring that the impact of these genes is minimized in subsequent analysis. The obtained batch feature evaluation data provides target genes and correction parameters for the correction of batch effects, improving data consistency and accuracy. Principal component analysis (PCA) is a data dimensionality reduction technology that can compress high-dimensional gene expression data into a few principal components that retain most of the variation information of the data. Dimensionality reduction operations can reduce computational complexity and improve the efficiency of data analysis. By selecting principal components whose cumulative contribution rate reaches a preset threshold (such as 80%-90%), the main expression variation features are retained and redundant information is eliminated. This not only reduces the dimensionality problem of batch effects, but also improves computational performance and controllability in subsequent batch corrections. Use batch feature evaluation data to correct the reduced-dimensional data to ensure that the deviation caused by batch effects is eliminated or minimized in multiple batches of data. The batch correction step can balance gene expression data from different sources so that they perform consistently in subsequent analyses. This improves the repeatability and accuracy of experimental results and avoids false positive or false negative results caused by batch effects. The corrected batch feature data can more realistically reflect biological characteristics and facilitate reliable comparison and analysis of group data.

[0121] Preferably, step S4 comprises the following steps:

[0122] Step S41: establishing a cell type marker gene database using gene expression statistical feature data, wherein the cell type marker gene database includes known cell type specific genes and expression pattern features;

[0123] The embodiment of the present invention establishes a cell type marker gene database according to the gene expression statistical feature data. In actual operation, the database can be established by referring to existing databases, such as Human Cell Atlas and CellMarker, in conjunction with known cell type specific genes in the experiment. These genes are marker genes highly expressed in different types of cells, and the expression pattern characteristics of each cell type will also be recorded. For example, through literature review and experimental verification, known marker genes of types such as nerve cells, immune cells, epithelial cells are collected and organized into a database. This database should include gene names, expression patterns and their performance in different cell types.

[0124] Step S42: comparing and analyzing the gene expression profile in the batch feature correction data with the cell type marker gene database to identify the type marker gene of each cell, thereby generating cell type feature data;

[0125] The embodiment of the present invention compares and analyzes the gene expression spectrum in the batch feature correction data with the cell type marker gene database. During the specific operation, the gene expression pattern of each cell can be matched with the marker gene in the cell type marker gene database by artificial or calculation methods. First, the gene expression data of each cell after the selected batch correction is then compared with the marker gene of each cell type in the database. The cell type closest to the marker gene expression pattern in the database can be identified by calculating the similarity of the expression value of each gene with the expression pattern of the marker gene in the database (such as by correlation or Euclidean distance), thereby distributing a label to each cell.

[0126] Step S43: constructing a cell similarity network based on the cell type feature data, and using a clustering algorithm to group the cells, thereby obtaining preliminary cell subpopulation classification data;

[0127] The embodiment of the present invention constructs a cell similarity network based on cell type feature data, and uses a clustering algorithm to group cells. In the experiment, the similarity of each pair of cells is first calculated. The similarity between cells can be quantified using traditional statistical methods such as the Pearson correlation coefficient by comparing the gene expression patterns of the cells. Afterwards, based on the similarity between cells, a cell similarity network is constructed, in which the nodes in the network represent cells, and the weights of the edges represent the similarity between cells. For cell grouping, methods such as hierarchical clustering or k-means clustering can be used to manually select the appropriate number of clusters, and use distance metrics (such as Manhattan distance or Euclidean distance) to cluster and group cells.

[0128] Step S44: performing cell subpopulation feature analysis on the preliminary cell subpopulation classification data to obtain cell subpopulation classification data, wherein the cell subpopulation feature analysis includes differential gene expression analysis and functional enrichment analysis;

[0129] The embodiment of the present invention performs cell subpopulation feature analysis on the preliminary cell subpopulation classification data. This step mainly analyzes the biological characteristics of each subpopulation in depth through differential gene expression analysis and functional enrichment analysis. Differential gene expression analysis can manually calculate the gene expression differences between each cell subpopulation and other subpopulations. Specifically, statistical methods such as t-test and Wilcoxon rank sum test are used to find out the genes specifically expressed in each cell subpopulation. Functional enrichment analysis uses GO (Gene Ontology) and KEGG pathway analysis to evaluate the biological processes, molecular functions and related signal pathways involved in the gene set. Enrichment analysis can be performed manually by referring to relevant databases, such as searching the GO database for the enrichment degree of genes in specific biological functions.

