A method for single-cell sequencing data sample provenance based on genotype dosage

Abstract

The application discloses a method for realizing single cell sequencing data sample tracing based on genotypic dose, comprising the following steps: obtaining the genotypic dose of common SNPs of a sample through lcWGS genotypic interpolation; preparing a single cell sequencing library of a mixed sample and obtaining scRNA-seq and scDNA-seq data through high-throughput sequencing; splitting each cell sequencing data according to a cell barcode and obtaining the genotypic dose of common SNPs of a single cell through variation detection; and taking the genotypic dose of SNPs as a feature to calculate the minimum distance between a cell to be predicted and the lcWGS sample through a supervised machine learning model K nearest neighbor algorithm to predict the sample source of the cell. The application has the characteristics of simple operation and high operation efficiency, and can effectively reduce the scRNA-seq and scDNA-seq cost of a single sample. The application has a single cell sample source prediction accuracy of more than 99.99%, and a mixed cell recognition accuracy of more than 94%, and compared with other genotypic tracing tools, the result is more accurate.

Description

Technical Field

[0001] This invention belongs to the field of single-cell transcriptomics and genome sequencing analysis, specifically involving a method for tracing the source of single-cell sequencing data samples based on genotype dosage. Background Technology

[0002] Single-cell transcriptome sequencing (scRNA-seq) and single-cell genome sequencing (scDNA-seq) are techniques for analyzing transcriptome and genome characteristics at the single-cell level, characterized by high resolution and high throughput. Currently, the mainstream single-cell sequencing technologies include 10x Genomics... TM This platform is the primary platform for testing. It requires that the samples be pre-prepared as single-cell suspensions. The platform automates the process of encapsulating single cells in microdroplets and constructing independent RNA reverse transcription or DNA fragmentation reaction spaces. Simultaneously, it adds molecular barcodes to the RNA or DNA molecules in each microdroplet reaction to achieve single-cell and single-molecule data labeling, ultimately constructing scRNA-seq or scDNA-seq libraries. After sequencing, these libraries can yield 10,000 (10x Genomics Chromium) or 20,000 (10x Genomics Chromium) single cells, effectively covering multiple cell types.

[0003] Based on the aforementioned advantages, single-cell sequencing technology is widely used in research fields such as transcriptome mapping, identification of new cell subpopulations, analysis of tumor microenvironment heterogeneity, understanding of viral infection processes, and exploration of mechanisms regulating organismal growth and development. However, standard microfluidic-based single-cell sequencing can only label RNA or DNA molecules in each cell, not sample information. Therefore, each single-cell sequencing reaction can theoretically only analyze one sample. The application cost of this technology is currently high, approximately 20,000 RMB per sample, which greatly limits its large-scale application. This is also the main reason why the sample size in currently reported single-cell transcriptome or genome sequencing studies is generally small. Therefore, mixing multiple samples in a single reaction for single-cell sequencing has become an application strategy to reduce the cost per sample, but the technical challenge of distinguishing the source of each cell sample (i.e., tracing its origin) remains.

[0004] Currently, various techniques have been developed to address the challenge of tracing the origin of each cell sample when mixing multiple samples for single-cell sequencing in a single reaction. These methods can be broadly categorized into two types based on whether additional reactants are added during mixing. One type involves adding sample tags, where a specific DNA sequence tag is added to the single-cell suspension prepared from each sample. These DNA sequences are conjugated to antibodies or other vectors, either specifically (DNA-conjugated antibodies, such as Biolegend's Hashtags technology) or non-specific (DNA-conjugated non-antibody molecules, 10×Genomics...). TM Labeling methods involve attaching small lipid molecules to the cell membrane, amplifying them in microfluidically generated droplets, and imbuing them with cell barcode molecules. This barcode is then detected in the sequencing data to identify the sample. However, these methods, which involve adding exogenous reactants, typically require cell suspensions with extremely high cell viability and low levels of impurities and cell debris. While effective in labeling peripheral blood leukocytes and cell lines, they are less effective in labeling tissue samples, resulting in high negative cell rates and high costs. Furthermore, sample processing can lead to the death of specific cell types, significantly impacting single-cell sequencing results. Another type of single-cell sample tracing strategy is label-free methods, which primarily rely on genotype information from different cells in mixed sample data for cell kinship identification. For example, Souporcell (a method for single-cell clustering and genotype inference in scRNA-seq data without a reference genome) calculates the genotype likelihood of each cell and uses unsupervised machine learning to cluster all cells, thus identifying the cell group. However, while this algorithm demonstrates high discriminative power with fewer than five pooled samples, its classification performance deteriorates drastically with pools exceeding five. In simulated scRNA-seq data with a pool of 10 samples, the misclassification rate approached 10%. More critically, although the tool can distinguish cells from different individuals within the pooled sample data, it cannot match the classified samples with the actual samples. Therefore, this technology is difficult to widely implement in practical pooled sample applications.

