limma powers differential expression analyses for RNA-sequencing and microarray studies

May combine two claims.

limma contains features for handling complex experimental designs and for information borrowing to overcome small sample sizes.

“It contains rich features for handling complex experimental designs and for information borrowing to overcome the problem of small sample sizes.”

• AssumptionRests on an assumption: Gene expression studies often involve only a small number of biological replicates.

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

All features of limma's statistical models can be accessed for gene-wise expression analyses and for higher level analyses of gene expression signatures.

“All the features of the statistical models can be accessed not just for gene-wise expression analyses but also for higher level analyses of gene expression signatures.”

Classified as probably not a substantial claim.

limma is designed so that, after initial pre-processing and normalization, the same analysis pipeline is used for data from all technologies.

“The package is designed in such a way that, after initial pre-processing and normalization, the same analysis pipeline is used for data from all technologies.”

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

limma has proven a popular choice for analysis of data from experiments involving microarrays, high-throughput PCR, protein arrays and other platforms.

“It has proven a popular choice for the analysis of data from experiments involving microarrays (3,4), high-throughput polymerase chain reaction (PCR) (5), protein arrays (6) and other platforms.”

• cites [3] Peart, 2005, Identification and functional significance of genes regulated by structurally different histone deacetylase inhibitors, Proc. Natl. Acad. Sci. U.S.A. sourceNot checked yet

  Not checked in the free run.

• cites [4] Caiazzo, 2011, Direct generation of functional dopaminergic neurons from mouse and human fibroblasts, Nature sourceNot checked yet

  Not checked in the free run.

• cites [5] Hubert, 2009, Aire-deficient c57bl/6 mice mimicking the common human 13-base pair deletion mutation present with only a mild autoimmune phenotype, J. Immunol. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

limma can perform both differential expression and differential splicing analyses of RNA sequencing (RNA-seq) data.

“First, the package can now perform both differential expression and differential splicing analyses of RNA sequencing (RNA-seq) data.”

• Also stated in: § INTRODUCTION · plain

  “First, the package can now perform both differential expression (DE) and differential splicing analyses of RNA sequencing (RNA-seq) data (7,8).”

• cites [7] Liu, 2014, Pax5 loss imposes a reversible differentiation block in B progenitor acute lymphoblastic leukemia, Genes Dev. sourceNot checked yet

  Not checked in the free run.

• cites [8] Su, 2014, A comprehensive assessment of RNA-seq accuracy, reproducibility and information content by the Sequencing Quality Control Consortium, Nat. Biotechnol. sourceNot checked yet

  Not checked in the free run.

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

limma includes pre-processing methods such as variance stabilization to reduce noise.

“(iv) pre-processing methods such as variance stabilization to reduce noise.”

May combine two claims.

The voom function converts the mean-variance trend into precision weights, which are incorporated into the analysis of log-transformed RNA-seq counts using the same linear modelling commands as for microarrays.

“The mean-variance trend is converted by the voom function into precision weights, which are incorporated into the analysis of log-transformed RNA-seq counts using the same linear modelling commands as for microarrays.”

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

Output in other formats can be read if the appropriate column names are supplied.

“Output in other formats can be read if the appropriate column names are supplied.”

May combine two claims.

Weights can be used to give more emphasis to control probes during normalization, or to down-weight less reliable measurements or samples.

“Weights can be used to give more emphasis to control probes during normalization, or can be used to down-weight measurements or samples that are less reliable in a gene expression analysis.”

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

read.maimages directly supports formats written by many different image analysis programs including GenePix, Agilent Feature Extraction, ArrayVision, BlueFuse, ImaGene, QuantArray and SPOT.

“This function directly supports formats written by many different image analysis programs including GenePix, Agilent Feature Extraction, ArrayVision, BlueFuse, ImaGene, QuantArray and SPOT (Table 1).”

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

The logCPM values can be normalized between samples by the voom function or can be pre-normalized by adding normalization factors within edgeR.

