CoChat
Sign Up
Cited by 2,231Single-cell and spatial transcriptomicsGene expression and cancer classificationGenomics and Phylogenetic Studies

Current best practices in single-cell RNA-seq analysis: a tutorial

Open access
Luecken, Malte D, Theis, Fabian J·2019·Molecular Systems Biology
DOI 10.15252/msb.20188746PMC PMC6582955
Open this paper in CoChat — free accountOpen PDF

Abstract

Single‐cell RNA‐seq has enabled gene expression to be studied at an unprecedented resolution. The promise of this technology is attracting a growing user base for single‐cell analysis methods. As more analysis tools are becoming available, it is becoming increasingly difficult to navigate this landscape and produce an up‐to‐date workflow to analyse one's data. Here, we detail the steps of a typical single‐cell RNA‐seq analysis, including pre‐processing (quality control, normalization, data correction, feature selection, and dimensionality reduction) and cell‐ and gene‐level downstream analysis. We formulate current best‐practice recommendations for these steps based on independent comparison studies. We have integrated these best‐practice recommendations into a workflow, which we apply to a public dataset to further illustrate how these steps work in practice. Our documented case study can be found at https://www.github.com/theislab/single-cell-tutorial . This review will serve as a workflow tutorial for new entrants into the field, and help established users update their analysis pipelines.

AI snapshot

Written from the paper’s full text.

TL;DR

This tutorial/review outlines current best-practice steps for single-cell RNA-seq analysis from count matrices through pre-processing and downstream cell- and gene-level analysis, based on independent comparison studies. It also provides a documented case-study workflow applied to a public mouse intestinal epithelium dataset.

Key findings

  • Recommends joint QC on count depth, number of genes, and mitochondrial fraction, with permissive thresholds; in the Haber et al dataset example, thresholds are applied at 1,500 counts and 700 genes.
  • Recommends scran normalization for non-full-length datasets, log(x+1) transformation, no gene scaling, ComBat for batch correction when cell-type/state compositions are consistent, and separate methods for data integration versus batch correction.
  • Recommends selecting 1,000–5,000 highly variable genes; UMAP for exploratory visualization, PCA for general summarization, diffusion maps for trajectory inference; Louvain clustering on a KNN graph; Slingshot for simple trajectories and PAGA for complex trajectories; MAST or limma for differential expression testing.
  • Notes the field had 385 analysis tools as of 7 March 2019, datasets can contain over 20,000 genes and over 15,000 dimensions after QC, and expression recovery should be used cautiously, perhaps only for visual display.

Methods

  • Literature review/tutorial synthesizing independent method-comparison studies and current tools; no new primary experimental data are generated.
  • Case study applies the best-practice workflow to a public mouse intestinal epithelium dataset from Haber et al (2017), integrating R and Python tools in a Jupyter–IPython notebook with rpy2, available at https://www.github.com/theislab/single-cell-tutorial.
  • Covers QC, normalization, data correction/integration, feature selection, dimensionality reduction, clustering, trajectory inference, differential expression, gene set analysis, and gene regulatory network inference.

Limitations

  • Stated: available method comparisons lag behind the latest method developments, so individual tool recommendations will require updates; there is no consensus on gene scaling, how denoised data should be used, or best practice for gene temporal dynamics.
  • Stated: no single normalization method is appropriate for all scRNA-seq data; data correction methods inevitably over- or under-correct; expression recovery may introduce false correlations and has scalability issues.
  • Evident: the recommendations are not a systematic benchmark or standardized pipeline, but a tutorial synthesis of cited comparisons plus one public dataset case study.

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

Integrity signals

Registry

Crossref · 10.15252/msb.20188746

Cites

168 of 168 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 162 cited sources checked · 0 supported · 0 flagged

  • Findingreasoning

    Datasets that contain heterogeneous mixtures of cell types may exhibit multiple QC covariate peaks.

    “Datasets that contain heterogeneous mixtures of cell types may exhibit multiple QC covariate peaks.”
    Quote in the paperSubstantial claim
  • Findingown data

    Molecular counts can differ strongly between cells.

    “Indeed, molecular counts can differ strongly between cells (see case study on project github).”
    Quote in the paperSubstantial claim
  • Findingreasoning

    Regressing out count depth effects can improve the performance of trajectory inference algorithms.

    “Regressing out count depth effects can improve the performance of trajectory inference algorithms, which rely on finding transitions between cells (see case study on project github).”
    Quote in the paperNot a substantial claim
  • Findingcitation

    Significant marker genes are found even when clustering random data generated by splatter.

    “We thus find significant marker genes even when clustering random data generated by splatter (Zappia et al, 2017) (see Appendix Supplementary Text S3).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Zappia et al, 2017 source

      Not checked in the free run.

  • Interpretationreasoning

    Count depths for identical cells can differ due to variability inherent in capture, reverse transcription, and sequencing.

    “Count depths for identical cells can differ due to the variability inherent in each of these steps.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 124
  • May combine two claims. 44

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

    Findingnone given

    Proliferating cell populations can be identified based on cell cycle scores.

    “For example, proliferating cell populations can be identified based on cell cycle scores (see case study on project github).”
    Quote in the paperNot a substantial claim

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

    Findingnone given

    Typically between 1,000 and 5,000 HVGs are selected for downstream analysis.

    “Depending on the task and the complexity of the dataset, typically between 1,000 and 5,000 HVGs are selected for downstream analysis (see Fig EV1 and Dataset EV1).”
    Quote in the paperNot a substantial claim

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

    Findingown data

    The 'elbow' of the variance explained plot for the mouse intestinal epithelium data lies between PCs 5 and 7.

    “The “elbow” of this plot, which is used to select relevant PCs to analyse the dataset, lies between PCs 5 and 7.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Cells with a comparatively high fraction of mitochondrial counts may be involved in respiratory processes.

    “For example, cells with a comparatively high fraction of mitochondrial counts may be involved in respiratory processes.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Sufficient data quality cannot be determined a priori; it is judged based on downstream analysis performance.