[0005] Besides the issue of sample origination, single-cell multi-sample pooled sequencing often involves sequencing excessive numbers of cells to achieve higher cost-effectiveness. This leads to a high rate of heterologous doublets, where, after sample suspension preparation and mixing, a single droplet microreaction system encapsulates more than two cells from different samples. Doublet generation interferes with subsequent single-cell transcriptome and genomic analyses, especially when the cell count in a single reaction is large, where the doublet proportion can exceed 10%. Therefore, efficiently and accurately identifying heterologous doublets is another challenge faced by pooled scRNA-seq or scDNA-seq technologies. For example, data from Souporcell shows poor accuracy in identifying heterologous doublets in multi-sample pooled datasets, resulting in too many cells being mislabeled as doublets and subsequently removed, leading to a significant loss of valuable data.

[0006] In summary, while label-free pooled single-cell sequencing methods offer advantages such as low cost and relatively simple operation, the performance of existing methods for sample provenance tracing and allogeneic doublet identification still needs optimization. Improving the accuracy and computational efficiency of label-free sample provenance tracing, especially methods based on individual genotypes, will help reduce costs and promote the application of single-cell sequencing technology in large-scale samples. Summary of the Invention

[0007] The purpose of this invention is to address the above-mentioned technical problems by providing a method for tracing the source of multi-sample mixed single-cell sequencing data that is simple to operate, highly accurate, and low in cost.

[0008] To achieve the above-mentioned objectives, this invention provides a method for tracing the source of single-cell sequencing data samples based on genotype dosage, comprising the following steps:

[0009] S1. Genotypic doses of common SNPs in the sample were obtained by lcWGS genotyping;

[0010] S2. Prepare single-cell sequencing libraries of mixed samples and obtain scRNA-seq and scDNA-seq data through high-throughput sequencing;

[0011] S3. Sequencing data for each cell is split based on cell barcodes, and genotype doses of common SNPs in individual cells are obtained through variant detection;

[0012] S4. Using the genotype dose of SNPs as a feature, the minimum distance between the cell to be predicted and the lcWGS sample is calculated using the supervised machine learning model K-nearest neighbor algorithm to predict the sample origin of the cell.

[0013] As a preferred implementation, lcWGS data alignment is performed using the bwa mem algorithm with GRCh38 as the reference genome, resulting in a BAM file.

[0014] As a preferred implementation, the lcWGS genotype imputation operation takes the PL matrix (vcf) format of the reference sample as input, uses 1000 Genomes Project data as the haplotype reference template, performs imputation according to the default parameters of GLIMPSE, performs strict quality control on the obtained genotypes, retains SNPs with MAF>0.01 and INFO value>0.9, and extracts the genotype doses of these SNPs as reference data values ​​for the KNN model.

[0015] As a preferred implementation, single-cell sequencing data alignment is performed using Seurat or other tools, with GRCh38 as the reference genome, to obtain BAM files. Based on cell barcodes, the data is split to obtain BAM files for individual cells.

