CoChat
Sign Up
Cited by 1,193Single-cell and spatial transcriptomicsGene Regulatory Network AnalysisCell Image Analysis Techniques

Single-Cell Multi-omic Integration Compares and Contrasts Features of Brain Cell Identity

Open access
Welch, Joshua D., Kozareva, Velina, Ferreira, Ashley, Vanderburg, Charles, Martin, Carly, Macosko, Evan Z.·2019·Cell
DOI 10.1016/j.cell.2019.05.006PMC PMC6716797
Open this paper in CoChat — free account

Abstract

Defining cell types requires integrating diverse single-cell measurements from multiple experiments and biological contexts. To flexibly model single-cell datasets, we developed LIGER, an algorithm that delineates shared and dataset-specific features of cell identity. We applied it to four diverse and challenging analyses of human and mouse brain cells. First, we defined region-specific and sexually dimorphic gene expression in the mouse bed nucleus of the stria terminalis. Second, we analyzed expression in the human substantia nigra, comparing cell states in specific donors and relating cell types to those in the mouse. Third, we integrated in situ and single-cell expression data to spatially locate fine subtypes of cells present in the mouse frontal cortex. Finally, we jointly defined mouse cortical cell types using single-cell RNA-seq and DNA methylation profiles, revealing putative mechanisms of cell-type-specific epigenomic regulation. Integrative analyses using LIGER promise to accelerate investigations of cell-type definition, gene regulation, and disease states.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper introduces LIGER, an integrative non-negative matrix factorization method that jointly models shared and dataset-specific features across single-cell datasets. It applies LIGER to mouse and human brain data, spanning scRNA-seq, spatial transcriptomics, and DNA methylation, to define cell types and regulatory relationships.

Key findings

  • LIGER jointly analyzed 204,737 mouse BNST nuclei, identifying 106,728 neurons, 70.2% BNST-localized, 41 transcriptionally distinct BNST-localized neuronal populations, and 85.9% inhibitory cells; it also found cell-type-specific sexual dimorphism, including high dimorphic gene counts in BNSTpr_St18 and BNSTpr_Esr2.
  • In 44,274 human substantia nigra nuclei from seven donors, LIGER identified 24 clusters across known cell classes and detected donor-specific glial signatures, including trauma-associated immediate early genes in donor MD5828 and amyloid-response genes in donor MD5840.
  • Integrating 71,000 Drop-seq frontal cortex profiles with 2,500 STARmap cells spatially located joint populations and increased STARmap resolution from 4 to 7 interneuron clusters and from 2 to 5 glial clusters.
  • Joint analysis of 55,803 scRNA-seq and 3,378 DNA methylation profiles identified 37 clusters, including 11 MGE interneuron populations; methylation-expression analyses found Mecp2 expression correlated with global methylation (ρ = 0.46, p = 0.0039) and Tet3 anticorrelated (ρ = −0.57, p = 0.0002).

Methods

  • LIGER uses integrative non-negative matrix factorization (iNMF) to learn shared and dataset-specific metagenes, followed by shared factor neighborhood graph construction and Louvain community detection for joint clustering.
  • The paper applies LIGER to four analyses: mouse BNST scRNA-seq, human substantia nigra scRNA-seq across donors and mouse comparison, mouse frontal cortex scRNA-seq plus STARmap spatial data, and mouse frontal cortex scRNA-seq plus single-cell DNA methylation.
  • Performance was assessed with alignment and agreement metrics, comparing LIGER to Seurat on published datasets; parameter selection used heuristics based on factor entropy and dataset alignment.

Limitations

  • The paper states that existing multi-omic integration methods, including LIGER, cannot incorporate different types of features, such as gene expression and intergenic methylation, in the definition of cell types.
  • The paper notes that LIGER solves a non-convex optimization problem and can produce different factorizations depending on initialization, and that manual tuning of parameters can sometimes improve results.
  • The paper states that technical factors cannot be completely ruled out in the observed relationship between gene length and methylation-expression correlation; more broadly, the text makes evident that benchmarking focuses mainly on comparisons with Seurat and selected published datasets.

