CoChat
Sign Up
Cited by 515Single-cell and spatial transcriptomicsGenomics and Phylogenetic StudiesGene expression and cancer classification

Souporcell: robust clustering of single-cell RNA-seq data by genotype without reference genotypes

Open access
Heaton, Haynes, Talman, Arthur M., Knights, Andrew, Imaz, Maria, Gaffney, Daniel J., Durbin, Richard, Hemberg, Martin, Lawniczak, Mara K. N.·2020·Nature Methods
DOI 10.1038/s41592-020-0820-1PMC PMC7617080
Open this paper in CoChat — free account

Abstract

Methods to deconvolve single-cell RNA-sequencing (scRNA-seq) data are necessary for samples containing a mixture of genotypes, whether they are natural or experimentally combined. Multiplexing across donors is a popular experimental design that can avoid batch effects, reduce costs and improve doublet detection. By using variants detected in scRNA-seq reads, it is possible to assign cells to their donor of origin and identify cross-genotype doublets that may have highly similar transcriptional profiles, precluding detection by transcriptional profile. More subtle cross-genotype variant contamination can be used to estimate the amount of ambient RNA. Ambient RNA is caused by cell lysis before droplet partitioning and is an important confounder of scRNA-seq analysis. Here we develop souporcell, a method to cluster cells using the genetic variants detected within the scRNA-seq reads. We show that it achieves high accuracy on genotype clustering, doublet detection and ambient RNA estimation, as demonstrated across a range of challenging scenarios.

AI snapshot

Written from the paper’s full text.

TL;DR

souporcell clusters single-cell RNA-seq cells by genotype without reference genotypes, using variants detected directly in the reads. It jointly calls genotype clusters, cross-genotype doublets, and ambient RNA, and the paper reports that it outperforms demuxlet, vireo, and scSplit across synthetic, maternal/fetal, and *Plasmodium falciparum* benchmarks.

Key findings

  • On a synthetic mixture of five human iPSC lines, souporcell assigned 6612/6622 singletons and 415/451 doublets correctly; four singletons were falsely labeled as doublets, 35 doublets were misidentified as singletons, and one doublet and four singletons were unassigned.
  • For doublet detection on a synthetic mixture with 6% doublets and 10% ambient RNA, ROC AUC was 0.98 for souporcell and 0.91 for vireo; demuxlet’s posterior doublet probability was reported as poorly calibrated, with nearly 40% false positive doublets at its default threshold.
  • Cell assignment versus demuxlet on three experimental iPSC mixture replicates gave Adjusted Rand Index 1.0 for souporcell and vireo and an average of 0.97 for scSplit.
  • Across ambient RNA from 2.5%–50%, souporcell and vireo retained high accuracy, with souporcell more robust at calling doublets in high-ambient-RNA cases; all tools performed well down to a minority cluster of 1.2% (50 cells), but only souporcell and vireo correctly identified all minority sample singletons down to 0.6% of all cells.
  • In maternal/fetal samples, souporcell clusters agreed with demuxlet and segregated with expected cell types, while vireo and scSplit had major discordances; in *P. falciparum* datasets, souporcell had superior ARI across the board, with vireo performing poorly on one dataset (ARI versus demuxlet 0.24).

Methods

  • Reads are remapped with minimap2 rather than using STAR-aligned BAMs directly, putative SNPs are called with freebayes, and per-cell allele counts are obtained with vartrix.
  • Clustering uses a sparse mixture model fit by a deterministic annealing variant of expectation maximization, with cluster centers represented as alternate allele fractions; doublets are detected with a beta-binomial model, and genotype plus ambient RNA are co-inferred in STAN by maximum likelihood/gradient descent.
  • Validation used five HipSci human iPSC lines sequenced individually and in mixtures with 10x Chromium, synthetic mixtures with simulated doublets and ambient RNA, maternal/fetal samples from Vento et al., and three *P. falciparum* datasets containing six genetically distinct strains with 1893–2608 cells and median UMIs of ~1000.

Limitations

  • The paper states that limitations include low signal-to-noise due to decreased UMI per cell and high numbers of donors causing increased local maxima.
  • The text also makes evident that genotype accuracy from scRNA-seq is significantly lower than from whole-genome sequencing because most variants have very low coverage per cluster.
  • The paper notes that no good generative models were available for batch effects, allele-specific expression, ambient RNA, and doublets in scRNA-seq, so realistic benchmarking required experimentally generated mixtures rather than fully in silico data.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41592-020-0820-1

Cites

25 of 25 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

4 of 5 cited sources checked · 2 supported · 0 flagged · accuracy not yet measured

  • Findingown data

    Demuxlet significantly overestimates doublets versus expectations based on the number of cells loaded, especially as ambient RNA increases.