“The logCPM values can be normalized between samples by the voom function or can be pre-normalized by adding normalization factors within edgeR.”

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

Raw read counts are assembled outside limma using tools such as featureCounts, HTSeq-counts or RSEM.

“Raw read counts are assembled outside limma using tools such as featureCounts (29), HTSeq-counts (30) or RSEM (31).”

• cites [29] Liao, 2014, featureCounts: an efficient general-purpose read summarization program, Bioinformatics sourceNot checked yet

  Not checked in the free run.

• cites [30] Anders, 2015, HTSeq—a Python framework to work with high-throughput sequencing data, Bioinformatics sourceNot checked yet

  Not checked in the free run.

• cites [31] Li, 2011, RSEM: accurate transcript quantification from RNA-Seq data with or without a reference genome, BMC Bioinformatics sourceNot checked yet

  Not checked in the free run.

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

read.maimages also supports the Stanford Microarray Database format.

“It also supports the Stanford Microarray Database format.”

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

These ideas were part of the original limma package submitted to Bioconductor in 2003.

“These ideas were part of the original limma package submitted to Bioconductor in 2003.”

May combine two claims.

Linear models permit very general analyses, including adjusting for multiple experimental factors, batch effects, time course effects, regression splines, gene expression covariates, and testing interaction effects or complex customized comparisons.

“Linear models permit very general analyses. Researchers can adjust for the effects of multiple experimental factors or can adjust for batch effects. The linear model might include time course effects or regression splines. The linear model could even include the expression values themselves of one or more genes as covariates, allowing researchers to test for inter-gene dependencies. Linear models”

May combine two claims.

Weights can be applied to genes, RNA samples, or individual expression values.

“Weights can be applied to genes or to RNA samples or to individual expression values.”

May combine two claims. The quote only partly states this claim.

limma generalized the concept of an MA-plot to apply to sets of single-channel expression values and to fitted model objects.

“limma generalized the concept of an MA-plot in two ways. First, the idea was extended to apply to sets of single-channel expression values. In this case, the plot is used to compare each sample to the average of all other samples. A virtual array is constructed by averaging the log-expression value for all the samples other than the sample of interest, and then a mean-difference plot is made”

May combine two claims.

limma implements global covariance models to estimate correlations between genes or the relatedness between the DE profiles resulting from different comparisons.

“A closely related statistical approach implemented in limma is to fit global covariance models, either to estimate correlations between genes or to estimate the relatedness between the DE profiles resulting from difference comparisons.”

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

Two-colour and single-channel data are both supported.

“Two-colour and single-channel data are both supported.”

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

The nec function implements normexp background correction for Illumina BeadChips using control probes specific to these arrays.

“The nec function implements normexp background correction for Illumina BeadChips making special use of the control probes that are specific to these arrays (24).”

• cites [24] Shi, 2010, Optimizing the noise versus bias trade-off for Illumina Whole Genome Expression Beadchips, Nucleic Acids Res. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

This strategy allows gene-wise models to incorporate correlations between duplicate probes, correlations between related RNA samples, or variations in quality between RNA samples.

“This strategy allows the gene-wise models to incorporate such things as correlations between duplicate probes for the same gene, or correlations between related RNA samples, or variations in quality between the RNA samples.”

May combine two claims.

limma includes a maximum likelihood implementation of the normal-exponential convolution model for background correction and an implementation of loess curves and normalization using quantitative weights.

“Notable are the maximum likelihood implementation of the normal-exponential convolution model for background correction (22) and the implementation of loess curves and normalization using quantitative weights.”

• cites [22] Silver, 2009, Microarray background correction: maximum likelihood estimation for the normal–exponential convolution, Biostatistics sourceNot checked yet

  Not checked in the free run.

May combine two claims.

One way is through estimating a mean-variance trend, which can either be incorporated into the empirical Bayes procedure or used to generate observation weights.