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

Integrity signals

Registry

Crossref · 10.1016/j.cell.2019.05.006

Cites

70 of 70 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 48 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    BNST_Vip and BNSTp_Cplx3 clusters expressed markers of CGE-derived interneurons found in cortex and hippocampus.

    “Two clusters, BNST_Vip and BNSTp_Cplx3, expressed markers of caudal ganglionic eminence (CGE)-derived interneurons found in cortex and hippocampus.”
    Quote in the paperSubstantial claim
  • Findingown data

    Another population aligned to cluster 3, containing striatal eSPNs.

    “A second population of our nuclei aligned to cluster 3, containing the striatal eSPNs we recently described (Saunders et al., 2018).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Saunders et al., 2018 source

      Not checked in the free run.

  • Findingown data

    A fourth population, cluster 4, expressed markers localizing exclusively to the rhomboid nucleus of BNST.

    “A fourth population, cluster 4, expressed markers localizing it exclusively to the rhomboid nucleus of BNST (Figures 3G and 3H).”
    Quote in the paperSubstantial claim
  • Findingown data

    Clustering found 41 transcriptionally distinct populations of BNST-localized neurons.

    “Clustering analysis revealed 41 transcriptionally distinct populations of BNST-localized neurons (Figure 3A).”
    Quote in the paperSubstantial claim
  • Findingown data

    The authors characterized gene expression differences by region and sex in the mouse bed nucleus of the stria terminalis.

    “First, we defined region-specific and sexually dimorphic gene expression in the mouse bed nucleus of the stria terminalis.”
    Quote in the paperSubstantial claim
  • May combine two claims. 30
  • Classified as probably not a substantial claim. 16
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    On human blood scRNA-seq datasets with mainly technical differences, LIGER and Seurat had similarly high alignment, and LIGER's joint clusters matched published assignments.

    “Indeed, LIGER and Seurat show similarly high alignment statistics (Figures 2A-C), and LIGER’s joint clusters match the published cluster assignments for the individual datasets.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For human and mouse pancreatic data, LIGER and Seurat performed similarly, with LIGER slightly higher in alignment.

    “LIGER and Seurat also performed similarly when integrating human and mouse pancreatic data, with LIGER showing slightly higher alignment (Figure 2C).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The authors used scRNA-seq and DNA methylation profiles together to define mouse cortical cell types and suggest mechanisms of epigenomic regulation.

    “Finally, we jointly defined mouse cortical cell types using single-cell RNA-seq and DNA methylation profiles, revealing putative mechanisms of cell-type-specific epigenomic regulation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Heuristics based on factor entropy and dataset alignment generally worked well across analyses, though manual tuning sometimes improved results.

    “Overall, these heuristics performed well across different analyses (Figure S1I-J), though we have observed that manual tuning can sometimes improve the results.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Initial clustering found 106,728 neurons, with 70.2% localized to BNST by ABA marker expression.

    “Initial clustering identified 106,728 neurons, of which 70.2% were localized to BNST by examination of marker expression in the Allen Brain Atlas (ABA) (Lein et al., 2007) (Figure S3B).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Lein et al., 2007 source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    85.9% of cells were inhibitory and 14% excitatory, matching previous estimates.

    “In agreement with previous estimates (Kudo et al., 2012), 85.9 % of the cells were inhibitory (expressing Gad1 and Gad2), while the remaining 14% were excitatory (expressing Slc17a6 (9.4%) or Slc17a8 (4.7%) (Figure S3C).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Kudo et al., 2012 source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    LIGER showed minimal false alignment between hippocampal oligodendrocytes and interneurons and preserved complex internal substructure despite changes in dataset proportion.