    “As “sufficient data quality” cannot be determined a priori, it is judged based on downstream analysis performance (e.g., cluster annotation).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Outlier barcodes in QC covariate distributions can correspond to dying cells, cells whose membranes are broken, or doublets.

    “These outlier barcodes can correspond to dying cells, cells whose membranes are broken, or doublets.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Weinreb et al used a simple extension of CPM that excludes genes that account for at least 5% of the total counts in any cell when calculating size factors.

    “For example, Weinreb et al (2018) used a simple extension of CPM that excludes genes that account for at least 5% of the total counts in any cell, when calculating their size factors.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Weinreb et al, 2018 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    Downsampling increases technical dropout rates, which CPM and other global scaling normalization methods do not.

    “While downsampling throws away data, it also increases technical dropout rates which CPM and other global scaling normalization methods do not.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Filtering out genes that are not expressed in more than a few cells can drastically reduce the number of genes and remove genes not informative of cellular heterogeneity.

    “This number can be drastically reduced by filtering out genes that are not expressed in more than a few cells and are thus not informative of the cellular heterogeneity.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Cells with high counts may be larger in size.

    “Cells with low counts and/or genes may correspond to quiescent cell populations, and cells with high counts may be larger in size.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Cells with low counts and/or genes may correspond to quiescent cell populations.

    “Cells with low counts and/or genes may correspond to quiescent cell populations, and cells with high counts may be larger in size.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The quality of a dataset can be determined by experimental QC metrics.

    “The quality of a dataset can be determined by experimental QC metrics (see Appendix Supplementary Text S2).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Appendix Supplementary Text S2

      Not checked in the free run.

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

    Premisenone given

    The most commonly used normalization protocol is count depth scaling, also referred to as CPM normalization.

    “The most commonly used normalization protocol is count depth scaling, also referred to as “counts per million” or CPM normalization.”
    Quote in the paperNot a substantial claim

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

    Interpretationcitation

    This approach allows for molecular count variability in few highly expressed genes.

    “This approach allows for molecular count variability in few highly expressed genes.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The growing number of analysis methods (385 tools as of 7 March 2019) and exploding dataset sizes are challenges to standardization.

    “The challenges to standardization include the growing number of analysis methods (385 tools as of 7 March 2019) and exploding dataset sizes (Angerer et al, 2017; Zappia et al, 2018).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Angerer et al, 2017

      Not checked in the free run.

    • cites Zappia et al, 2018

      Not checked in the free run.

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

    Premisecitation

    Analysis tools for scRNA-seq data are written in a variety of programming languages, most prominently R and Python.

    “Analysis tools for scRNA‐seq data are written in a variety of programming languages—most prominently R and Python (Zappia et al, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Zappia et al, 2018

      Not checked in the free run.

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

    Interpretationnone given

    The decision of which covariates to consider for data correction will depend on the intended downstream analysis.

    “Instead, the decision of which covariates to consider will depend on the intended downstream analysis.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Normalization attempts to remove the effects of count sampling.

    “Normalization as described above attempts to remove the effects of count sampling.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The popular Seurat tutorials generally apply gene scaling.

    “While the popular Seurat tutorials (Butler et al, 2018) generally apply gene scaling, the authors of the Slingshot method opt against scaling over genes in their tutorial (Street 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.

  • May combine two claims.

    Interpretationreasoning

    Batch effects exist on multiple levels: between groups of cells in an experiment, between experiments performed in the same laboratory, or between datasets from different laboratories.

    “The resulting effects exist on multiple levels: between groups of cells in an experiment, between experiments performed in the same laboratory or between datasets from different laboratories.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Cole et al find that different normalization methods perform optimally for different datasets and argue that their scone tool should be used to select the appropriate normalization method for a specific dataset.

    “Indeed Cole et al (2019) find that different normalization methods perform optimally for different datasets and argue that their scone tool should be used to select the appropriate normalization method for a specific dataset.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Cole et al, 2019

      Not checked in the free run.

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

    Interpretationreasoning

    These approaches may be particularly relevant for plate-based scRNA-seq datasets where the larger variation of count depths per cell can mask the heterogeneity between cells.

    “These approaches may be particularly relevant for plate‐based scRNA‐seq datasets where the larger variation of count depths per cell can mask the heterogeneity between cells.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Mayer et al fit a negative binomial model to count data, using technical covariates such as read depth and number of counts per gene to fit model parameters.

    “For example, Mayer et al (2018) fit a negative binomial model to count data, using technical covariates such as the read depth and the number of counts per gene to fit the model parameters.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Mayer et al, 2018 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationnone given

    Data from full-length protocols may benefit from normalization methods that take into account gene length, while 3′ enrichment data do not.

    “Data from full‐length protocols may benefit from normalization methods that take into account gene length (e.g. Patel et al, 2014; Kowalczyk et al, 2015; Soneson & Robinson, 2018), while 3′ enrichment data do not.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Batch effects are typically corrected using linear methods, while non-linear approaches are used for data integration.

    “While batch effects are typically corrected using linear methods, non‐linear approaches are used for data integration.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    The most prominent technical covariates in single-cell data are count depth and batch.

    “The most prominent technical covariates in single‐cell data are count depth and batch.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    CPM, high-count filtering CPM, and scran use linear, global scaling to normalize count data.

    “CPM, high‐count filtering CPM, and scran use linear, global scaling to normalize count data.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Variation in cell size accounts for the transcriptomic effect generally attributed to the cell cycle.

    “Finally, it has been argued that variation in cell size accounts for the transcriptomic effect generally attributed to the cell cycle (McDavid et al, 2016).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites McDavid et al, 2016 source

      Not checked in the free run.

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

    Interpretationreasoning

    An alternative to regression-based strategies for removing count effects is to use a more rigorous normalization procedure such as downsampling or non-linear normalization methods.

    “An alternative to regression‐based strategies for removing count effects is to use a more rigorous normalization procedure such as downsampling or non‐linear normalization methods (see “Normalization” section).”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The authors of the Slingshot method opt against scaling over genes in their tutorial.

    “While the popular Seurat tutorials (Butler et al, 2018) generally apply gene scaling, the authors of the Slingshot method opt against scaling over genes in their tutorial (Street et al, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Street et al, 2018 source

      Not checked in the free run.