[0130] Step S45: Perform machine learning-based prediction and filling of missing values ​​in the cell subpopulation classification data to generate complete cell type annotation data, wherein the prediction and filling include inference based on similar cell features and completion based on expression patterns.

[0131] The embodiment of the present invention predicts and fills the missing values ​​in the cell subpopulation classification data based on machine learning. In actual operation, it is first necessary to check the missing values ​​in the cell subpopulation classification data, which may be due to the missing or incomplete gene expression data of some cells. In this case, similar cell characteristics can be used to infer the missing values. For example, the missing data can be estimated by finding other cells that are similar to the cells with missing data in expression patterns and using their known expression values ​​as references. In addition, regression analysis can also be used to predict missing values ​​based on known gene expression patterns. For example, based on other expression patterns of cells, the known data of the nearest neighbors are used to fill in the missing data. This process is usually performed by repeatedly adjusting and verifying the filling results to ensure that the filled missing values ​​do not deviate from the overall expression pattern of the cell population.

[0132] In the analysis of multi-source single-cell data of the present invention, batch effects can introduce systematic biases, making it difficult to compare data from different experimental batches during analysis. Extracting expression difference features and calculating the degree of variation of gene expression between batches helps to identify these batch effects. By constructing a batch effect feature matrix, the impact of batch effects on data can be clearly quantified and displayed. This step provides a data basis for subsequent batch effect correction, which helps to ensure the consistency of data from different batches, thereby avoiding misleading analysis. Calculating the correlation coefficient of gene expression between batches can assess the similarity of gene expression between different batches and identify gene sets significantly affected by batch effects. Identifying gene sets significantly affected by batch effects can focus on the areas with the strongest batch effects and provide target gene sets for correction steps. This enhances the pertinence and efficiency of data analysis, allowing downstream analysis to ignore or reduce deviations affected by batch effects. Principal component analysis (PCA) is a commonly used dimensionality reduction method that converts high-dimensional data into a few principal components by extracting the main sources of variation in the data for easy analysis. PCA dimensionality reduction can increase analysis speed, reduce the impact of noise on data, and help highlight the most valuable features in the data. Selecting principal components whose cumulative contribution rate reaches a preset threshold (such as 80%-90%) ensures that most of the information in the data is retained while reducing the computational complexity. This step provides more simplified and focused feature data for subsequent batch effect correction. Batch effect correction is performed based on batch feature evaluation data so that the reduced-dimensional data performs consistently across different sources, eliminating or reducing the impact of batch effects. Batch effect correction ensures the comparability and consistency of multi-source data. It significantly reduces the bias introduced by batch effects in downstream analysis and improves the accuracy and reliability of data integration and analysis results. This makes gene expression features more realistically reflect biological phenomena, helping to increase confidence in biological conclusions and the accuracy of interpretation.

[0133] The present invention also provides a system for online integration of single-cell data from multiple sources, which is used to perform the above-mentioned method for online integration of single-cell data from multiple sources. The system for online integration of single-cell data from multiple sources comprises:

[0134] The data quality control module is used to obtain single-cell transcriptome data from multiple sources; perform data quality assessment on the single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; set standardized filtering parameters based on the initial quality control data, and perform quality control parameter filtering on the single-cell transcriptome data to obtain filtered cell data;

[0135] The feature extraction module is used to perform gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; perform spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; perform differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data;

[0136] The batch correction module is used to identify batch effects based on gene expression difference data, perform batch feature evaluation, and generate batch feature evaluation data; perform data dimensionality reduction analysis on filtered cell data, and perform batch correction processing on batch feature evaluation data to generate batch feature correction data;

[0137] The cell annotation module is used to identify cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classify cell type feature data into cell subpopulations to obtain cell subpopulation classification data; fill missing values ​​in cell subpopulation classification data to generate complete cell type annotation data;

[0138] The integrated analysis module is used to build a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis and generate single-cell integrated data.