    “LIGER generated minimal false alignment between these classes and demonstrated a good preservation of complex internal substructure (Figures 2D-F, S2A-C), even across considerable changes in dataset proportion (Figures 2G-H).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Most high-error genes either showed technical differences between STARmap and Drop-Seq or had no obvious spatial pattern.

    “Most high-error genes either showed technical differences in measurement between STARmap and Drop-Seq (e.g., Aldoc, Tsnax, Hlf) or possessed no obvious spatial pattern (e.g., Elmol, Glul, Scg2) (Figure S5E-G).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    LIGER successfully integrated scRNA-seq and DNA methylation datasets, jointly identifying frontal cortex neuronal cell types and matching published analyses.

    “Indeed, LIGER successfully integrated the datasets, jointly identifying the neuronal cell types of the frontal cortex and according well with the published analyses of each dataset (Figure 6A-C).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: It was unknown whether gene-expression clusters reflect epigenetic distinctions and vice versa.

  • May combine two claims.

    Findingown data

    Canonical marker genes have relatively low overall methylation levels, making cell-type assignment from methylation alone difficult.

    “The canonical marker genes have relatively low overall methylation levels, making it challenging to assign the identity of this cell type from methylation alone.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The larger astrocyte population expressed high Mfge8 and Htral; the second expressed high Gfap.

    “The larger population expressed high levels of Mfge8 and Htral, while the second population showed high expression of Gfap (Figure 5F).”
    Quote in the paperNot a substantial claim

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

    Findingown data

    One STARmap replicate showed a chain of endothelial cells through cortex, likely a contiguous vasculature segment.

    “One replicate of the STARmap data also showed a chain of endothelial cells running through the cortex, likely a contiguous segment of vasculature (Figure 5D).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    The 5 glial clusters included two astrocyte clusters, polydendrocytes, and two oligodendrocyte clusters.

    “The 5 glial clusters we identified included two astrocyte clusters, polydendrocytes, and two clusters of oligodendrocytes (Wang et al., 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Wang et al., 2018 source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Integrated analysis spatially located each jointly defined population and reflected known mouse cortex spatial features.

    “Our integrated analysis spatially located each of the jointly defined populations (Figure 5D) and reflected the known spatial features of the mouse cortex, including meninges and sparse layer 1 interneurons at the surface, excitatory neurons organized in layers 2-6, and oligodendrocyte-rich white matter below the cortex (Figure 5D).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Joint analysis found corresponding broad cell classes across species and subtler cell types within each class after a second round.

    “The joint analysis identified both corresponding broad cell classes across species, and subtler cell types within each class after a second round of analysis (Figures S4B-S4F).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    LIGER correctly defined joint cell populations across scRNA-seq and STARmap datasets, with marker genes confirming correspondence.

    “Nevertheless, LIGER correctly defined joint cell populations across the datasets (Figure 5A-B), with expression of key marker genes confirming the correspondence of cells across these different modalities (Figure 5C).”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Spatial context is an important part of cellular identity, but most studies define cell types using scRNA-seq from dissociated cells.

  • May combine two claims.

    Findingown data

    New dimorphic genes were identified, often with complex cell-type-specific patterns across BNST subpopulations.

    “We also identified new dimorphic genes, often with complex cell-type-specific dimorphisms across the many BNST subpopulations (Figure 3L and Supplementary File 1).”
    Quote in the paperSubstantial claim

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

    Findingown data

    The standard multiplicative update NMF algorithm converges much more slowly than the authors' ALS implementation on non-sparse methylation data.

    “the standard multiplicative update NMF algorithm converges extremely slowly compared to our ALS implementation when running on the non-sparse methylation data”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingreasoning

    Because each subproblem is solved exactly at each iteration, the algorithm converges very quickly; previous benchmarks and the authors' own show block coordinate descent for NMF generally converges in many fewer iterations than multiplicative updates.