    “Demuxlet significantly overestimates doublets versus expectations based on the number of cells loaded11 (Table S2) especially as ambient RNA increases (Fig. 2j).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [11] Zheng, G. X. Y. et al. Massively parallel digital transcriptional profiling of single cells. Nat. Commun. 8, 14049 (2017). source
  • Findingown data

    The default doublet probability threshold for demuxlet gives nearly 40% false positive doublets.

    “The default doublet probability threshold for demuxlet gives nearly 40% false positive doublets.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Souporcell is able to accurately estimate the amount of ambient RNA in the experiment.

    “With these data we also show that souporcell is able to accurately estimate the amount of ambient RNA in the experiment (Fig. 2k).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Allele counts for locus i of each cluster j are drawn from a binomial distribution with an alternative allele fraction of (1-ρ)fij+ρ*ai, where fij is 0, 0.5, or 1 (with a haploid mode limited to 0 and 1), ρ is the amount of ambient RNA, and ai is the average allele fraction in the experiment.
  • Findingown data

    The pattern of souporcell agreement with demuxlet and vireo/scSplit discordance is similar for the other samples tested.

    “This is similar for the other samples tested (Fig. S2, Table S3).”
    Quote in the paperNot a substantial claim
  • Findingown data

    The ARI of scSplit and demuxlet suffered due to poor doublet detection.

    “The ARI of scSplit and demuxlet suffered due to poor doublet detection.”
    Quote in the paperSubstantial claim
  • May combine two claims. 21
  • Classified as probably not a substantial claim. 3

  • May combine two claims.

    Premisecitation

    Vartrix avoids reference bias due to ambiguous support such as alignment end effects and corrects bases with duplicate reads via the UMI.

    “Next, we count alleles per cell with vartrix14(Fig. 1c) which avoids reference bias due to ambiguous support such as alignment end effects and corrects bases with duplicate reads via the UMI.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [14] Petti, A. A. et al. A general approach for detecting expressed mutations in AML cells using single cell RNA-sequencing. Nat. Commun. 10, 3660 (2019). source

      The Methods section (‘Single-cell mutation identification and analysis’; page unavailable) identifies VarTrix as a tool for automated cell-variant assignment. It also describes resolving discordant duplicate reads by UMI, but attributes that procedure to the authors’ Pysam-based workflow, not to VarTrix. Our copy does not say that VarTrix avoids reference bias or alignment-end effects. (LLM, conf=0.85)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationnone given

    If using known variable sites such as from the 1k genomes project or if the genotypes of the donors are known, the remapping process is significantly less important.

    “If using known variable sites such as from the 1k genomes project24 or if the genotypes of the donors are known, the remapping process is significantly less important.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [24] 1000 Genomes Project Consortium. A global reference for human genetic variation. Nature 526, 68–74 (2015). source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    GATK recommendations for variant calling on bulk RNAseq involve removing these regions of the alignments prior to variant calling.

    “This has been noted by others in the past, and GATK recommendations for variant calling on bulk RNAseq involve removing these regions of the alignments prior to variant calling23.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [23] Sahraeian, S. M. E. et al. Gaining comprehensive biological insight into the transcriptome by performing a broad-spectrum RNA-seq analysis. Nat. Commun. 8, 59 (2017). source

      The Results section (‘Variant calling’; page unavailable) says GATK applies RNA-specific filtering steps to handle splicing in RNA-seq data. It does not say that GATK recommends removing the regions in question from alignments before variant calling, so that specific claim is unsupported. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    Analysis of the expression profile of one Plasmodium dataset reveals that the genotypes are distributed across the Plasmodium intra-erythrocytic cycle while being well separated in normalized loss cluster space.

    “Analysis of the expression profile of one of these (see Fig. S3 for the others) reveals that the genotypes are distributed across the Plasmodium intra-erythrocytic cycle (Fig. 3b) while being well separated in normalized loss cluster space (Fig. 3c,d).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the synthetic mixture, souporcell assigned 6612/6622 singletons and 415/451 doublets correctly; four singletons were falsely labeled as a doublet, 35 doublets were misidentified as singletons, and one doublet and four singletons were unassigned.

    “For these data souporcell assigned 6612/6622 singletons and 415/451 doublets correctly; four singletons were falsely labeled as a doublet, 35 doublets were misidentified as singletons, and one doublet and four singletons were unassigned.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    STAR handles spliced reads that do not match the reference well by introducing multiple splice events that force the read to fit the reference in a statistically spurious fashion, causing variant callers to count these reads as supporting the reference allele and introducing reference bias.

