A method for splitting a multi-sample mixed single-cell data mix

Abstract

The application relates to a method for sample-mixed single-cell data mixing and splitting, belonging to the technical field of single-cell analysis. The method comprises the following steps: respectively performing multi-sample mixed single-cell sequencing and batch mitochondrial sequencing of each sample; extracting a mitochondrial variation site frequency matrix; screening a high-variable site with large difference; calculating a correlation coefficient of the high-variable site frequency of each cell and each sample; taking a corresponding sample with a correlation coefficient value greater than 0 as the sample label of the cell; taking the correlation coefficient of the site frequency between cells in the single-cell mitochondrial variation site frequency matrix as an edge, taking each single cell as a point, starting from the point with a known sample label, propagating the label to the point with an unknown sample label through the edge until a model stopping condition is reached, and outputting the predicted label of each single cell. The method has high result accuracy and high calculation efficiency.

Description

Technical Field

[0001] This invention relates to a method for splitting mixed single-cell data from multiple samples, belonging to the field of single-cell analysis technology. Background Technology

[0002] In recent years, single-cell RNA-sequencing (scRNA-seq) has been widely used. While scRNA-seq assays can robustly identify single-cell clusters and cell types corresponding to genetic markers, variations in different rounds of experiments can introduce unwanted noise into the dataset, also known as batch effects. Furthermore, the high cost of conventional scRNA-seq experiments limits their application in analyzing large numbers of samples.

[0003] To address these challenges, various techniques have been developed to split data from mixed samples in scRNA-seq experiments using chemical probes or genetically engineered exogenous barcodes. While chemical probes can effectively split label-based data, this increases cost and sample preparation time due to the need for universal antibodies suitable for various analytes. On the other hand, genetically engineered exogenous barcodes do not require any antibodies, but their application is limited to cell culture systems or model organisms, making them unsuitable for obtaining single-cell data from mixed clinical samples. Furthermore, the experimental expertise required to create exogenous barcodes and correctly map them to cells also hinders the widespread adoption of these techniques.

[0004] The splitting of scRNA-seq data can also be achieved by analyzing natural genomic variations. Some methods rely on whole-genome sequencing or whole-exome sequencing as a reference. Sequencing reference-dependent methods can assign the split cell groups to individual samples, while sequencing reference-independent methods can only divide the split cells into multiple groups, without reconstructing the specific sample origin of each cell. However, these methods require the detection and analysis of single-nucleotide polymorphisms (SNPs) throughout the genome as endogenous barcodes for cell splitting, making them computationally lengthy and complex. Furthermore, assigning the split cell groups to specific samples still requires whole-genome sequencing or whole-exome sequencing data from each sample, significantly increasing detection costs. Therefore, there is an urgent need in this technical field for a splitting method with high accuracy and significantly improved computational efficiency.

[0005] Similar to SNPs found in the nuclear genome, the mitochondrial genome also contains numerous variations. Previous studies have validated that mitochondrial genome variations can be used as endogenous barcodes for pedigree analysis. Based on this, we hypothesize that it is feasible to use SNPs found in the mitochondrial genome to separate scRNA-seq data from mixed samples, and therefore developed this algorithm to separate mixed scRNA-seq samples by referencing SNPs found in mitochondrial RNA. Summary of the Invention

[0006] The purpose of this invention is to solve the technical problem of how to obtain a splitting method with high accuracy and significantly improved computational efficiency.

[0007] To address the aforementioned problems, the present invention provides a method for splitting mixed single-cell data based on mitochondrial variation, comprising the following steps:

[0008] Step 1: For samples that require multi-sample fusion sequencing, perform multi-sample fusion single-cell sequencing and individual batch mitochondrial sequencing for each sample.

[0009] Step 2: After quality control and comparison of sequencing data, extract the frequency matrix of mitochondrial variation sites in single cells and each sample;

[0010] Step 3: Screen for highly variable sites with significant differences in the frequency matrix, calculate the correlation coefficient between the frequency of highly variable sites for each cell and each sample, and take the sample with the largest correlation coefficient value and greater than 0 in each single cell as the sample label for that cell.

[0011] Step 4: Using a semi-supervised machine learning model, the correlation coefficients of inter-cell site frequencies in the single-cell mitochondrial variation site frequency matrix are used as edges. Each single cell is treated as a point. Starting from the point with known sample labels, the labels are propagated to the points with unknown sample labels through the edges until the model stops, and the predicted label for each single cell is output.