[0139] The module of the present invention is used to collect and preliminarily evaluate the quality of multi-source single-cell transcriptome data, and data quality is evaluated by parameters such as gene expression, mitochondrial gene ratio, and cell data depth. Through this initial quality control process, high-quality single-cell data can be screened out to ensure that downstream analysis is based on reliable input data. The filtering of quality control parameters and data standardization help to eliminate potential technical noise and abnormal data points, thereby improving the overall accuracy of the analysis and the repeatability of the results. Deep feature extraction is performed on the filtered cell data, especially analyzing the expression characteristics of high-variable genes and counting their distribution in time and space. The module also includes differential expression analysis to identify important gene expression patterns. High-variable gene expression characteristics help identify important genes that drive biological processes, allowing researchers to focus on the most biologically significant parts. Spatiotemporal distribution statistics provide data support for studying the dynamic behavior of cells under different conditions, and differential expression analysis further explores potential biomarkers and regulatory mechanisms. Batch effects are identified and evaluated through gene expression difference data, and dimensionality reduction analysis is performed on the data at the same time to handle the complexity of large data sets and ultimately achieve batch effect correction. Batch effect correction improves the comparability between different data sources, reduces the impact of technical variation, and ensures the biological interpretability of the analysis results. Dimensionality reduction reduces data dimensions, optimizes computational efficiency, and maintains key biological information. Based on the batch-corrected data, cell type marker genes are identified through gene expression statistical characteristics to achieve cell type annotation and classification. At the same time, missing values ​​are predicted and filled to ensure data integrity. Accurate cell type annotation is key to studying heterogeneity and specific biological processes. Cell subpopulation classification reveals the differentiation and functional characteristics of cell populations. Filling in missing values ​​makes the data more complete, avoids bias in analysis results, and improves the overall quality of the data. A single-cell data integration model based on batch-corrected data and cell type annotation data is constructed, and the model is used for dynamic cell type prediction analysis. The integrated analysis module can realize the integration of single-cell data across experimental conditions or samples, providing a holistic perspective for complex biological problems. Dynamic predictive analysis enables researchers to explore changes in cell states and types under different conditions, revealing cell differentiation trajectories and potential regulatory pathways.

[0140] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0141] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for online integration of multi-source single-cell data, characterized in that: The following steps are involved: Step S1: Acquire single-cell transcriptome data from multiple sources; Perform data quality assessment on single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; Standardized filtering parameters are set according to the initial quality control data, and the single-cell transcriptome data are filtered by quality control parameters to obtain filtered cell data; Step S2: performing gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; performing spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; performing differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data; Step S3: performing batch effect identification based on the gene expression difference data, and performing batch feature evaluation to generate batch feature evaluation data; performing data dimensionality reduction analysis on the filtered cell data, and performing batch correction processing on the batch feature evaluation data to generate batch feature correction data; Step S4: identifying cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classifying cell type feature data into cell subpopulations to obtain cell subpopulation classification data; filling missing values ​​in the cell subpopulation classification data to generate complete cell type annotation data; Step S5: construct a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis to generate single-cell integrated data.

2. The method for online integration of multi-source single-cell data according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Acquire single-cell transcriptome data from multiple sources; Step S12: Calculate the total expression of each cell and the average expression level of each gene in the expression spectrum for the single-cell transcriptome data, thereby obtaining gene expression data; Step S13: Calculate the mitochondrial gene expression ratio according to the gene expression data to generate mitochondrial gene ratio data; Step S14: Analyze the number of genes detected in each cell of the single-cell transcriptome data to obtain cell data depth data; Step S15: performing cell data quality assessment based on gene expression data, mitochondrial gene ratio data, and cell data depth data, thereby obtaining initial quality control data; Step S16: Standardized filtering parameters are set according to the initial quality control data, and the single-cell transcriptome data is filtered in multiple dimensions, and cells that do not meet the gene number threshold, mitochondrial gene ratio threshold and sequencing depth threshold are eliminated in turn to obtain filtered cell data, wherein the standardized filtering parameters include the minimum number of genes, the maximum number of genes, the minimum number of cells, the maximum mitochondrial gene ratio and the minimum sequencing depth.