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

    Interpretationnone given

    Cells may differ in size and therefore in mRNA molecule counts.

    “For example, cells may differ in size and therefore in mRNA molecule counts.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    A commonly used normalization method for full-length scRNA-seq data is TPM normalization.

    “A commonly used normalization method for full‐length scRNA‐seq data is TPM normalization (Li et al, 2009), which comes from bulk RNA‐seq analysis.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Li et al, 2009

      Not checked in the free run.

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

    Premisenone given

    The most common biological data correction is to remove the effects of the cell cycle on the transcriptome.

    “The most common biological data correction is to remove the effects of the cell cycle on the transcriptome.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The differing environments experienced by cells in distinct groups can have an effect on the measurement of the transcriptome or on the transcriptome itself.

    “The differing environments experienced by the cells can have an effect on the measurement of the transcriptome or on the transcriptome itself.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Correction for biological covariates serves to single out particular biological signals of interest.

    “While correcting for technical covariates may be crucial to uncovering the underlying biological signal, correction for biological covariates serves to single out particular biological signals of interest.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    After normalization, data matrices are typically log(x+1)-transformed.

    “After normalization, data matrices are typically log(x+1)‐transformed.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Correcting for technical covariates may be crucial to uncovering the underlying biological signal.

    “While correcting for technical covariates may be crucial to uncovering the underlying biological signal, correction for biological covariates serves to single out particular biological signals of interest.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Reduced data would be the natural candidate for a broadly applicable data layer.

    “As the final pre‐processing stage, reduced data would be the natural candidate for a broadly applicable data layer.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The corrected data layer represents the 'cleanest' version of the data, closest to the underlying biological signal.

    “The corrected data layer represents the “cleanest” version of the data, which is the closest approximation of the underlying biological signal.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    In non-linear dimensionality reduction, interpretability of reduced dimensions is sacrificed.

    “Especially in the non‐linear case, the interpretability of the reduced dimensions is sacrificed in this process.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    A particularly prominent aspect of this noise is dropout.

    “A particularly prominent aspect of this noise is dropout.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    PCA is commonly used as a pre-processing step for non-linear dimensionality reduction methods.

    “Indeed, PCA is commonly used as a pre‐processing step for non‐linear dimensionality reduction methods.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given

    A dedicated visualization technique will typically provide a better representation of the variability.

    “However, a dedicated visualization technique will typically provide a better representation of the variability.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    A 2-dimensional visualization output should not be used to summarize a dataset, but a summarization method can be used to visualize data using leading reduced components.

    “While a 2‐dimensional visualization output should not be used to summarize a dataset, a summarization method can be used to visualize the data using the leading reduced components.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    There is currently no consensus on how denoised data should be used.

    “There is currently no consensus on how denoised data should be used in the light of these considerations (see “Stages of processed data” section).”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Feature selection methods that use gene expression means and variances cannot be used when gene expression values have been normalized to zero mean and unit variance, or when residuals from model fitting are used as normalized expression values.

    “Feature selection methods that use gene expression means and variances cannot be used when gene expression values have been normalized to zero mean and unit variance, or when residuals from model fitting are used as normalized expression values.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Scalability to large datasets is still an issue for currently available expression recovery methods.

    “Furthermore, scalability to large datasets is still an issue for currently available expression recovery methods.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Cell-level analysis typically focuses on the description of two structures: clusters and trajectories.

    “Cell‐level analysis typically focuses on the description of two structures: clusters and trajectories.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Testing for differential expression is only biologically interpretable in gene space, which is not fully represented in reduced data.

    “However, testing for differential expression is only biologically interpretable in gene space, which is not (fully) represented in reduced data.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Summarization does not prescribe the number of output components; higher components become less important.

    “Summarization does not prescribe the number of output components. Instead, higher components become less important for describing the variability present in the data.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    PCA summarizes a dataset via its top N principal components, where N can be determined by elbow heuristics or the jackstraw method.

    “Typically, PCA summarizes a dataset via its top N principal components, where N can be determined by “elbow” heuristics (see Fig 4F) or the permutation‐test‐based jackstraw method (Chung & Storey, 2015; Macosko et al, 2015).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Chung & Storey, 2015

      Not checked in the free run.

    • cites Macosko et al, 2015 source

      Not checked in the free run.

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

    Interpretationreasoning

    The simple linearity of PCA has the advantage that distances in reduced dimensional space have a consistent interpretation in all regions.

    “The simple linearity of PCA has the advantage that distances in reduced dimensional space have a consistent interpretation in all regions of this space.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    t-SNE visualizations may exaggerate differences between cell populations and overlook potential connections.

    “Thus, these visualizations may exaggerate differences between cell populations and overlook potential connections between these populations.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Louvain has been shown to outperform other clustering methods for single-cell RNA-seq data, and flow and mass cytometry data.

    “It has been shown to outperform other clustering methods for single‐cell RNA‐seq data (Duò et al, 2018; Freytag et al, 2018), and flow and mass cytometry data (Weber & Robinson, 2016).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Duò et al, 2018

      Not checked in the free run.

    • cites Freytag et al, 2018 source

      Not checked in the free run.

    • cites Weber & Robinson, 2016

      Not checked in the free run.

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

    Premisenone given

    PCA and diffusion maps are popular dimensionality reduction techniques that are principally summarization methods.

    “Two popular dimensionality reduction techniques that are principally summarization methods are principal component analysis (PCA; Pearson, 1901) and diffusion maps (Coifman et al, 2005) which were popularized for single‐cell analysis by Haghverdi et al (2015).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Pearson, 1901

      Not checked in the free run.

    • cites Coifman et al, 2005 source

      Not checked in the free run.

    • cites Haghverdi et al (2015) source

      Not checked in the free run.

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

    Premisenone given

    Organizing cells into clusters is typically the first intermediate result of any single-cell analysis.

    “Organizing cells into clusters is typically the first intermediate result of any single‐cell analysis.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    There are two main objectives of dimensionality reduction methods: visualization and summarization.