[0016] As a preferred implementation, the genotype similarity of the BAM file is calculated using the default parameters of bcftools mpileup, and the PL value in the generated vcf file is the genotype similarity value of the SNP; for lcWGS samples, the PL matrix will be imputed; for single-cell samples, the PL value is converted using the bcftools dosage module to obtain the genotype dosage value of common SNPs covered by the cell.

[0017] As a preferred implementation, sample source prediction for individual cells is performed using a KNN model. The genotype dose of common SNPs in each cell is intersected with the dose of reference samples. This intersection data is used as a feature value of the model to calculate the distance from the cell to be predicted to all reference samples. The cell with the smallest distance is considered to be the sample source.

[0018] Compared with the prior art, the method of the present invention has the following advantages:

[0019] (1) The present invention can significantly reduce the cost of single-cell transcription and DNA sequencing of a single sample, and the cost of lcWGS for the additional expenditure per sample is extremely low.

[0020] (2) The main computational steps of this invention all implement multi-threaded and distributed computing, which significantly improves the computational efficiency of genotype similarity.

[0021] (3) The present invention achieves a traceability accuracy of over 99% for single cells and an accuracy of over 94% for identifying heterologous doublets.

[0022] (4) By comparing single-cell and lcWGS sequencing data, this invention can directly identify the specific sample to which a single cell belongs, without the need to add additional tags to the single-cell reaction system to achieve single-cell sample source identification. Attached Figure Description

[0023] Figure 1 This invention illustrates the analytical process for tracing the origin of single cells after single-cell pooled library construction and sequencing.

[0024] Figure 2 The results of the analysis of a simulated single-cell sequencing dataset containing 10 samples and the key factors affecting the accuracy of allogeneic doublet detection are shown. (A) Correlation between allogeneic doublet detection accuracy and the INFO value selected by lcWGS to fill SNPs. (B) Correlation between allogeneic doublet detection accuracy and the read ratio of the two cell sequences in the mixture. (C) Distribution of the second relative genetic distance between allogeneic doublets and singlets. (D) Principal component analysis (PCA) results. Detailed Implementation

[0025] The present invention will be further described below with reference to specific embodiments and accompanying drawings. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0026] Figure 1 This invention illustrates the analytical process for tracing the origin of single cells after single-cell pooled library construction and sequencing.

[0027] This invention provides a method for tracing the origin of samples and identifying allogeneic doublets in mixed scRNA-seq or scDNA-seq cells based on genotype dosage. The method includes the following steps:

[0028] (1) Genotype doses of common SNPs in the sample were obtained by lcWGS genotype imputation;

[0029] (2) Prepare single-cell sequencing libraries of mixed samples and obtain scRNA-seq and scDNA-seq data through high-throughput sequencing;

[0030] (3) Sequencing data of each cell is split according to cell barcode and genotype dose of common SNPs in a single cell is obtained by mutation detection;

[0031] (4) Using the SNP genotype dose as a feature, the minimum distance between the cell to be predicted and the lcWGS sample is calculated by the supervised machine learning model K-nearest neighbor algorithm to predict the sample source of the cell.

[0032] The specific implementation process is as follows.

[0033] The lcWGS data alignment was performed using the bwa mem algorithm, with the GRCh38 genome as the reference genome, resulting in a BAM file.

[0034] Single-cell sequencing data alignment was performed using Seurat or other tools, with GRCh38 as the reference genome, resulting in BAM files. Based on cell barcodes, the data was split to obtain individual cell BAM files.

[0035] Genotype likelihood (PL) values ​​in the BAM file are calculated using the default parameters of bcftools mpileup. The PL values ​​in the resulting VCF file represent the genotype similarity values ​​of the SNPs. For lcWGS samples, the PL matrix undergoes genotype imputation. For single-cell samples, the PL values ​​are converted using the bcftools dosage module to obtain the genotype dosage values ​​of common SNPs covered by the cell.

[0036] The lcWGS genotyping operation takes the PL matrix (vcf) format of the reference sample as input, uses 1000 Genomes Project data as the haplotype reference template, and performs imputation according to the default parameters of GLIMPSE. The obtained genotypes are strictly quality-controlled, retaining SNPs with MAF > 0.01 and INFO value > 0.9, and extracting the genotype doses of these SNPs as reference data values ​​for the KNN model.