    “Additionally, because we solve each subproblem exactly at each iteration, the algorithm converges very quickly; previous empirical benchmarks (Kim, 2014) and our own have shown that block coordinate descent algorithms for NMF generally converge in many fewer iterations than multiplicative updates (Figure S1A-B).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Kim, 2014 source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    A third group of DMRs upstream of the Arx site lies in a highly conserved region (not a UCE), with three conservation spikes aligning precisely with the DMR locations.

    “A third group of DMRs upstream of the Arx site lies in a region of very high conservation (though not a UCE), with three clear spikes in conservation that aligned precisely with the locations of the DMRs (Figure 7H, left).”
    Quote in the paperNot a substantial claim

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

    Findingreasoning

    LIGER uses integrative nonnegative matrix factorization, solves a non-convex optimization problem, and produces different factorizations depending on initialization.

    “In contrast, LIGER uses integrative nonnegative matrix factorization, which solves a non-convex optimization problem, and thus produces a different factorization depending on the initialization used.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    The authors developed an R package that supports analysis of large-scale datasets and includes functions for tuning parameters, visualizing results, and quantifying integrative performance.

    “To facilitate adoption of the tool in the community, we have developed an R package that supports analysis of large-scale datasets and includes ancillary functions for tuning algorithmic parameters, visualizing results, and quantifying integrative performance.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    LIGER can identify shared cell types across individuals, species, and modalities, and also dataset-specific features, providing unified analysis of heterogeneous single-cell data.

    “We show here that LIGER enables the identification of shared cell types across individuals, species, and multiple modalities (gene expression, epigenetic or spatial data), as well as dataset-specific features, offering a unified analysis of heterogeneous single-cell datasets.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Defining cell types requires combining diverse single-cell measurements from multiple experiments and biological contexts.

  • May combine two claims.

    Premisecitation

    Widely used alignment methods treat dataset differences as technical and remove them, or project datasets into a shared latent space using maximally correlated dimensions.

    “the most widely used data alignment approaches (Butler et al., 2018; Haghverdi et al., 2018; Johnson et al., 2007; Risso et al., 2014) implicitly assume that the differences between datasets arise entirely from technical variation and attempt to eliminate them, or map datasets into a completely shared latent space using dimensions of maximum correlation (Butler et al., 2018).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Butler et al., 2018 source

      Not checked in the free run.

    • cites Haghverdi et al., 2018 source

      Not checked in the free run.

    • cites Johnson et al., 2007 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    These analyses indicate LIGER detects common populations without spurious alignment and preserves complex substructure across divergent datasets.

    “Together, these analyses indicate that LIGER sensitively detects common populations without spurious alignment and preserves complex substructure, even when applied across divergent datasets.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The main challenge of integrative analysis is reconciling the large heterogeneity across individual datasets.

  • Classified as probably not a substantial claim.

    Premisecitation

    New technologies can quantify DNA methylation, chromatin accessibility, and in situ expression across thousands to millions of cells.

    “new technologies can now measure DNA methylation (Luo et al., 2017; Mulqueen et al., 2018), chromatin accessibility (Cusanovich et al., 2018), and in situ expression (Coskun and Cai, 2016; Moffitt and Zhuang, 2016; Wang et al., 2018), in thousands to millions of cells.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Luo et al., 2017 source

      Not checked in the free run.

    • cites Cusanovich et al., 2018 source

      Not checked in the free run.

    • cites Wang et al., 2018 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    NG cells had not been described in BNST, but NG-like cells occur in amygdala.

    “Although, to our knowledge, NG cells have not been described in the BNST before, cells with NG-like anatomy and physiology have been observed within the amygdala (Manko et al., 2012), a structure with related functional roles.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Manko et al., 2012 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Chromatin remodeling genes showed less expression conservation, suggesting species differences in epigenetic regulation.

    “genes involved in chromatin remodeling (sharing the GO function “chromatin organization”) also showed less expression conservation, hinting at species differences in epigenetic regulation.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Comparing across species often yields deeper understanding of cell types.