    “The first artifact is due to the way STAR handles spliced reads when the read does not match the reference well. STAR will take such a read and introduce multiple splice events that force it to fit the reference in a statistically spurious fashion.”
    • Also stated in: § Supplementary Methods · plain
      “Consequently, for variants in these loci, the variant callers will count these reads as supporting the reference allele, thereby introducing reference bias and noise to the downstream clustering.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    On a synthetic mixture with 6% doublets and 10% ambient RNA, the ROC AUC for doublet calls was 0.98 for souporcell and 0.91 for vireo.

    “we calculated a receiver-operator characteristic (ROC) curve of the doublet calls (Fig. 2i) on a synthetic mixture with 6% doublets and 10% ambient RNA that showed the area under the curve values of 0.98 and 0.91 for souporcell and vireo, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The Adjusted Rand Index of remaining cell assignments versus demuxlet was 1.0 for souporcell and vireo across the three replicates and an average of 0.97 for scSplit.

    “The Adjusted Rand Index (ARI) of the remaining cell assignments versus demuxlet (Table S2) were 1.0 (fully concordant) for souporcell and vireo across the three replicates and an average of 0.97 for scSplit.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Souporcell and vireo retain high accuracy across ambient RNA levels, with souporcell being more robust at accurately calling doublets in high ambient RNA cases.

    “We found that souporcell and vireo retain high accuracy with souporcell being more robust at accurately calling doublets in high ambient RNA cases (Fig. 2k).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Souporcell outperforms vireo and scSplit and surpasses demuxlet on both cell assignment and doublet accuracy.

    “We show that souporcell not only outperforms these new methods, but also surpasses demuxlet on both cell assignment and doublet accuracy.”
    • Also stated in: § Discussion · plain
      “Our benchmarks show that souporcell can outperform all other currently available methods, including those that require genotypes a priori.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Donor of origin is the most significant driver of expression, but donor cells overlap in expression patterns and it is not possible to assign a donor to each cell based solely on expression patterns.

    “Indeed, the most significant driver of expression appears to be the donor of origin, but the donor cells overlap in expression patterns and it is not possible to assign a donor to each cell based solely on expression patterns.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Souporcell is highly accurate for challenging datasets involving closely related maternal/fetal samples, and varying mixtures of Plasmodium falciparum strains.

    “Moreover, souporcell is highly accurate for challenging datasets involving closely related maternal/fetal samples, and varying mixtures of Plasmodium falciparum strains.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Souporcell clusters agree with demuxlet and segregate with the expected cell type clusters, while vireo and scSplit have major discordances with demuxlet.

    “While souporcell clusters agree with demuxlet and segregate with the expected cell type clusters, vireo and scSplit have major discordances with demuxlet.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    All tools performed well down to a minority cell cluster comprising only 1.2% (50 cells) of total cells, but only souporcell and vireo were able to correctly identify all minority sample singletons as their own cluster down to 0.6% of all cells.

    “We found that all tools performed well down to the minority cell cluster comprising only 1.2% (50 cells) of total cells (Fig. 2m), but only souporcell and vireo were able to correctly identify all minority sample singletons as their own cluster down to 0.6% of all cells.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingreasoning

    The most common error mode for vireo and scSplit is calling homozygous reference loci as heterozygous variants, which is expected when ambient RNA is not accounted for.

    “The most common error mode for vireo and scSplit is calling homozygous reference loci as heterozygous variants (Fig. S1j) which is expected when ambient RNA is not accounted for, as it is not in these two tools.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The ARI for each method on the three Plasmodium datasets shows superior performance for souporcell across the board, with scSplit suffering on all datasets and vireo performing poorly on one dataset with an ARI versus demuxlet of 0.24.

    “The ARI for each method (Table S4) on the three Plasmodium data sets show superior performance for souporcell across the board, with scSplit suffering on all datasets and vireo performing poorly on one, which had an ARI versus demuxlet of 0.24.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Demuxlet's posterior doublet probability output did not have enough significant digits and is 1.0 until it starts varying with 27% false positives.

    “Demuxlet’s posterior doublet probability output did not have enough significant digits and is 1.0 until it starts varying with 27% false positives.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    The indel penalty relative to the mismatch penalty is much higher in STAR than other aligners, causing alignments to choose many mismatches over a single or few indels and thus create false positive variants.