    “There are two main objectives of dimensionality reduction methods: visualization and summarization.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    UMAP's speed and ability to scale to large numbers of cells set it apart.

    “What sets UMAP apart in this comparison is its speed and ability to scale to large numbers of cells (Becht et al, 2018).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Becht et al, 2018 source

      Not checked in the free run.

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

    Interpretationreasoning

    Diffusion components emphasize transitions in the data, so they are principally used when continuous processes such as differentiation are of interest.

    “As diffusion components emphasize transitions in the data, they are principally used when continuous processes such as differentiation are of interest.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Downstream analysis methods extract biological insights and describe the underlying biological system by fitting interpretable models to the data.

    “After pre‐processing, methods that we call downstream analysis are used to extract biological insights and describe the underlying biological system. These descriptions are obtained by fitting interpretable models to the data.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    ComBat uses all cells in a batch to fit batch parameters, which confounds batch effect with biological differences not shared among datasets.

    “When estimating batch effects, ComBat uses all cells in a batch to fit batch parameters. This approach will confound the batch effect with biological differences between cell types or states that are not shared among datasets.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Data integration methods have increased degrees of freedom of non-linear approaches, leading to risk of over-correction.

    “While data integration methods can also be applied to simple batch correction problems, we recommend to be wary of over‐correction given the increased degrees of freedom of non‐linear data integration approaches.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Data correction may not be necessary for single batch datasets.

    “For example, data correction may not be necessary for single batch datasets.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    PCA does not capture structure in few dimensions as well as non-linear methods, but it is the basis of many analysis tools for clustering or trajectory inference.

    “While PCA does not capture the structure of the data in few dimensions as well as non‐linear methods, it is the basis of many currently available analysis tools for clustering or trajectory inference.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Data integration tools may over-correct simple batch effects.

    “Data integration tools may over‐correct simple batch effects.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    Clusters allow us to infer the identity of member cells.

    “Clusters allow us to infer the identity of member cells.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    Principal components can be correlated with quantities of interest to assess their importance.

    “Thus, we can correlate quantities of interest with principal components to assess their importance.”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    The k-means clustering algorithm requires an input of the number of clusters expected, which is usually unknown and must be calibrated heuristically.

    “This approach requires an input of the number of clusters expected, which is usually unknown and must be calibrated heuristically.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    If the clustering is biologically meaningful, the top-ranked marker genes will still be the best marker gene candidates.

    “Assuming the clustering is biologically meaningful, the top‐ranked marker genes will still be the best marker gene candidates.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The clustering is biologically meaningful.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    These databases greatly facilitate cell identity annotation.

    “These databases greatly facilitate cell identity annotation.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Currently few dedicated gene temporal dynamics tools exist.

    “Currently few dedicated gene temporal dynamics tools exist.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The validity of identified clusters can only be determined by successful annotation of the represented biology.

    “As any clustering algorithm will produce a partition of the data, the validity of the identified clusters can only be determined by successful annotation of the represented biology.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Given the few available tools, a best practice for investigating gene temporal dynamics cannot yet be determined.

    “Given the few available tools, a best practice for investigating gene temporal dynamics cannot yet be determined.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Further information is needed to validate whether a biological process was indeed captured by a trajectory.

    “Thus, further information is needed to validate whether a biological process was indeed captured.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    BEAM is a tool integrated into the Monocle TI pipeline that allows for detection of branch-specific gene dynamics.

    “BEAM is a tool integrated into the Monocle TI pipeline (Qiu et al, 2017a), which allows for detection of branch‐specific gene dynamics.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Qiu et al, 2017a

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    Increases in speed go together with compromises in flexibility.

    “Increases in speed go together with compromises in flexibility.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    With the current set-up, P-values for marker genes are often inflated, which can lead to an overestimation of the number of marker genes.

    “With the current set‐up, the P‐values are often inflated, which can lead to an overestimation of the number of marker genes.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: The obtained cell clusters represent the biological ground truth.

  • May combine two claims.

    Premisereasoning

    Marker genes differentiate a cluster from others in the dataset and are dependent on both the cell cluster and the dataset composition.

    “Secondly, marker genes differentiate a cluster from others in the dataset and are thus dependent not only on the cell cluster, but also on the dataset composition.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    PAGA was recommended for more complex trajectories.

    “If more complex trajectories are expected, PAGA (Wolf et al, 2019) was recommended by the authors.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Wolf et al, 2019

      Not checked in the free run.

    • cites Saelens et al, 2018

      Not checked in the free run.

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

    Interpretationnone given

    If cells are sampled in an unbiased manner, dense regions along a trajectory indicate preferred transcriptomic states.

    “Assuming that cells were sampled in an unbiased manner, dense regions along a trajectory indicate preferred transcriptomic states.”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Cells were sampled in an unbiased manner.

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

    Premisereasoning

    Several top-performing TI methods rely on clustered data, so TI is typically performed after clustering.

    “Furthermore, as several top‐performing TI methods rely on clustered data, TI is typically performed after clustering.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    As marker genes are expected to have strong differential expression effects, simple statistical tests such as Wilcoxon rank-sum or t-test are often used to rank genes.

    “As marker genes are expected to have strong differential expression effects, simple statistical tests such as the Wilcoxon rank‐sum test or the t‐test are often used to rank genes by their difference in expression between these two groups.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    It is not always clear what constitutes a cell type.

    “Firstly, it is not always clear what constitutes a cell type.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    Marker genes characterize the cluster and are used to annotate it with a meaningful biological label.

    “These so‐called marker genes characterize the cluster and are used to annotate it with a meaningful biological label.”
    Quote in the paperSubstantial claim

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

    Premisenone given

    Automated cluster annotation methods offer a vast speedup of the iteration of clustering, cluster annotation, re- or subclustering and re-annotation.

    “Automated cluster annotation methods offer a vast speedup of this process.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Regression frameworks differ in their noise model assumptions and the class of function used to describe the expression as a function of pseudotime.

    “Regression frameworks differ in their noise model assumptions and the class of function used to describe the expression as a function of pseudotime.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Pseudotime is often interpreted as a proxy for developmental time.