[0012] Preferably, the specific steps in step 1 above are as follows: first, collect multiple samples that need to be mixed for single-cell sequencing, and divide the single cells to be sequenced in each sample into two groups: one group is used for batch mitochondrial transcriptome sequencing of each sample independently, and the other group is used for single-cell transcriptome sequencing of multiple samples.

[0013] Preferably, in step 1 above, the sample-independent batch mitochondrial transcriptome sequencing needs to be performed on each sample individually. The mitochondria of a single sample are isolated using a cell mitochondrial isolation kit, and then RNA is extracted and purified from the isolated mitochondria using a total DNA / RNA isolation kit. Subsequently, library preparation, reverse transcription, and amplification for batch mitochondrial transcriptome sequencing are performed.

[0014] Preferably, in step 1 above, single-cell transcriptome sequencing of multiple mixed samples requires first mixing the cell samples together, using a chip manufacturing, barcode bead preparation, and sequencing library preparation process built using Well-Paired-Seq or other single-cell sequencing platforms, and then sequencing the library on a sequencing instrument.

[0015] Preferably, the quality control and alignment of sequencing data in step 2 above specifically involves: during sequence alignment, using sequence alignment software CellRanger or zUMIs to align single-cell sequencing data to a reference genome; using quality control software cutadapt to perform quality filtering on batch RNA-seq data, removing short reads containing known artificial adapter sequences and short reads with more than 30 consecutive A, T, or G bases; and using sequence alignment software STAR to align the batch sequencing data of each sample to a mitochondrial reference genome; in order to improve alignment quality and reduce false positives at identified variant sites. The positive rate was determined by using the sequence alignment software Minimap2 to extract short reads from the obtained sequence alignment files, re-aligning them to the reference genome to obtain re-aligned files for single cells and samples. For single-cell re-aligned files, only alignment results with both base quality and alignment quality higher than 20 were retained. For sample re-aligned files, only alignment results with both base quality and alignment quality higher than 10 were retained. Based on the alignment results, the sequencing depth of each base at each mitochondrial locus for each cell or sample was extracted, the variation frequency of the locus was calculated, and the results were integrated to generate a mitochondrial variation site frequency matrix for single cells and samples.

[0016] Preferably, in step 3 above, the screening of highly variable sites with significant differences is specifically as follows: for the frequency matrix of mitochondrial variant sites generated by the sample, the highly_variable_genes function of the single-cell analysis Python package scanpy is used, with the calculation parameters set as flavor = 'seurat_v3', span = 0.3, n_top_genes = 2000, and the obtained mitochondrial variants are taken as highly variable sites.

[0017] Preferably, the correlation coefficient in step 3 above is specifically the Pearson correlation coefficient.

[0018] Preferably, the semi-supervised machine learning model in step 4 above is: the LabelPropagation Algorithm machine learning model, which uses a Gaussian kernel function (rbf) with a clamping factor of 0.8 and 100 iterations.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. This invention uses mitochondrial variation as an endogenous barcode, resulting in high accuracy of mixed sample separation.

[0021] 2. This invention reduces the reliance on internal barcode data sources for mixed sample splitting, significantly improving computational efficiency;

[0022] 3. This invention utilizes mitochondrial sequencing instead of whole genome sequencing or whole exome sequencing, thereby reducing analysis costs. Attached Figure Description

[0023] Figure 1 The flowchart illustrates a method for splitting mixed single-cell data based on mitochondrial variation in a multi-sample mixture, as provided in this embodiment of the invention.

[0024] Figure 2 This is a schematic diagram illustrating the computational principle of a method for splitting mixed single-cell data based on mitochondrial variation in multi-sample mixing according to the present invention.

[0025] Figure 3 This is a computational performance comparison chart of a method for splitting mixed single-cell data based on mitochondrial variation in multi-sample mixing according to the present invention. Detailed Implementation

[0026] To make the present invention more apparent and understandable, preferred embodiments are described in detail below with reference to the accompanying drawings:

[0027] This invention provides a method for splitting mixed single-cell data based on mitochondrial variation:

[0028] 1. For samples that require multi-sample fusion sequencing, perform multi-sample fusion single-cell sequencing and batch mitochondrial sequencing for each sample separately.

[0029] Specifically: such as Figure 1 Step 1: Sample Sequencing. First, collect multiple samples that require mixed single-cell sequencing. Divide the single cells to be sequenced in each sample into two groups: one group is used for independent batch mitochondrial transcriptome sequencing of each sample, and the other group is used for mixed single-cell transcriptome sequencing of multiple samples.