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

    Premisecitation

    The Arx locus contains 8 ultraconserved elements (UCEs) that are perfectly conserved among human, mouse, and rat.

    “The Arx locus harbors 8 ultraconserved elements (UCEs)—long stretches of sequence showing perfect conservation among human, mouse, and rat (Bejerano et al., 2004; Colasante et al., 2008).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Bejerano et al., 2004 source

      Not checked in the free run.

    • cites Colasante et al., 2008 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    The results suggest BNST contains SPN-like neurons highly homologous to striatal SPNs and at least one Ppp1r1b+ population with tissue-specific specializations.

    “These results suggest that the BNST contains a combination of SPN-like neurons with high homology to striatal SPNs, while also harboring at least one Ppp1r1b+ population with tissue-specific specializations.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The molecular relationship between striatal SPNs and Ppp1r1b+ BNST cells was unknown.

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

    Premisecitation

    Gadd45b is thought to regulate DNA demethylation by recruiting TETs.

    “Consistent with our analysis, Gadd45b is thought to regulate DNA demethylation by recruiting TETs (Bayraktar and Kreutz, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Bayraktar and Kreutz, 2018

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    BNST shows significant sexual dimorphism in size and gene expression.

    “BNST also displays significant sexual dimorphism, both in size (Allen and Gorski, 1990; Hines et al., 1992) and gene expression (Xu et al., 2012).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Allen and Gorski, 1990 source

      Not checked in the free run.

    • cites Hines et al., 1992 source

      Not checked in the free run.

    • cites Xu et al., 2012

      Not checked in the free run.

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

    Premisecitation

    Several distal regulatory elements, including some within neighboring UCEs, have been demonstrated to regulate Arx expression.

    “Several distal regulatory elements, including some located within neighboring UCEs, have recently been demonstrated to regulate Arx expression (Colasante et al., 2008; Dickel et al., 2018).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Colasante et al., 2008 source

      Not checked in the free run.

    • cites Dickel et al., 2018 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The BNSTp and BNSTpr clusters were previously shown to be the most dimorphic anatomical region of the BNST.

    “were previously shown to be the most dimorphic anatomical region of the BNST (Xu et al., 2012)”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Xu et al., 2012

      Not checked in the free run.

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

    Premisenone given

    Multiplicative updates often used for NMF-like optimization problems do not have a convergence guarantee.

    “In contrast, the multiplicative updates often used for NMF-like optimization problems do not have a convergence guarantee.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    LIGER's inference of shared and dataset-specific factors enables a more transparent and nuanced definition of how cells correspond across datasets.

    “Finally, LIGER’s inference of both shared and dataset-specific factors enables a more transparent and nuanced definition of how cells correspond across datasets.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    LIGER infers interpretable shared and dataset-specific factors that often correspond to important biological signals, including signals not orthogonal to cell type, or technical signals, enabling their removal from downstream analysis.

    “LIGER infers interpretable shared and dataset-specific factors, which often correspond to important biological signals, including signals that are not orthogonal to cell type, or technical signals, enabling their removal from downstream analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisereasoning

    CCA solves a convex optimization problem and guarantees a deterministic, globally optimal solution.

    “CCA solves a convex optimization problem and thus guarantees a deterministic, globally optimal solution.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    LIGER's nonnegativity constraint yields interpretable factors, each generally corresponding to a biologically meaningful signal.

    “First, the nonnegativity constraint of NMF yields interpretable factors, such that each factor generally corresponds to a biologically meaningful signal.”
    Quote in the paperNot a substantial claim

  • The quote only partly states this claim.

    Interpretationreasoning

    When running on non-sparse methylation data, NMF factor loadings need to be centered as well as scaled during SFN clustering.