    “While this variable is related to transcriptional distances from a root cell, it is often interpreted as a proxy for developmental time (Moignard et al, 2015; Haghverdi et al, 2016; Fischer et al, 2018; Griffiths et al, 2018).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Moignard et al, 2015 source

      Not checked in the free run.

    • cites Haghverdi et al, 2016 source

      Not checked in the free run.

    • cites Fischer et al, 2018

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Genes that vary smoothly across pseudotime characterize the trajectory and can be used to identify the underlying biological process.

    “Genes that vary smoothly across pseudotime characterize the trajectory and can be used to identify the underlying biological process.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    Marker genes for the same cell-identity cluster may differ between datasets purely due to dataset cell type and state compositions.

    “Note that marker genes for the same cell‐identity cluster may differ between datasets purely due to dataset cell type and state compositions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationcitation

    Subclustering can allow identification of cell states within cell-type clusters, but may also lead to patterns that arise only from noise.

    “Such subclustering can allow the user to identify cell states within cell‐type clusters (Wagner et al, 2016), but may also lead to patterns that arise only from noise in the data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Wagner et al, 2016 source

      Not checked in the free run.

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

    Premisecitation

    Salmonella infection increases the proportion of enterocytes in the mouse intestinal epithelium.

    “For example, salmonella infection has been shown to increase the proportion of enterocytes in the mouse intestinal epithelium (Haber et al, 2017).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Haber et al, 2017 source

      Not checked in the free run.

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

    Premisecitation

    Regulator genes help us understand how and why biological processes are triggered and represent potential drug targets.

    “Regulator genes help us understand how and why biological processes are triggered and represent potential drug targets (Gashaw et al, 2012).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Gashaw et al, 2012

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Interpretationnone given

    Investigating compositional changes in single-cell data requires sufficient cell numbers to robustly assess cell-identity cluster proportions, and sufficient sample numbers to evaluate expected background variation in cell-identity cluster compositions.

    “Investigating compositional changes in single‐cell data requires sufficient cell numbers to robustly assess cell‐identity cluster proportions, and sufficient sample numbers to evaluate expected background variation in cell‐identity cluster compositions.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    This information can come in the form of perturbation experiments, inferred regulatory gene dynamics, and support from RNA velocity.

    “This information can come in the form of perturbation experiments, inferred regulatory gene dynamics, and support from RNA velocity.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    Reference atlases will not contain exactly the same cell identities as the dataset under investigation.

    “As mentioned above, reference atlases will not contain exactly the same cell identities as the dataset under investigation.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Even when design matrices are not full rank, DE testing tools will often adapt the matrix and run without outputting a warning.

    “Even when design matrices are not full rank, DE testing tools will often adapt the matrix and run without outputting a warning.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Gene-level analysis methods often produce long lists of candidate genes that are difficult to interpret.

    “Gene‐level analysis methods often produce long lists of candidate genes that are difficult to interpret.”
    Quote in the paperSubstantial claim

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

    Premisenone given

    In the scenario we describe here, the condition covariate is determined in the experimental set-up.

    “In the scenario we describe here, the condition covariate is determined in the experimental set‐up.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Differential expression testing, gene set analyses and gene regulatory network inference directly investigate molecular signals in the data.

    “Differential expression testing, gene set analyses and gene regulatory network inference directly investigate molecular signals in the data.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    As uncorrected, measured data should be used for DE testing, accounting for confounding factors is crucial to robust estimation of differentially expressed genes.

    “As uncorrected, measured data should be used for DE testing, accounting for confounding factors is crucial to robust estimation of differentially expressed genes.”
    Quote in the paperSubstantial claim

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

    Interpretationnone given

    When correcting for multiple categorical batch covariates, it becomes increasingly difficult to visually spot confounding groups of covariates.

    “When correcting for multiple categorical batch covariates, it becomes increasingly difficult to visually spot confounding groups of covariates.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Gene regulatory network inference is performed based on measurements of gene co-expression such as correlation, mutual information, or via regression models.

    “Gene regulatory network inference is performed based on measurements of gene co‐expression such as correlation, mutual information, or via regression models (Chen & Mar, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chen & Mar, 2018

      Not checked in the free run.

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

    Premisecitation

    While pseudogenes or non-coding RNAs can be informative, they are often ignored in the analysis.

    “While pseudogenes or non‐coding RNAs can be informative (An et al, 2017), they are often ignored in the analysis.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites An et al, 2017

      Not checked in the free run.

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

    Premisenone given

    These approaches use cellular heterogeneity as context in which gene expression is to be understood.

    “Rather than describing the cellular heterogeneity, these approaches use this heterogeneity as context in which gene expression is to be understood.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Interaction between cell clusters is inferred from the expression of receptors and their cognate ligands.

    “Here, interaction between cell clusters is inferred from the expression of receptors and their cognate ligands.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    We can facilitate the interpretation of these results by grouping the genes into sets based on shared characteristics and testing whether these characteristics are overrepresented in the candidate gene list.

    “We can facilitate the interpretation of these results by grouping the genes into sets based on shared characteristics and testing whether these characteristics are overrepresented in the candidate gene list.”
    Quote in the paperSubstantial claim

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

    Premisenone given

    Thousands of genes may be differentially expressed between treated and control cells.

    “For example, thousands of genes may be differentially expressed between treated and control cells.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Bulk methods were developed to accurately estimate gene variance from few samples, while single-cell data do not present this problem.

    “While bulk methods were developed to accurately estimate gene variance from few samples, single‐cell data do not present this problem.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    To interpret DE results, we typically group genes based on involvement in common biological processes.

    “To interpret DE results, we typically group genes based on involvement in common biological processes.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Genes do not function independently.

    “Genes do not function independently.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Single-cell data contain unique technical noise artefacts such as dropout, and high cell-to-cell variability.

    “On the other hand, single‐cell data contain unique technical noise artefacts such as dropout, and high cell‐to‐cell variability (Hicks et al, 2017; Vallejos et al, 2017).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Hicks et al, 2017 source

      Not checked in the free run.