[0030] In the case of sample-independent batch mitochondrial transcriptome sequencing, each sample needs to be performed individually. The mitochondria of a single sample are isolated using a cell mitochondrial isolation kit, and then RNA is extracted and purified from the isolated mitochondria using a total DNA / RNA isolation kit. Subsequently, library preparation, reverse transcription, and amplification are performed for batch mitochondrial transcriptome sequencing.

[0031] Among them, single-cell transcriptome sequencing with multiple samples requires first mixing the cell samples together, using a process of chip manufacturing, barcode bead preparation, and sequencing library preparation built using Well-Paired-Seq or other single-cell sequencing platforms, and then sequencing the library on a sequencing instrument.

[0032] 2. After quality control and comparison of sequencing data, the frequency matrix of mitochondrial variation sites in single cells and each sample was extracted.

[0033] Specifically: such as Figure 1 Step 2: Sequence alignment and variant identification.

[0034] During sequence alignment, CellRanger or zUMIs sequence alignment software was used to align single-cell sequencing data to a reference genome. Cutadapt quality control software was used to filter batch RNA-seq data, removing short reads containing known artificial adapter sequences and short reads exceeding 30 consecutive A, T, or G bases. STAR sequence alignment software was used to align the batch sequencing data of each sample to a mitochondrial reference genome. To improve alignment quality and reduce the false positive rate of identified variant sites, Minimap2 sequence alignment software was used to extract short reads from cells or samples from the obtained sequence alignment files and re-align them to the reference genome, obtaining re-aligned files of single cells and samples.

[0035] For single-cell re-alignment files, only alignment results with both base quality and alignment quality higher than 20 are retained;

[0036] For the sample re-alignment files, only the alignment results with both base quality and alignment quality higher than 10 are retained.

[0037] Based on the alignment results, the sequencing depth of each base at each mitochondrial locus of each cell or sample is extracted, the mutation frequency of the locus is calculated, and the mitochondrial mutation site frequency matrix of single cells and samples is integrated to generate a matrix.

[0038] 3. Screen for highly variable sites with significant differences between samples, and determine the cell sample label based on the correlation between single cells and highly variable sites between samples.

[0039] Specifically: such as Figure 1Step 3: Single-cell-sample correlation calculation. For the generated mitochondrial variant frequency matrix of the samples, the `highly_variable_genes` function of the `scanpy` Python package for single-cell analysis is used, with the calculation parameters set to `flavor = 'seurat_v3', span = 0.3, n_top_genes = 2000`. The obtained mitochondrial variants are considered as highly variable sites, which show significant differences across different samples. The rows containing the highly variable sites in the single-cell and sample variant frequency matrices are extracted. The Pearson correlation coefficient of each single-cell and sample pair at the frequency of highly variable sites is calculated. For each single cell, the sample corresponding to the highest correlation coefficient exceeding 0 is set as the sample label for that cell.

[0040] 4. Determine the sample labels of the remaining cells using a semi-supervised machine learning model.

[0041] Specifically: such as Figure 1 Step 4: Cell Sample Label Determination. Using the correlation coefficients of inter-cell site frequencies in the single-cell mitochondrial variation site frequency matrix as edges, and each single cell as a point, the labels are propagated from points with known sample labels to points with unknown sample labels via these edges. The Label Propagation Algorithm in a semi-supervised machine learning model is used, with a Gaussian kernel function (rbf), a clamping factor of 0.8, and 100 iterations. When the model stops, the predicted label for each single cell is output.

[0042] Figure 3 Datasets 1-3 contain three test datasets: 4 samples with 3,177 single cells, 3 samples with 25,078 single cells, and 10 samples with 60,000 single cells, respectively. Each column displays a mixed-sample splitting algorithm; the first column is the present invention, and the second to fifth columns are other algorithms. Each row displays computational performance: AUC represents computational accuracy, ranging from 0 to 1 (the redder the heatmap and the higher the value, the more accurate the computation); Time (h) represents the total computation time in hours (the bluer the heatmap and the lower the value, the higher the computational efficiency and the less time consumed); MEM (G) represents the peak memory required for computation in gigabytes (the bluer the heatmap and the lower the value, the less memory required); NA indicates that the algorithm crashed due to excessive memory requirements or excessive computation time, resulting in computation failure.