    “the NMF factor loadings need to be centered as well as scaled during the SFN clustering procedure”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductcitation

    The block coordinate descent algorithm satisfies the requirements of Bertsekas's theorem because each subproblem is convex with respect to the optimized block, and thus is guaranteed to converge to a fixed point (local minimum).

    “As described in Kim et al, our block coordinate descent algorithm satisfies the requirements of the theorem of Bertsekas (Bertsekas, 1999), because each of the subproblems is convex with respect to the block of variables being optimized. Thus, the algorithm is guaranteed to converge to a fixed point (local minimum).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Bertsekas, 1999

      Not checked in the free run.

    • cites Kim, 2014 source

      Not checked in the free run.

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

    Study design and conductnone given

    The mouse bed nucleus and human substantia nigra single-nucleus RNA-seq data were deposited in the Gene Expression Omnibus under accession GSE126836.

    “The mouse bed nucleus and human substantia nigra single nucleus RNA-seq data have been deposited in the Gene Expression Omnibus under accession code GSE126836.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Existing multi-omic integration methods, including LIGER, cannot incorporate different feature types such as gene expression and intergenic methylation in cell-type definitions.

    “We note that one limitation of existing methods for multi-omic integration, including LIGER, is their inability to incorporate different types of features, such as gene expression and intergenic methylation, in the definition of cell types.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    In comparisons with Seurat, the authors followed all documented procedures for running Seurat, including selecting the number of canonical components.

    “For comparisons with Seurat, we followed all documented procedures for running the method, including choosing the number of canonical components.”
    Quote in the paperNot a substantial claim

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

    Study design and conductcitation

    For the PBMC and pancreas datasets, the authors reproduced the analyses as described by Butler et al. and later Seurat tutorials.

    “For the PBMC and pancreas datasets, we reproduced the analyses as described by Butler et al. and subsequent Seurat tutorials (Butler et al., 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Butler et al., 2018 source

      Not checked in the free run.

45 of 45

  • The paper relies on this without support.

    Defining cell types requires combining diverse single-cell measurements from multiple experiments and biological contexts.

    “Defining cell types requires integrating diverse single-cell measurements from multiple experiments and biological contexts.”

  • The paper relies on this without support.

    Before this study, scRNA-seq had not been performed on BNST.

    “To date, scRNA-seq has not been performed on BNST, providing an opportunity to clarify how cell types are shared between this structure and datasets generated from related tissues.”

  • The paper relies on this without support.

    The molecular relationship between striatal SPNs and Ppp1r1b+ BNST cells was unknown.

    “The molecular relationship between striatal SPNs and these BNST cells is not known.”

  • The paper relies on this without support.

    It was unknown whether gene-expression clusters reflect epigenetic distinctions and vice versa.

    “First, it is unknown whether clusters defined from gene expression reflect epigenetic distinctions and vice versa.”

  • The paper relies on this without support.

    Many ante- and postmortem variables produce complex technical variation in gene expression, making biological variation in cell state harder to identify.

    “However, many ante- and postmortem variables create complex technical variation in gene expression, complicating efforts to identify biological variation in cell state.”

  • The paper relies on this without support.

    Spatial context is an important part of cellular identity, but most studies define cell types using scRNA-seq from dissociated cells.

    “Spatial context is an important aspect of cellular identity, but most studies have used scRNA-seq from dissociated cells to define cell types.”

  • The paper relies on this without support.

    The main challenge of integrative analysis is reconciling the large heterogeneity across individual datasets.

    “The major challenge of integrative analysis lies in reconciling the immense heterogeneity observed across individual datasets.”

  • The paper relies on this without support.

    In many analyses, both similarities and differences between datasets are biologically important.

    “However, in many kinds of analysis, both dataset similarities and differences are biologically important, such as when we seek to compare and contrast scRNA-seq data from healthy and disease-affected individuals.”

  • The paper relies on this without support.

    Glial activation is an important hallmark and driver of many brain diseases, including neurodegeneration and traumatic brain injury.