    “The final issue is that the indel penalty relative to the mismatch penalty is much higher in STAR than other aligners. This causes the alignments to choose many mismatches over a single or few indels when possible and thus create false positive variants.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Limitations of souporcell include low signal to noise due to decreased UMI per cell and high numbers of donors causing increased local maxima.

    “Limitations of souporcell include low signal to noise due to decreased UMI per cell and high numbers of donors causing increased local maxima.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    Souporcell is robust across a large range of parameters, and more so than any other currently available method.

    “Using more realistic and challenging test cases than previous studies, we show that souporcell is robust across a large range of parameters, and more so than any other currently available method.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The expression PCA and normalized cell-cluster loss PCA of the experimental mixture were similar to the synthetic mixture, indicating that the synthetic mixtures were an accurate approximation of real mixtures.

    “The expression PCA (Fig. 2e) and normalized cell-cluster loss PCA (Fig. 2g,h) of the experimental mixture were similar to the synthetic mixture indicating that the synthetic mixtures were an accurate approximation of real mixtures.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationnone given

    Souporcell clusters cells by genotype, calls doublet-cell barcodes, and infers the amount of ambient RNA in the experiment without the use of a genotype reference.

    “We present souporcell, a method to cluster cells by genotype, call doublet-cell barcodes, and infer the amount of ambient RNA in the experiment without the use of a genotype reference.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    Filtering loci to those with at least n cells supporting each allele provides fairly robust SNPs that have a good chance of aiding the clustering process.

    “This provides us with fairly robust SNPs that have a good chance of aiding the clustering process.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Souporcell enables valuable genotype information to be extracted and ambient RNA estimation at no additional cost.

    “souporcell also enables valuable genotype information to be extracted and ambient RNA estimation at no additional cost.”
    Quote in the paperSubstantial claim
24 of 24

  • The authors state this as an assumption.

    Allele counts for locus i of each cluster j are drawn from a binomial distribution with an alternative allele fraction of (1-ρ)fij+ρ*ai, where fij is 0, 0.5, or 1 (with a haploid mode limited to 0 and 1), ρ is the amount of ambient RNA, and ai is the average allele fraction in the experiment.

    “we assume that the allele counts for locus i of each cluster j are drawn from a binomial distribution with an alternative allele fraction of (1-ρ)fij+ρ*ai, where fij is 0, 0.5, or 1 (with a haploid mode limited to 0 and 1), ρ is a parameter representing the amount of ambient RNA and ai is the average allele fraction in the experiment.”

  • The paper relies on this without support.

    There are no good generative models available for batch effects, allele-specific expression, ambient RNA, and doublets in scRNAseq that can be used to generate in silico data for testing methods that cluster by genotype.

    “Currently, there are no good generative models available for batch effects, allele-specific expression, ambient RNA, and doublets in scRNAseq that can be used to generate in silico data for testing methods that cluster by genotype.”

  • The paper relies on this without support.

    Malaria infections often contain parasites from multiple different genetic backgrounds, and it is not possible to separate the strains prior to sequencing.

    “Malaria infections often contain parasites from multiple different genetic backgrounds, and it is not possible to separate the strains prior to sequencing.”

  • The paper relies on this without support.

    The decidua is primarily composed of maternal cells with some invading fetal trophoblasts, while the placenta is largely composed of cells of fetal origin with the exception of maternal macrophages.

    “The decidua is primarily composed of maternal cells with some invading fetal trophoblasts, while the placenta is largely composed of cells of fetal origin with the exception of maternal macrophages.”

  • The paper relies on this without support.

    In scRNAseq data most variants have very low coverage per cluster compared to what would be generated from WGS data, so genotype accuracy is significantly lower than one would attain with genome sequencing.

    “In scRNAseq data most variants have very low coverage per cluster compared to what would be generated from WGS data, thus the genotype accuracy is significantly lower than one would attain with genome sequencing.”

  • The authors state this as an assumption.

    The majority of barcodes contain a single cell.

    “It is important for some assumptions in this model that the majority of barcodes contain a single cell.”

  • The authors state this as an assumption.

    Only biallelic variants are considered.

    “We will assume only biallelic variants.”

  • The authors state this as an assumption.

    Ploidy is limited to 1 or 2.

    “We assume ploidy is limited to 1 or 2.”

  • The authors state this as an assumption.

    Each cluster is independent and each locus is independent.

    “We treat each cluster as independent and each locus as independent, before marginalizing across the possible genotypes.”

  • The authors state this as an assumption.

    The observed allele fractions are drawn from a binomial distribution with a probability skewed away from p = g/P by the level of ambient RNA ρ.

    “The observed allele fractions are assumed to have been drawn from a binomial distribution with a probability that was skewed away from p = g/P by the level of ambient RNA ρ.”