[0037] Sample origin prediction for individual cells is performed using the KNN model from Scikit-learn. The intersection of the genotype dose of common SNPs in each cell and the dose of the reference sample is taken. This intersection data is used as a feature value of the model to calculate the distance from the cell to be predicted to all reference samples. The cell with the smallest distance is regarded as the sample origin.

[0038] The key modules and algorithm designs of this invention include:

[0039] -lcWGS's genotype filling and sample kinship identification module;

[0040] -An algorithm that splits data into individual cells after alignment with scRNA-seq or scDNA-seq;

[0041] - An algorithm for calculating the dose of common SNP genotypes covered by single-cell sequencing data;

[0042] -A supervised machine learning algorithm that calculates the distance between a single cell and an lcWGS sample and performs sample source tracing based on genotype dose as features;

[0043] - A module for recognizing allogeneic doublets generated from single-cell sequencing pool library construction.

[0044] The calculation instructions for each module are as follows:

[0045] The samples used for pooled scRNA-seq or scDNA-seq sequencing library construction were derived from unrelated individuals. All samples were pooled for sequencing library construction, while each sample underwent separate 1.5x whole-genome sequencing (lcWGS). The lcWGS data for each sample were imputed using GLIMPSE software, and the genotype doses of common SNPs were calculated. This genotype dose matrix served as a reference value for distinguishing the individual cell origins. The imputed genotype data was also used for Identity By Descent (IBD) analysis of the pooled samples to rule out interference from related samples in subsequent analyses.

[0046] The scRNA-seq or scDNA-seq data obtained from the pooled sample method were compared using 10x Genomics' Cell Ranger software or other software for single-cell alignment, resulting in a BAM file containing alignment information. The BAM file was further split based on cell barcodes to obtain the alignment sequence for each cell. The algorithm for splitting the single-cell data was implemented using a Python script, which supports multi-threaded parallel computation, offering advantages such as high processing speed and low memory consumption.

[0047] Using the bcftools mpileup software, the dosage values ​​of common SNP genotypes covered by the split individual cell alignment data are calculated. This process treats the dosage estimation of each cell genotype as an independent computational task, supports multi-threaded parallel computation for a large number of single-cell datasets, and supports distributed computation in a cluster, that is, it utilizes multiple computers to complete the computational task concurrently.

[0048] By comparing lcWGS and split scRNA-seq or scDNA-seq single-cell data, the genotype dose values ​​of common SNPs in reference samples and single cells are obtained. Then, the Euclidean distance from each cell to the reference sample is calculated using the genotype dose value as a feature. The sample closest to a single cell is predicted as the sample source for that cell. This algorithm uses the KNN submodule of the Python scientific computing module Scikit-learn for prediction, and the classification calculation process supports multi-threaded operation.

[0049] This invention also includes a computationally efficient and accurate allogeneic doublet identification module. Its core principle is to construct simulated reference samples by pairing and merging any two lcWGS raw sequencing data sets, creating a genetic background similar to that of allogeneic mixed multicellular doublets. A dose matrix of common SNP genotypes from the simulated reference samples is obtained through lcWGS interpolation, and this matrix, together with real individual reference samples, forms a genetic distance reference library for cell origin tracing. All cells predicted to originate from real individuals are considered true single cells (i.e., singlets), while those predicted to be simulated chimeric samples are considered mixed cells (i.e., allogeneic doublets) originating from different individuals during the experiment and should be excluded in subsequent analyses.

[0050] The single-cell sequencing method and sequencing data volume requirements are as follows:

[0051] For samples requiring single-cell library construction, use one sample from the same individual for lcWGS library construction. Prepare single-cell suspensions, mix 2-15 samples according to experimental requirements, and then adjust the concentration using 10x Genomics. TM Follow the instructions on the single-cell library preparation platform to perform microdroplet generation and reverse transcription or DNA fragmentation reactions to complete the construction of scRNA-seq or scDNA-seq libraries.