“One way is through estimating a mean-variance trend, which can either be incorporated into the empirical Bayes procedure as mentioned above or used to generate observation weights (10).”

• cites [10] Law, 2014, Voom: precision weights unlock linear model analysis tools for RNA-seq read counts, Genome Biol. sourceNot checked yet

  Not checked in the free run.

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

The main limma function to read image output files is read.maimages.

“The main limma function to read image output files is read.maimages.”

May combine two claims.

A separate model is fitted for each gene, but the gene-wise models can be linked by global parameters or global hyper-parameters.

“A separate model is fitted for each gene, but the gene-wise models can be linked by global parameters or global hyper-parameters.”

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

The voom function converts read counts into log2 counts per million with associated precision weights.

“The read counts are processed by the voom function in limma to convert them into log2 counts per million (logCPM) with associated precision weights.”

May combine two claims.

Use of global parameters shares information between genes and can be used even for the smallest experiments because the global parameters can be estimated from the entire data set.

“The use of global parameters is a simple means of sharing information between genes that can be used even for the smallest experiments, because the global parameters can be estimated from the entire data set involving all the genes at once.”

May combine two claims.

This allows a robust empirical Bayes procedure in which hyper-variable genes are identified and treated separately.

“This allows a sophisticated robust empirical Bayes procedure in which hyper-variable genes are identified and treated separately (18,19).”

• cites [18] Phipson, 2013, Empirical Bayes in the presence of exceptional cases, with application to microarray dataNot checked yet

  Not checked in the free run.

• cites [19] Phipson, 2013, Empirical Bayes modelling of expression profiles and their associationsNot checked yet

  Not checked in the free run.

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

RNA-seq data can be explored using boxplots or mean-difference plots, similarly to single-channel microarray data.

“For example, RNA-seq data can be explored using boxplots or mean-difference plots, similarly to single-channel microarray data.”

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

Illumina BeadChips need special treatment: output from Illumina's GenomeStudio can be read by read.ilmn if exported as a text file or by read.idat if in binary format.

“Illumina BeadChips need special treatment: output from Illumina's GenomeStudio can be read by read.ilmn if exported as a text file or by read.idat if in binary format.”

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

limma can accept a DGEList object from the edgeR package.

“Alternatively it can accept a DGEList object from the edgeR package.”

May combine two claims.

Use of weights increases power to detect differentially expressed genes, and a model-based approach avoids ad hoc decisions about which observations or samples to filter out.

“The use of weights increases power to detect differentially expressed genes, and having a model based approach avoids the need for ad hoc decisions about which observations or samples to filter out (11).”

• cites [11] Ritchie, 2006, Empirical array quality weights in the analysis of microarray data, BMC Bioinformatics sourceNot checked yet

  Not checked in the free run.

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

limma accepts RNA-seq data in the form of a matrix of read counts, with rows for genomic features and columns for RNA samples.

“limma accepts RNA-seq data in the form of a matrix of read counts, with rows for genomic features and columns for RNA samples.”

May combine two claims.

The resulting pipeline gives comparable performance to the best negative binomial-based software packages but with greater speed and reliability for large data sets.

“The resulting pipeline gives comparable performance to the best of the negative binomial-based software packages but with greater speed and reliability for large data sets (10,21).”

• cites [10] Law, 2014, Voom: precision weights unlock linear model analysis tools for RNA-seq read counts, Genome Biol. sourceNot checked yet

  Not checked in the free run.

• cites [21] Rapaport, 2013, Comprehensive evaluation of differential gene expression analysis methods for RNA-seq data, Genome Biol. sourceNot checked yet

  Not checked in the free run.

May combine two claims.

Analysing data as a whole shares information between samples and allows modelling correlations between samples due to repeated measures or other causes.

“This has the effect of sharing information between samples. Analysing the data as a whole also allows us to model correlations that may exist between samples due to repeated measures or other causes.”