  • The authors state this as an assumption.

    Allele counts of each cell at each locus are drawn from a beta-binomial distribution from either a single cluster or a pair of clusters.

    “We treat the allele counts of each cell at each locus as random variables drawn from a beta-binomial distribution from either a single cluster or a pair of clusters.”

  • The authors state this as an assumption.

    The expected allele fractions of a doublet coming from cluster i and cluster j is the average of the allele fractions of the two clusters.

    “The expected allele fractions of a doublet coming from cluster i, and cluster j is the average of the allele fractions of the two clusters.”

  • The authors state this as an assumption.

    The beta-binomial is used to model uncertainty in the binomial parameter p.

    “The beta-binomial is used to model our uncertainty in the binomial parameter p.”

  • The authors state this as an assumption.

    Each locus in each cluster comes from one of three genotypes for diploid (0/0, 0/1, 1/1) and two genotypes for haploid (0, 1).

    “The model treats each locus in each cluster as coming from one of three genotypes for diploid (0/0, 0/1, 1/1, here denoted by g = 0, 1, or 2) and two genotypes from haploid (0, 1).”
14 of 14

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Outside of human samples, SNP-based demultiplexing has been applied to demultiplex plasmodium samples and across mouse strains.

    “Outside of human samples, SNP-based demultiplexing has been applied to demultiplex other species: plasmodium samples, and across mouse strains (Heaton et al, 2020; Mylka et al, 2022).”

    From Evaluation of genetic demultiplexing of single-cell sequencing data from model species (2023) · cites this paper as Heaton et al, 2020

    The Validation and Benchmarking section reports using souporcell to separate six genetically distinct Plasmodium falciparum strains, supporting the Plasmodium example. The cited paper does not report demultiplexing across mouse strains, so it supports only one of the two examples in the sentence. (LLM, conf=0.87)

    Experimental — accuracy not yet measured

  • SoupOrCell uses the identification of conflicting genotypes to identify ambient RNA contamination, limiting its application to mixed-genotype experiments.

    “SoupOrCell [21] uses the identification of conflicting genotypes to identify ambient RNA contamination, limiting its application to mixed-genotype experiments.”

    From SoupX removes ambient RNA contamination from droplet-based single-cell RNA sequencing data (2020) · cites this paper as [21]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Reference-allele rate is defined inconsistently

    Minor

    The methods define g as an integer number of reference-allele copies, but also call g the expected reference-allele rate. A later passage gives that rate as g/P, which differs from g when the ploidy is two and g is one or two.

  • Cluster parameter switches between reference and alternate allele fractions

    Minor

    The definition says φ is the fraction of reference alleles, whereas Equation 1 applies φ to the alternate-allele count and (1−φ) to the reference-allele count. Those are opposite definitions of the same parameter.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Heaton, H., Talman, A. M., Knights, A., Imaz, M., Gaffney, D. J., Durbin, R., Hemberg, M., & Lawniczak, M. K. N.
Affiliations
  • Wellcome Sanger Institute (GB)
  • Centre National de la Recherche Scientifique (FR)
  • University of Cambridge (GB)
Corresponding author
  • Heaton, Haynes
  • Hemberg, Martin
  • Lawniczak, Mara K. N.
ORCID
Venue
Nature Methods
Publisher
Springer Science and Business Media LLC
Published
2020-05-04
Volume
17
Issue
6
Pages
615-620
ISSN
  • 1548-7091 (ISSN-L)
  • 1548-7105
DOI
10.1038/s41592-020-0820-1
PMC
PMC7617080
OpenAlex
W3023290106
Cited by
506
References
25
Citations by year
2026: 95 · 2025: 88 · 2024: 126 · 2023: 77 · 2022: 69 · 2021: 38 · 2020: 11 · 2018: 1
FWCI
13.53
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature Methods (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (submittedVersion)
  • HAL (Le Centre pour la Communication Scientifique Directe) (submittedVersion)
  • Apollo (University of Cambridge) (submittedVersion)
  • Apollo
Funders
  • Wellcome Trust (206194/Z/17/Z)
  • British Heart Foundation (RG/13/13/30194; RG/18/13/33946)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene expression and cancer classification
Keywords
single-cell RNA sequencing, genotype clustering, cell lysis, batch effects
MeSH terms
RNA-Seq, Algorithms, Base Sequence, Cell Line, Genotype, Humans, RNA, Sensitivity and Specificity, Software, Cluster Analysis, Polymorphism, Single Nucleotide, Single-Cell Analysis

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-13.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-13.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review