[0052] The constructed lcWGS, scRNA-seq, or scDNA-seq libraries were subjected to high-throughput sequencing to obtain raw sequencing data. The WGS sequencing data was approximately 4.8 Gb (i.e., approximately 1.5 times the coverage of the human genome), while the scRNA-seq or scDNA-seq data contained at least 0.003 Gb of sequencing data per cell.

[0053] The specific steps of the genotype imputation analysis process for lcWGS are as follows:

[0054] (1) First, low-quality sequences with an average Q30 of less than 80% in the sequencing data were removed using the quality control tool Fastp, and the quality-controlled sequences were aligned to the human reference genome (GRCh38) using bwa to obtain the alignment file in BAM format.

[0055] (2) Select SNPs with minor allele frequency (MAF) greater than 0.01 from the 1000 Genomes Project database. Use the mpileup function of bcftools to detect the genotypic similarity (PL value) of common sites covered by each BAM file. After merging all samples, a PL value matrix (vcf format) is obtained. Removing sites with MAF < 0.01 can greatly reduce the number of SNPs that need to be calculated and improve computational efficiency.

[0056] (3) Imputation of lcWGS: Using the 1000 Genomes Project database as a haplotype reference template, genotype imputation was performed on the lcWGS genotype similarity matrix according to the standard GLIMPSE imputation procedure to obtain genotype information of common variant sites in all samples. The genotype dose of common SNPs in each lcWGS sample was extracted using the dosage function of bcftools. The genotype dose matrix obtained in this step will serve as the reference values ​​for subsequent sample classification.

[0057] (4) Based on lcWGS-based genotyping, after rigorous quality control to retain high-quality SNPs (INFO value greater than 0.9), the genetic analysis tool PLINK was used to calculate the kinship of the mixed samples (IBD analysis). Simultaneously, the samples were sexed based on genotype information. The results of this step will be used for quality control of the mixed data. Kinship analysis can avoid mixing samples with first- or second-degree kinship, while sex determination can verify the accuracy of single-cell tracing.

[0058] The algorithm implementation steps for aligning and splitting scRNA-seq or scDNA-seq data are as follows:

[0059] (1) After removing low-quality sequences with an average Q30 of less than 80% from the sequencing data obtained by pooled sequencing using the quality control tool Fastp, the quality-controlled sequences are aligned using Cell Ranger or other single-cell alignment tools to obtain BAM files. Each line of this file contains alignment information, nucleic acid sequence information, and cell barcode molecular tag information.

[0060] (2) Quality control is performed on the transcriptome or genome data of each cell through single-cell transcriptome or genome sequencing data analysis. The main quality control conditions are: for single-cell transcriptome data, remove low-quality cells or fragments with fewer than 500 genes and fewer than 1000 RNA molecules detected in a single cell; for single-cell DNA sequencing data, remove low-quality cells or fragments with fewer than 200 genes and fewer than 500 DNA fragments detected in a single cell.

[0061] (3) Obtain the molecular barcodes of high-quality cells (10x Genomics analysis tool: CB:Z:), and design a program to support concurrent splitting of single-cell sequencing data based on this. The core algorithm is implemented as follows:

[0062] a) Read in the list of all cell and molecular barcode (CB:Z:) tags and use them as the key list to build the default Python dictionary.

[0063] b) Open the BAM file generated by Cell Ranger or other single-cell alignment software using the samtools view method, and set a fixed number of rows to be read at a time using Python's itertools module (default is 1 million rows).

[0064] c) By calling the multiprocessing module, multiple interval alignment sequences of the BAM file are processed simultaneously in parallel mode, and the sequence of each cell is output to a separate file.

[0065] d) Detect the genotypic dose of common variant sites covered by the sequencing data of each cell using the bcftools mpileup module.