May combine two claims.

Weights can be preset based on external quality information or estimated from the expression data itself.

“The weights can be preset based on external quality information, or may be estimated from the expression data itself.”

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

limma is the only software package to provide methods for error rate control across multiple contrasts and genes simultaneously.

“limma is the only software package to provide methods for error rate control across multiple contrasts and genes simultaneously.”

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

Figure 3A shows that total variability asymptotes to biological variability as count sizes increase.

“Total variability asymptotes to biological variability as count sizes increases.”

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

The propexpr function estimates the total proportion of probes on each array that correspond to expressed genes, providing an estimate of transcriptome size.

“The propexpr function compares intensities to those of negative control probes to estimate the total proportion of probes on each array that correspond to expressed genes (34).”

• cites [34] Shi, 2010, Estimating the proportion of microarray probes expressed in an RNA sample, Nucleic Acids Res. sourceNot checked yet

  Not checked in the free run.

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

Many limma functions are generic or operate appropriately on objects of different classes.

“Many limma functions are generic or operate appropriately on objects of different classes.”

May combine two claims.

limma provides the possibility of analysing two-colour microarrays as if they were single channel microarrays with two separate samples hybridized to each physical array, allowing intensities to be directly compared between microarrays.

“limma also provides the possibility of analysing two-colour microarrays as if they were single channel microarrays with two separate samples hybridized to each physical array. This provides a very powerful type of analysis in which intensities can be directly compared between microarrays.”

May combine two claims.

The treat function tests whether the log2-fold-change is greater than a threshold rather than merely different to zero, which can be effective for prioritizing biologically as well as statistically significant results.

“the treat function can be used to test whether the log2-fold-change is greater than a threshold rather than merely different to zero (45). This can be effective for prioritizing results that are biologically as well as statistically significant.”

• cites [45] McCarthy, 2009, Testing significance relative to a fold-change threshold is a TREAT, Bioinformatics sourceNot checked yet

  Not checked in the free run.

May combine two claims.

All linear model fits in limma produce a fitted model object with the same structure, allowing the same framework for DE to be used for all experimental designs and platform technologies.

“All the linear model fits, whether using lmFit or lmscFit, produce a fitted model object with the same structure.”

• Also stated in: § Finding differentially expressed genes · plain

  “This consistency allows the same framework for DE to be used for all experimental designs and platform technologies.”

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

The neqc function implements quantile normalization for Illumina BeadChips using control probes specific to these arrays.

“The neqc function implements quantile normalization for Illumina BeadChips making special use of the control probes specific to these arrays (24).”

• cites [24] Shi, 2010, Optimizing the noise versus bias trade-off for Illumina Whole Genome Expression Beadchips, Nucleic Acids Res. sourceNot checked yet

  Not checked in the free run.

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

Figure 4C shows a strong positive concordance between Pax5 restoration and the large to small cell transition during normal B cell development.

“The plot shows a strong positive concordance between Pax5 restoration and the large to small cell transition.”

• cites [7] Liu, 2014, Pax5 loss imposes a reversible differentiation block in B progenitor acute lymphoblastic leukemia, Genes Dev. sourceNot checked yet

  Not checked in the free run.

• cites [47] Hoffmann, 2002, Changes in gene expression profiles in developing b cells of murine bone marrow, Genome Res. sourceNot checked yet

  Not checked in the free run.

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

The memory requirements of limma are linear in the number of genes and the number of samples.

“The memory requirements are linear in the number of genes and the number of samples.”

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

Barcodeplot is similar to the set location plot introduced by Subramanian et al., with enhancements including the ability to incorporate genes with positive and negative prior directions.

“Barcodeplot is similar to the set location plot introduced by Subramanian et al. (63), with a number of enhancements, especially the ability to incorporate genes with positive and negative prior directions.”

• cites [63] Subramanian, 2005, Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles, Proc. Natl. Acad. Sci. U.S.A. sourceNot checked yet

  Not checked in the free run.