3. The method for online integration of multi-source single-cell data according to claim 2, characterized in that: Step S2 includes the following steps: Step S21: constructing the original count matrix of each cell according to the filtered cell data, and performing logarithmic normalization transformation on the original count matrix to obtain logarithmic expression matrix data; Step S22: Calculating the coefficient of variation of genes based on the average expression amount and the standard deviation according to the logarithmic expression matrix data, thereby obtaining coefficient of variation data; Step S23: Arrange the genes in descending order according to the coefficient of variation data, and select the top 2000 genes with the highest coefficient of variation as high-variability genes, thereby obtaining the expression characteristic data of high-variability genes; Step S24: performing spatiotemporal distribution statistical analysis on the highly variable gene expression feature data, thereby generating gene expression statistical feature data; Step S25: using the gene expression statistical feature data to perform differential expression analysis on the highly variable gene expression feature data, thereby generating gene expression differential data.

4. The method for online integration of multi-source single-cell data according to claim 3, characterized in that: Step S21 includes the following steps: Step S211: establishing a cell-gene correspondence relationship based on the filtered cell data, and constructing an original expression matrix including cells and genes; Step S212: performing gene expression count statistics on the original expression matrix in the cell, and filling the count values ​​in the original expression matrix, thereby obtaining an original count matrix; Step S213: calculating the total expression level of each cell in the original count matrix, thereby obtaining cell size factor data; Step S214: Perform natural logarithm transformation on each expression value in the original count matrix, and standardize the logarithm-transformed values ​​according to the cell size factor data, thereby obtaining logarithmic expression matrix data.

5. The method for online integration of multi-source single-cell data according to claim 4, characterized in that: Step S24 includes the following steps: Step S241: grouping cells by time points according to the highly variable gene expression feature data, thereby generating time series grouping data; Step S242: Calculate the expression level distribution of each highly variable gene in the time series grouping data, so as to obtain the expression distribution data at the time point; Step S243: fitting the expression distribution data at each time point using a kernel density estimation method to generate expression distribution curve data, wherein the fitting process specifically includes using a Gaussian kernel function as a kernel function for density estimation, calculating a bandwidth parameter using Scott's law, performing kernel density estimation on the expression data at each time point, and calculating a 95% confidence interval to evaluate the uncertainty of the estimate; Step S244: analyzing the distribution differences between adjacent time points in the expression distribution curve data, thereby obtaining time dynamic characteristic data; Step S245: constructing a spatial autocorrelation matrix of gene expression based on the highly variable gene expression feature data, thereby obtaining spatial co-expression data; Step S246: Perform module recognition analysis on the spatial co-expression data to identify gene modules with spatial co-expression patterns, thereby generating gene expression statistical feature data.

6. The method for online integration of multi-source single-cell data according to claim 5, characterized in that: Step S25 includes the following steps: Step S251: constructing a zero-inflated negative binomial distribution model of gene expression based on the gene expression statistical feature data, wherein the model construction specifically includes establishing a negative binomial distribution part for fitting non-zero expression values, establishing a Bernoulli distribution part for fitting zero expression probability, and constructing a complete zero-inflated negative binomial distribution model based on the above two parts; Step S252: Calculate the initial parameter value of the zero-inflated negative binomial distribution model based on the moment estimation method, and use the EM algorithm to perform iterative optimization until the relative change of the parameter estimate is less than a preset threshold and the iteration is stopped, thereby generating model parameter data, wherein the EM algorithm includes an E step for calculating the posterior probability that each observation value comes from the zero-inflated part, and an M step for updating the negative binomial distribution parameters and the zero-inflated probability; Step S253: performing a likelihood ratio test on the zero-inflated negative binomial distribution model according to the zero-inflated negative binomial distribution model, and evaluating the significance of differential expression, thereby generating significance test data; Step S254: performing multiple test correction based on the Benjamini-Hochberg method on the significance test data to generate corrected significance data; Step S255: screening differentially expressed genes according to the corrected significance data and the preset effect size, thereby obtaining a differentially expressed gene set; Step S256: performing functional enrichment analysis on the differentially expressed gene set to obtain functional enrichment result data; Step S257: The differentially expressed gene set, the corrected significance data, and the functional enrichment result data are processed to generate gene expression difference data.