[0066] The implementation of supervised machine learning methods for accurate source tracing of single-cell samples is as follows:

[0067] Using the lcWGS data to impute the genotype dose matrix as a reference, the intersection SNPs (Special Nominal Points) with each cell were obtained by finding the variant sites shared by each cell. The Euclidean distance (considered as genetic distance) between the test cell and all reference samples was calculated using the Scikit-learn KNN algorithm, and the reference sample closest to the test cell was identified as the sample source.

[0068] The model uses genotype dose values ​​as a feature, and the Euclidean distance between cells and reference samples is calculated using the following formula:

[0069]

[0070] Note: n represents the number of SNPs, p1 and q1 represent the genotype dose of a single cell and reference sample located at the first SNP, p2 and q2 represent the genotype dose of a single cell and reference sample located at the second SNP, and so on.

[0071] The implementation steps of the heterogeneous doublets identification module algorithm are as follows:

[0072] Allogeneic doublets are chimeras formed when two or more cells from different sample sources are encapsulated by microdroplets and labeled as single cells, creating a genetic background with different samples. They interfere with downstream single-cell sequencing analysis. Therefore, identifying and removing allogeneic doublets is crucial for subsequent analysis.

[0073] Since most doublets (cells from two different samples) are chimeric cells generated when single-cell sequencing is performed using multi-sample pooling, this invention combines lcWGS data pairwise to simulate lcWGS sequencing data with mixed genetic backgrounds from two samples. For the simulated reference samples, GLIMPSE is used for genotyping to obtain genotype dose matrices for common variant sites. Taking a pool of 10 samples as an example, 45 simulated chimeric reference samples are needed, resulting in a total of 55 reference samples. The distance between the test cell and each reference sample is calculated using a KNN model; when the minimum distance falls within a simulated reference sample, the cell is identified as a heterologous doublet.

[0074] Simulated allogeneic doublets were constructed by randomly combining single-cell data from two different samples. The results showed that the detection accuracy of allogeneic doublets was closely related to the INFO value selected for SNP filling by lcWGS and the read ratio of the two mixed cell sequencing samples. Figure 2 (A, B). Since adjacent SNPs in the genome may exhibit linkage disequilibrium, redundancy may exist in the SNPs used to calculate genetic distance. To assess the impact of linked SNPs on the accuracy of sample tracing, we performed pruning on the SNPs included in the KNN model. Test results show that as the INFO value increases, the impact of LD-pruning on the accuracy of identifying allogeneic doublets gradually decreases. Figure 2 Another analysis found that when the SNP selection INFO value of the reference sample was greater than 0.9, the accuracy of identifying allogeneic doublets with a two-cell reads ratio in the range of 0.4-0.5 exceeded 90%. Figure 2 (B) In addition, this invention randomly selected cells with read counts between 4000 and 6000 to construct 1981 allogeneic doublets. The algorithm accurately identified 1875 of these allogeneic doublets, achieving an accuracy rate of 94.65%. Furthermore, after adding chimeric simulated reference samples, only 124 cells out of 38268 single cells were misidentified as doublets, with the remaining cells being accurately identified. Therefore, the data loss caused by a single cell being misidentified as an allogeneic doublet by this invention is less than 0.32%, far superior to other similar software.

[0075] By comparing the second smallest relative genetic distance (second distance to minimum distance) between allogeneic doublets and singlets using the algorithm for calculating genetic distance based on SNP genotype dosage proposed in this invention, significant differences in the distribution of the second relative genetic distance between the two groups of cells can be observed, further confirming that the algorithm has extremely high accuracy in distinguishing allogeneic doublets. Figure 2 Further principal component analysis (PCA) of the calculated genetic distances showed that samples predicted as singlets were clearly distinguishable at the PC1-PC2 level. Figure 2 D).