    • cites Vallejos et al, 2017 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    Obtained P-values for DE tests over conditions represent the expected measures of significance and must be corrected for multiple testing.

    “Obtained P‐values for DE tests over conditions represent the expected measures of significance and must be corrected for multiple testing.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Enrichment of annotations on the gene list can be tested using a vast array of tools, which are reviewed and compared in Huang et al (2009) and Tarca et al (2013).

    “Enrichment of annotations on the gene list can be tested using a vast array of tools, which are reviewed and compared in Huang et al (2009) and Tarca et al (2013).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Huang et al, 2009 source

      Not checked in the free run.

    • cites Tarca et al, 2013

      Not checked in the free run.

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

    Premisereasoning

    Gene-level analysis of single-cell data has a broader scope than characterizing cellular structures.

    “While we have so far focused on gene‐level analysis methods that characterize cellular structures, gene‐level analysis of single‐cell data has a broader scope.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    MAST was the best-performing single-cell DE testing method in the aforementioned study, and outperformed bulk and single-cell methods in a small-scale comparison on a single dataset.

    “It was the best‐performing single‐cell DE testing method in the aforementioned study (Soneson & Robinson, 2018), and outperformed bulk and single‐cell methods in a small‐scale comparison on a single dataset (Vieth et al, 2017).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Soneson & Robinson, 2018 source

      Not checked in the free run.

    • cites Vieth et al, 2017

      Not checked in the free run.

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

    Premisecitation

    Several single-cell GRN inference methods use trajectories with mechanistic differential equation models.

    “Indeed, several single‐cell GRN inference methods use trajectories with mechanistic differential equation models (Ocone et al, 2015; Matsumoto et al, 2017).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Ocone et al, 2015 source

      Not checked in the free run.

    • cites Matsumoto et al, 2017 source

      Not checked in the free run.

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

    Premisenone given

    A recent development in the single-cell analysis field is the use of paired gene labels to perform ligand-receptor analysis.

    “A recent development in the single‐cell analysis field is the use of paired gene labels to perform ligand–receptor analysis.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Biological process labels are stored in databases such as MSigDB, the Gene Ontology, or the pathway databases KEGG and Reactome.

    “Biological process labels are stored in databases such as MSigDB (Liberzon et al, 2011), the Gene Ontology (Ashburner et al, 2000; The Gene Ontology Consortium, 2017), or the pathway databases KEGG (Kanehisa et al, 2017) and Reactome (Fabregat et al, 2018).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Liberzon et al, 2011 source

      Not checked in the free run.

    • cites Ashburner et al, 2000 source

      Not checked in the free run.

    • cites Kanehisa et al, 2017 source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    This set-up distinguishes DE testing over conditions and DE testing over clusters.

    “This set‐up distinguishes DE testing over conditions and DE testing over clusters.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisenone given

    Clustering and trajectory inference represent two distinct views of single-cell data.

    “Clustering and trajectory inference represent two distinct views of single‐cell data.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    Although limma is a bulk DE testing method, limma-voom was shown to achieve comparable performance to MAST.

    “Although limma is a bulk DE testing method, limma–voom was shown to achieve comparable performance to MAST.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Law et al, 2014 source

      Not checked in the free run.

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

    Premisenone given

    Uncovering these regulatory interactions is the goal of gene regulatory network (GRN) inference methods.

    “Uncovering these regulatory interactions is the goal of gene regulatory network (GRN) inference methods.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Gene set information can be found in curated label databases for various applications.

    “Gene set information can be found in curated label databases for various applications.”
    Quote in the paperNot a substantial claim

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

    Premisereasoning

    The expression level of a gene is determined by a complex interplay of regulatory interactions with other genes and small molecules.

    “Instead, the expression level of a gene is determined by a complex interplay of regulatory interactions with other genes and small molecules.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    MAST has a 10-fold to 100-fold faster runtime than weighted bulk methods, and a further 10-fold speedup can be achieved using limma-voom.

    “While MAST has a 10‐fold to 100‐fold faster runtime than weighted bulk methods (Van den Berge et al, 2018), a further 10‐fold speedup can be achieved using limma–voom (Law et al, 2014).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Van den Berge et al, 2018

      Not checked in the free run.

    • cites Law et al, 2014 source

      Not checked in the free run.

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

    Premisecitation

    DE testing is a well-documented problem that originates from bulk gene expression analysis.

    “DE testing is a well‐documented problem that originates from bulk gene expression analysis (Scholtens & von Heydebreck, 2005).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Scholtens & von Heydebreck, 2005

      Not checked in the free run.

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

    Premisecitation

    Among command line platforms, Scater and Seurat easily interface with the large variety of analysis tools available via the R Bioconductor project.

    “Among command line platforms, Scater (McCarthy et al, 2017) and Seurat (Butler et al, 2018) easily interface with the large variety of analysis tools available via the R Bioconductor project (Huber et al, 2015).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites McCarthy et al, 2017 source

      Not checked in the free run.

    • cites Butler et al, 2018 source

      Not checked in the free run.

    • cites Huber et al, 2015

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    These platforms are especially useful for exploratory analysis.

    “These platforms are especially useful for exploratory analysis.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Modules of genes that are enriched for regulatory relationships will be more reliable than individual edges.

    “Modules of genes that are enriched for regulatory relationships will be more reliable than individual edges.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    As Web servers, these two platforms are readily available, yet computational infrastructure will limit their ability to scale to large datasets.

    “As Web servers, these two platforms are readily available, yet computational infrastructure will limit their ability to scale to large datasets.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    ASAP was tested on a dataset of only 92 cells.

    “For example, ASAP was tested on a dataset of only 92 cells.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Gardeux et al, 2017

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    Available method comparisons necessarily lag behind the latest method developments.

    “It should be noted that available method comparisons necessarily lag behind the latest method developments.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    Graphical user interface platforms enable non-expert users to build single-cell analysis workflows.

    “Graphical user interface platforms enable non‐expert users to build single‐cell analysis workflows.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationcitation

    As single-cell omic technologies improve, the need for integrated omic analysis pipelines will grow.