[0043] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any form or substance. It should be noted that those skilled in the art can make various improvements and additions without departing from the present invention, and these improvements and additions should also be considered within the scope of protection of the present invention. Any modifications, alterations, and equivalent changes made by those skilled in the art based on the above-disclosed technical content without departing from the spirit and scope of the present invention are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, and evolutions made to the above embodiments based on the essential technology of the present invention still fall within the scope of the technical solution of the present invention.

Claims

1. A method for mixture deconvolution of multi-sample mixed single-cell data based on mitochondrial variation, characterized in that, Includes the following steps: Step 1: For samples that require multi-sample fusion sequencing, perform multi-sample fusion single-cell sequencing and individual batch mitochondrial sequencing for each sample. Step 2: After quality control and comparison of sequencing data, extract the frequency matrix of mitochondrial variation sites in single cells and each sample; Step 3: Screen for highly variable sites with significant differences, calculate the correlation coefficient of the frequency of highly variable sites between each cell and each sample, and take the sample with the largest correlation coefficient value in each single cell that is greater than 0 as the sample label of that cell. Step 4: Using a semi-supervised machine learning model, the correlation coefficients of inter-cell site frequencies in the single-cell mitochondrial variation site frequency matrix are used as edges. Each single cell is treated as a point. Starting from the point with known sample labels, the labels are propagated to the points with unknown sample labels through the edges until the model stops, and the predicted label for each single cell is output.

2. The method of claim 1, wherein, The specific steps in step 1 are as follows: First, collect multiple samples that need to be mixed for single-cell sequencing. Divide the single cells to be sequenced in each sample into two groups: one group is used for batch mitochondrial transcriptome sequencing of each sample independently, and the other group is used for single-cell transcriptome sequencing of multiple samples.

3. The method of claim 2, wherein, In step 1, sample-independent batch mitochondrial transcriptome sequencing requires each sample to be performed individually. Using a cell mitochondrial isolation kit, mitochondria of individual samples are isolated. Then, RNA is extracted and purified from the isolated mitochondria using a total DNA / RNA isolation kit. Subsequently, library preparation, reverse transcription, and amplification are performed for batch mitochondrial transcriptome sequencing.

4. The method of claim 2, wherein, In step 1, single-cell transcriptome sequencing of multiple mixed samples requires first mixing the cell samples together, using a process of chip manufacturing, barcode bead preparation, and sequencing library preparation built using Well-Paired-Seq or other single-cell sequencing platforms, and then sequencing the library on a sequencing instrument.

5. The method for mixing and splitting single-cell data based on mitochondrial variation according to claim 1, characterized in that, Step 2, specifically the quality control and alignment of sequencing data, involves: during sequence alignment, using sequence alignment software CellRanger or zUMIs to align single-cell sequencing data to a reference genome; using quality control software CutAdapt to perform quality filtering on batch RNA-seq data, removing short reads containing known artificial adapter sequences and short reads exceeding 30 consecutive A, T, or G bases; using sequence alignment software STAR to align the batch sequencing data of each sample to a mitochondrial reference genome; and using sequence alignment software Mi... Nimap2 extracts short reads from cells or samples from the obtained sequence alignment files, re-aligns them to the reference genome, and obtains re-aligned files for single cells and samples. For single-cell re-aligned files, only alignment results with both base quality and alignment quality higher than 20 are retained. For sample re-aligned files, only alignment results with both base quality and alignment quality higher than 10 are retained. Based on the alignment results, the sequencing depth of each base at each mitochondrial locus of each cell or sample is extracted, the variation frequency of the locus is calculated, and the results are integrated to generate a mitochondrial variation site frequency matrix for single cells and samples.

6. The method for splitting mixed single-cell data based on mitochondrial variation according to claim 1, characterized in that, In step 3, highly variable sites with significant differences are selected. Specifically, for the mitochondrial variant site frequency matrix of the generated samples, the highly_variable_genes function of the single-cell analysis Python package scanpy is used, with the calculation parameters set as flavor = 'seurat_v3', span = 0.3, n_top_genes = 2000, and the obtained mitochondrial variants are taken as highly variable sites.

7. The method for splitting mixed single-cell data based on mitochondrial variation according to claim 1, characterized in that, The correlation coefficient in step 3 is specifically the Pearson correlation coefficient.

8. The method for splitting mixed single-cell data based on mitochondrial variation according to claim 1, characterized in that, The semi-supervised machine learning model in step 4 is specifically the LabelPropagation Algorithm machine learning model, which uses a Gaussian kernel function rbf, a clamping factor of 0.8, and 100 iterations.

Images