    “Glial activation is an important hallmark and driver of many brain diseases, including neurodegeneration and traumatic brain injury (TBI).”

  • The paper relies on this without support.

    Comparing across species often yields deeper understanding of cell types.

    “A deeper understanding of cell types often arises from comparisons across species.”

  • The paper relies on this without support.

    The STARmap data assay pre-selected markers.

    “Because the STARmap data assays pre-selected markers”

  • The paper relies on this without support.

    Gene expression and gene body mCH are generally anticorrelated.

    “Because gene expression and gene body mCH are generally anticorrelated”
12 of 12

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?

  • Integration using three independent methodologies (Harmony, Conos, and LIGER) yielded highly reliable cell type results based on visual inspection of UMAP distributions and low LISI values for cell type coherence when compared to Seurat integration.

    “In each case, highly reliable cell type results were observed, based on visual inspection of UMAP distributions and low Local Inverse Simpson Index (LISI) values for cell type coherence when compared to Seurat integration.”

    From A harmonized atlas of mouse spinal cord cell types and their spatial organization (2021) · cites this paper as [27]

  • The main integration was compared to three independent methods: Harmony, Conos, and LIGER.

    “This integration was compared to three independent methods (Harmony25, Conos26, and LIGER27).”

    From A harmonized atlas of mouse spinal cord cell types and their spatial organization (2021) · cites this paper as [27]

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.

  • Donor sex counts total eight, not seven

    Minor

    The stated five male and three female donors sum to eight, but the paper says there were seven donors.

    Reported 7 · recomputed 8

  • Factor-reduction instructions use conflicting conditions

    Minor

    The text gives k2 > k1 both for adding new factors and for selecting k2 factors from the existing factorization. The latter instruction conflicts with the former and with the algorithm table, which places factor selection under k2 < k1.

  • CGE sampling count does not match the stated sampling scheme

    Minor

    Sampling 100 cells from each of 11 subclusters would yield 1,100 cells, not the reported 1,060.

    Reported 1060 · recomputed 1100

3 of 3

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Welch, J. D., Kozareva, V., Ferreira, A., Vanderburg, C., Martin, C., & Macosko, E. Z.
Affiliations
Broad Institute (US)
Corresponding author
  • Welch, Joshua D.
  • Macosko, Evan Z.
ORCID
Venue
Cell
Publisher
Elsevier BV
Published
2019-06
Volume
177
Issue
7
Pages
1873-1887.e17
ISSN
0092-8674 (ISSN-L)
DOI
10.1016/j.cell.2019.05.006
PMC
PMC6716797
OpenAlex
W2948469692
Cited by
1,193
References
70
Citations by year
2026: 121 · 2025: 183 · 2024: 181 · 2023: 182 · 2022: 173 · 2021: 193 · 2020: 136 · 2019: 23 · 2018: 1
FWCI
43.78
Type
journal article
Language
en
Open access
Licence
Locations
  • Cell (publishedVersion)
  • PubMed (publishedVersion)
  • PubMed Central (acceptedVersion)
Funders
  • Stanley Center for Psychiatric Research
  • Chan Zuckerberg Initiative (2017-175259, 2018-183155)
  • NIH/NIMH BRAIN (1U19MH114821)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Single-cell and spatial transcriptomics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Gene Regulatory Network Analysis
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Biophysics › Cell Image Analysis Techniques
Keywords
LIGER, multi-omics integration, single-cell data integration, bed nucleus of the stria terminalis, sexual dimorphism, substantia nigra, frontal cortex, spatial transcriptomics, DNA methylation, epigenomic regulation, human-mouse comparison
MeSH terms
Adolescent, Adult, Aged, Animals, Female, Gene Expression Regulation, Humans, Male, Middle Aged, Septal Nuclei, Substantia Nigra, Sequence Analysis, RNA, DNA Methylation, Mice, 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 2026-10-08.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-08.

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