    “As single‐cell omic technologies improve, the need for integrated omic analysis pipelines will grow (Tanay & Regev, 2017).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Tanay & Regev, 2017

      Not checked in the free run.

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

    Premisenone given

    It leverages the increasing number of tools written in Python, which is particularly popular for machine learning applications.

    “It leverages the increasing number of tools written in Python, which is particularly popular for machine learning applications.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Alternatives to the Web-based GUI platforms are packages such as FASTGenomics, iSEE, IS-CellR, and Granatum run on a local server.

    “Alternatives to the Web‐based GUI platforms are packages such as FASTGenomics (preprint: Scholz et al, 2018), iSEE (Rue‐Albrecht et al, 2018), IS‐CellR (Patel, 2018), and Granatum run on a local server.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Scholz et al, 2018

      Not checked in the free run.

    • cites Rue‐Albrecht et al, 2018

      Not checked in the free run.

    • cites Patel, 2018

      Not checked in the free run.

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

    Premisenone given

    Scater has a particular strength in QC and pre-processing, while Seurat is arguably the most popular and comprehensive platform, which includes a large array of tools and tutorials.

    “Scater has a particular strength in QC and pre‐processing, while Seurat is arguably the most popular and comprehensive platform, which includes a large array of tools and tutorials.”
    Quote in the paperNot a substantial claim

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

    Premisereasoning

    Users are often guided through prescribed workflows that facilitate the analysis, but also limit user flexibility.

    “Users are often guided through prescribed workflows that facilitate the analysis, but also limit user flexibility.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Due to its flexibility to scale to large data, deep learning has revolutionized fields from computer vision to natural language processing, and is starting to have a strong impact in genomics.

    “Due to its flexibility to scale to large data, deep learning has revolutionized fields from computer vision to natural language processing, and is starting to have a strong impact in genomics (Webb, 2018).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Webb, 2018

      Not checked in the free run.

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

    Premisecitation

    Currently available platforms exist on the command line in R or Python, and as local applications or Web servers with graphical user interfaces (GUIs).

    “Currently available platforms exist on the command line in R (McCarthy et al, 2017; Butler et al, 2018) or Python (Wolf et al, 2018), and as local applications (Patel, 2018; preprint: Scholz et al, 2018) or Web servers (Gardeux et al, 2017; Zhu et al, 2017) with graphical user interfaces (GUIs).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites McCarthy et al, 2017 source

      Not checked in the free run.

    • cites Butler et al, 2018 source

      Not checked in the free run.

    • cites Wolf et al, 2018 source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Single-cell multi-omic integration can occur via consensus clustering approaches, multi-omic factor analysis, or multi-omic gene regulatory network inference.

    “Single‐cell multi‐omic integration can occur via consensus clustering approaches, multi‐omic factor analysis (Argelaguet et al, 2018), or multi‐omic gene regulatory network inference (Colomé‐Tatché & Theis, 2018).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Argelaguet et al, 2018 source

      Not checked in the free run.

    • cites Colomé‐Tatché & Theis, 2018

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Integrated modelling approaches such as this have the potential to replace current pipelines, which are often an agglomeration of individual tools.

    “Integrated modelling approaches such as this have the potential to replace current pipelines, which are often an agglomeration of individual tools.”
    Quote in the paperNot a substantial claim

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

    Premisereasoning

    These platforms provide a basis for the construction of analysis pipelines.

    “These platforms provide a basis for the construction of analysis pipelines.”
    Quote in the paperNot a substantial claim

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

    Premisereasoning

    To facilitate the movement of data between these tools, single-cell platforms have been developed around consistent data formats.

    “To facilitate the movement of data between these tools, single‐cell platforms have been developed around consistent data formats.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    scanpy is a growing Python-based platform, which exhibits improved scaling to larger numbers of cells.

    “A recent addition to this group is scanpy (Wolf et al, 2018), a growing Python‐based platform, which exhibits improved scaling to larger numbers of cells.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Wolf et al, 2018 source

      Not checked in the free run.

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

    Interpretationreasoning

    These are platforms and GUI wrappers that can scale with the locally available computational power.

    “These are platforms and GUI wrappers that can scale with the locally available computational power.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Platforms such as Granatum and ASAP differ in the tools they integrate, with Granatum including the larger variety of methods.

    “Platforms such as Granatum (Zhu et al, 2017) and ASAP (Gardeux et al, 2017) differ in the tools they integrate, with Granatum including the larger variety of methods.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Zhu et al, 2017

      Not checked in the free run.

    • cites Gardeux et al, 2017

      Not checked in the free run.

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

    Premisenone given

    Single-cell analysis workflows are collations of independently developed tools.

    “Single‐cell analysis workflows are collations of independently developed tools.”
    Quote in the paperNot a substantial claim
148 of 148

  • The paper relies on this without support.

    The lack of standardization in scRNA-seq analysis is due to the relative immaturity of the field and is a barrier of entry for novices.

    “Despite considerable effort being undertaken by the field to ensure the usability of individual tools, a barrier of entry for novices in single‐cell data analysis is the lack of standardization due to the relative immaturity of the field.”

  • The authors state this as an assumption.

    CPM normalization assumes that all cells in the dataset initially contained an equal number of mRNA molecules and count depth differences arise only due to sampling.

    “CPM normalization assumes that all cells in the dataset initially contained an equal number of mRNA molecules and count depth differences arise only due to sampling.”

  • The paper relies on this without support.

    Biological processes occur within the same organism and there exist dependencies between these processes.

    “Given that biological processes occur within the same organism, there exist dependencies between these processes.”

  • The paper relies on this without support.

    Single-cell datasets typically consist of heterogeneous cell populations with varying sizes and molecule counts.

    “As single‐cell datasets typically consist of heterogeneous cell populations with varying sizes and molecule counts, more complex normalization methods are usually appropriate.”

  • The paper relies on this without support.

    Raw count matrices often include over 20,000 genes.

    “Raw count matrices often include over 20,000 genes.”

  • The paper relies on this without support.

    A human single-cell RNA-seq dataset can contain expression values for up to 25,000 genes.

    “A human single‐cell RNA‐seq dataset can contain expression values for up to 25,000 genes.”