[0076] This invention tested the aforementioned algorithm by constructing a simulated scRNA-seq dataset (containing 38,268 cells) with 10 mixed samples. The results showed that only 2 cells had sample origination errors (i.e., when allogeneic doublets are not considered, the method's accuracy in predicting cell sample origin is 99.99%). In contrast, the Souporcell tool's accuracy in predicting cell sample origin in the same simulated data is less than 90%, and nearly 10% of cells cannot be clearly classified into specific samples and are discarded. In other words, Souporcell's sample identification is not only less accurate than the method disclosed in this invention, but also results in a significant loss of single-cell sequencing data. More importantly, Souporcell does not use the genotype information of the reference sample. Although its clustering method can distinguish the origin of cells from different samples, it cannot ultimately determine which specific sample each group of cells belongs to, thus failing to achieve the purpose of individual cell sample origination.

[0077] This invention verifies the source tracing results by merging the cells predicted by the KNN classifier according to their respective samples, using GLIMPSE to impute the merged dataset, and performing a kinship analysis between the obtained genotypes and the lcWGS imputed results. The results show that the KNN model predicted samples and lcWGS samples have an IBD ratio of over 95%, and are considered to be the same individual, indicating that the sample source tracing algorithm proposed in this invention has extremely high accuracy.

[0078] In summary, this invention provides a method for cell sample tracing and quality control using mixed scRNA-seq or scDNA-seq samples, with advantages including:

[0079] (1) The present invention can significantly reduce the cost of single-cell transcription and DNA sequencing of a single sample, and the cost of lcWGS for the additional expenditure per sample is extremely low.

[0080] (2) The main computational steps of this invention all implement multi-threaded and distributed computing, which significantly improves the computational efficiency of genotype similarity.

[0081] (3) The present invention achieves a traceability accuracy of over 99% for single cells and an accuracy of over 94% for identifying heterologous doublets.

[0082] (4) By comparing single-cell and lcWGS sequencing data, this invention can directly identify the specific sample to which a single cell belongs, without the need to add additional tags to the single-cell reaction system to achieve single-cell sample source identification.

Claims

1. A method for tracing the origin of single-cell sequencing data samples based on genotype dosage, characterized in that... Includes the following steps: S1. Genotypic doses of common SNPs in the sample were obtained by lcWGS genotyping; S2. Prepare single-cell sequencing libraries of mixed samples and obtain scRNA-seq and scDNA-seq data through high-throughput sequencing; S3. Sequencing data for each cell is split based on cell barcodes, and genotype doses of common SNPs in individual cells are obtained through variant detection; S4. Using the genotype dose of SNPs as a feature, the minimum distance between the cell to be predicted and the lcWGS sample is calculated using the supervised machine learning model K-nearest neighbor algorithm to predict the sample origin of the cell.

2. The method according to claim 1, characterized in that, The lcWGS data alignment was performed using the bwa mem algorithm, with the GRCh38 genome as the reference genome, resulting in a BAM file.

3. The method according to claim 1, characterized in that, The lcWGS genotyping operation takes the PL matrix vcf format of the reference sample as input, uses 1000 Genomes Project data as the haplotype reference template, performs imputation according to the default parameters of GLIMPSE, and performs strict quality control on the obtained genotypes, retains SNPs with MAF>0.01 and INFO value>0.9, and extracts the genotype doses of these SNPs as reference data values ​​for the KNN model.

4. The method according to claim 1, characterized in that, Single-cell sequencing data alignment was performed using Seurat or other tools, with GRCh38 as the reference genome, to obtain BAM files. Based on cell barcodes, the data was split to obtain BAM files for individual cells.

5. The method according to claim 1, characterized in that, Genotype similarity in BAM files is calculated using the default parameters of bcftoolsmpileup, and the PL values ​​in the resulting VCF file are the genotype similarity values ​​of the SNPs. For lcWGS samples, the PL matrix will be imputed with genotypes. For single-cell samples, the PL values ​​are converted using the bcftools dosage module to obtain the genotype dosage values ​​of common SNPs covered by the cells.

6. The method according to claim 1, characterized in that, To predict the sample origin of a single cell, a KNN model is used. The intersection of the genotype dose of common SNPs of each cell and the dose of the reference sample is taken. This intersection data is used as the feature value of the model to calculate the distance from the cell to be predicted to all reference samples. The cell with the smallest distance is regarded as the sample origin of the cell.

Images