7. The method for online integration of multi-source single-cell data according to claim 6, characterized in that: Step S256 includes the following steps: The differentially expressed gene sets were mapped to GO terms and KEGG pathways, and the enrichment significance was evaluated using the hypergeometric test. The enrichment score of each functional category was calculated, and the functional categories were sorted according to the enrichment score, and the top 50 functional categories with the highest scores were selected to obtain the functional enrichment result data. The formula for calculating the enrichment score of each functional category was: Among them, S is the enrichment score of the functional category, p is the p-value of the hypergeometric test, Os is the high-variability gene set, and Og is the gene set in the functional category.

8. The method for online integration of multi-source single-cell data according to claim 7, characterized in that: Step S3 includes the following steps: Step S31: extracting expression difference features between different data sources based on gene expression difference data, and calculating the degree of variation of gene expression between batches, thereby obtaining a batch effect feature matrix; Step S32: Calculate the correlation coefficient of gene expression between batches for the batch effect identification matrix, and identify the gene set significantly affected by the batch effect, so as to obtain batch feature evaluation data; Step S33: performing data dimensionality reduction based on principal component analysis on the filtered cell data, calculating the contribution rate of the principal component, and selecting the principal component whose cumulative contribution rate reaches a preset threshold, thereby generating dimensionality reduction feature data; Step S34: performing batch effect correction processing on the dimension-reduced feature data based on the batch feature evaluation data, thereby generating batch feature correction data.

9. The method for online integration of multi-source single-cell data according to claim 8, characterized in that: Step S4 includes the following steps: Step S41: establishing a cell type marker gene database using gene expression statistical feature data, wherein the cell type marker gene database includes known cell type specific genes and expression pattern features; Step S42: comparing and analyzing the gene expression profile in the batch feature correction data with the cell type marker gene database to identify the type marker gene of each cell, thereby generating cell type feature data; Step S43: constructing a cell similarity network based on the cell type feature data, and using a clustering algorithm to group the cells, thereby obtaining preliminary cell subpopulation classification data; Step S44: performing cell subpopulation feature analysis on the preliminary cell subpopulation classification data to obtain cell subpopulation classification data, wherein the cell subpopulation feature analysis includes differential gene expression analysis and functional enrichment analysis; Step S45: Perform machine learning-based prediction and filling of missing values ​​in the cell subpopulation classification data to generate complete cell type annotation data, wherein the prediction and filling include inference based on similar cell features and completion based on expression patterns.

10. A system for online integration of single-cell data from multiple sources, characterized in that: A method for online integration of single-cell data from multiple sources as claimed in claim 1, wherein the system for online integration of single-cell data from multiple sources comprises: The data quality control module is used to obtain single-cell transcriptome data from multiple sources; perform data quality assessment on the single-cell transcriptome data based on gene expression, mitochondrial gene ratio, and cell data depth to obtain initial quality control data; set standardized filtering parameters based on the initial quality control data, and perform quality control parameter filtering on the single-cell transcriptome data to obtain filtered cell data; The feature extraction module is used to perform gene expression feature analysis on the filtered cell data to generate high-variability gene expression feature data; perform spatiotemporal distribution statistics of gene expression based on the high-variability gene expression feature data to generate gene expression statistical feature data; perform differential expression analysis on the high-variability gene expression feature data to generate gene expression differential data; The batch correction module is used to identify batch effects based on gene expression difference data, perform batch feature evaluation, and generate batch feature evaluation data; perform data dimensionality reduction analysis on filtered cell data, and perform batch correction processing on batch feature evaluation data to generate batch feature correction data; The cell annotation module is used to identify cell type marker genes on batch feature correction data through gene expression statistical feature data to generate cell type feature data; classify cell type feature data into cell subpopulations to obtain cell subpopulation classification data; fill missing values ​​in cell subpopulation classification data to generate complete cell type annotation data; The integrated analysis module is used to build a single-cell data integration model based on batch feature correction data and cell type annotation data; use the single-cell data integration model to perform dynamic cell type prediction analysis and generate single-cell integrated data.

Images