  • The paper relies on this without support.

    Many genes in a scRNA-seq dataset will not be informative, and many genes will mostly contain zero counts.

    “Many of these genes will not be informative for a given scRNA‐seq dataset, and many genes will mostly contain zero counts.”

  • The paper relies on this without support.

    Even after filtering out zero count genes, the feature space for a single-cell dataset can have over 15,000 dimensions.

    “Even after filtering out these zero count genes in the QC step, the feature space for a single‐cell dataset can have over 15,000 dimensions.”

  • The authors state this as an assumption.

    The obtained cell clusters represent the biological ground truth.

    “Firstly, the P‐values obtained for marker genes are based on the assumption that the obtained cell clusters represent the biological ground truth.”

  • The authors state this as an assumption.

    The clustering is biologically meaningful.

    “Assuming the clustering is biologically meaningful, the top‐ranked marker genes will still be the best marker gene candidates.”

  • The authors state this as an assumption.

    Cells were sampled in an unbiased manner.

    “Assuming that cells were sampled in an unbiased manner, dense regions along a trajectory indicate preferred transcriptomic states.”
11 of 11

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?

  • Without identification and removal of doublets, cell type clustering, differential gene expression analysis, and inference of cell developmental trajectories are severely compromised.

    “Without identification and removal of doublets, cell type clustering, differential gene expression analysis, and inference of cell developmental trajectories are severely compromised [26, 27].”

    From Benchmarking of computational demultiplexing methods for single-nucleus RNA sequencing data (2025) · cites this paper as [26]

    The Quality control section recommends filtering potential doublets before analysis, and the Downstream analysis section identifies clustering and trajectories as later steps. The paper does not say that failing to remove doublets severely compromises clustering, differential expression analysis, or developmental trajectory inference. (LLM, conf=0.83)

    Experimental — accuracy not yet measured

  • Doublets can confound gene expression analysis, cell clustering, and cell type annotation if not considered in downstream analysis.

    “Doublets arising from the inclusion of either two independent nuclei or more nuclei that have become attached during isolation being encapsulated within a GEM resulting in the generation of a single transcriptome can, if not considered in downstream analysis, confound gene expression analysis as well as cell clustering and cell type annotation [26, 27].”

    From Benchmarking of computational demultiplexing methods for single-nucleus RNA sequencing data (2025) · cites this paper as [26]

    The Quality control section says that unusually high counts and gene numbers may indicate doublets and recommends filtering potential doublets before analysis. The Cluster analysis section explains how expression profiles inform clustering and annotation, but does not specifically say that doublets confound those steps. The cited passages also do not describe nuclei becoming attached or their encapsulation within a GEM; no page numbers are available in our copy. (LLM, conf=0.80)

    Experimental — accuracy not yet measured

  • Existing computational best practices and tool recommendations for scRNA-seq are outdated or incomplete.

    “although computational best practices and tool recommendations have previously been outlined for scRNA-seq2,5,6,17, they are either outdated or incomplete.”

    From Best practices for single-cell analysis across modalities (2023) · cites this paper as [5]

  • Community detection based on graph modularity optimization via the Louvain algorithm works best for cluster identification.

    “Independent benchmarks5,50,51 showed that community detection based on graph modularity optimization via the Louvain algorithm works best for cluster identification.”

    From Best practices for single-cell analysis across modalities (2023) · cites this paper as [5]

  • Inferred trajectories might not necessarily have biological meaning.

    “Inferred trajectories might not necessarily have biological meaning5.”

    From Best practices for single-cell analysis across modalities (2023) · cites this paper as [5]

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.

  • UMAP name differs from the cited reference

    Minor

    The text calls UMAP the "Uniform Approximation and Projection method," whereas the cited reference expands its name as "uniform manifold approximation and projection."

  • Regulatory-network method is named SCONE instead of SCODE

    Minor

    The text attributes a regulatory-network method called SCONE to Matsumoto et al. (2017), but the cited reference names the method SCODE.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. Vieth B, Parekh S, Ziegenhain C, Enard W, Hellmann I (2019) A systematic evaluation of single cell RNA‐seq analysis pipelines: library preparation and normalisation methods have the biggest impact on the performance of scRNA‐seq studies. bioRxiv https://doi.org/10.1101/583013 [PREPRINT] · DOI

Cited by

Sorted by citations

Related

Authors
Luecken, M. D., & Theis, F. J.
Affiliations
Helmholtz Munich (DE)
Corresponding author
Theis, Fabian J
ORCID
Venue
Molecular Systems Biology
Publisher
Springer Science and Business Media LLC
Published
2019-06-01
Volume
15
Issue
6
Pages
e8746-e8746
ISSN
1744-4292 (ISSN-L)
DOI
10.15252/msb.20188746
PMC
PMC6582955
OpenAlex
W2949237386
Semantic Scholar
d197473a348efadb23a6bb52de3476fdfdfe7dec
Cited by
2,231
References
168
Citations by year
2026: 410 · 2025: 303 · 2024: 345 · 2023: 318 · 2022: 307 · 2021: 306 · 2020: 205 · 2019: 34 · 2018: 1 · 2015: 1
FWCI
18.81
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Systems Biology (publishedVersion, other-oa)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
  • mediaTUM – the media and publications repository of the Technical University Munich (Technical University Munich) (submittedVersion, other-oa)
  • Site cant be reached (submittedVersion, cc-by)
Funders
  • Bundesministerium für Bildung und Forschung (01IS18036A, 01IS18053A)
  • Deutsche Forschungsgemeinschaft (1243, Subproject A17)
  • Helmholtz Association (ZT‐I‐0007)
  • Silicon Valley Community Foundation (182835)
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 expression and cancer classification
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
Keywords
single-cell RNA sequencing, data analysis pipeline, quality control, normalization, data correction, feature selection, dimensionality reduction, gene-level analysis, best practices
MeSH terms
Guidelines as Topic, Sequence Analysis, RNA, Internet, Gene Expression Profiling, Workflow, Single-Cell Analysis, High-Throughput Nucleotide Sequencing

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 2026-10-10.

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

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