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

The propTrueNull function estimates the proportion of true null hypotheses in a collection of hypothesis tests given a vector of P-values.

“The propTrueNull function estimates the number of truly differentially expressed genes that remain to be identified. Mathematically, it estimates the proportion of true null hypotheses in a collection of hypothesis tests given a vector of P-values.”

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

The romer function in limma implements a GSEA approach that is based on rotation instead of permutation.

“The romer function in limma implements a GSEA approach that is based on rotation instead of permutation.”

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

The romer function can be used with a battery of gene sets and with any linear model.

“Like camera and mroast, it can be used with a battery of gene sets and with any linear model.”

May combine two claims.

limma includes a robustified shrinkage strategy that allows gene-wise shrinkage factors to be estimated, ensuring unusually large variances are not squeezed too heavily, reducing the chance they appear statistically significant, while more consistently expressed genes are squeezed more severely.

“limma includes a robustified shrinkage strategy that allows for gene-wise shrinkage factors to be estimated (18). This ensures unusually large variances are not squeezed too heavily, reducing the chance that they will appear statistically significant, while more consistently expressed genes are squeezed more severely towards the common variance.”

• cites [18] Phipson, 2013, Empirical Bayes in the presence of exceptional cases, with application to microarray dataNot checked yet

  Not checked in the free run.

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

Figure 3A shows that technical variability decreases with count size for RNA-seq data.

“This plot shows that technical variability decreases with count size.”

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

The limma authors maintain mouse and human versions of the Molecular Signatures Database collections in R binary format that can be conveniently used with camera, mroast or romer.

“The limma authors maintain mouse and human versions of the Molecular Signatures Database collections (64) in R binary format that can be conveniently used with camera, mroast or romer”

• cites [64] Liberzon, 2011, Molecular signatures database (MSigDB) 3.0, Bioinformatics sourceNot checked yet

  Not checked in the free run.

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

The limma software is freely available online as part of the Bioconductor project.

“The limma software is freely available online as part of the Bioconductor project (http://www.bioconductor.org).”

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

Most estimation procedures in limma finish in a few seconds on a standard desktop computer and virtually all in less than a minute.

“Most estimation procedures finish in a few seconds on a standard desktop computer and virtually all in less than a minute.”

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

The geneSetTest and wilcoxGST functions perform rank-based tests that assess whether a specified set of genes is more highly ranked in an ordered list of all genes than would be expected by chance.

“The simplest approach is implemented in the geneSetTest and wilcoxGST functions, which perform rank-based tests (13). These tests assess whether the specified set of genes is more highly ranked in an ordered list of all genes than would be expected by chance.”

• cites [13] Michaud, 2008, Integrative analysis of runx1 downstream pathways and target genes, BMC Genomics sourceNot checked yet

  Not checked in the free run.

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

Each function in limma has its own documentation page that specifies the input data, options and output format of the function.

“First, each function has its own documentation page that concisely but completely specifies the input data, options and output format of the function.”

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

All of the data classes in limma obey many analogies with matrices.

“All of these data classes obey many analogies with matrices.”

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

Camera has been used successfully in a number of biomedical projects.

“Camera has been used successfully in a number of biomedical projects (57).”

• cites [57] Wu, 2013, Gene-expression data integration to squamous cell lung cancer subtypes reveals drug sensitivity, Br. J. Cancer sourceNot checked yet

  Not checked in the free run.

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

The plotMDS function uses multi-dimensional scaling to plot differences in expression profiles between samples, with distances representing 'leading fold change' defined as the root-mean-square average of log-fold-changes for the genes best distinguishing each pair of samples.

“The plotMDS function uses multi-dimensional scaling to plot differences in expression profiles between different samples (Figure 3C). Distances between samples on the plot represent ‘leading fold change’, which is defined as the root-mean-square average of the log-fold-changes for the genes best distinguishing each pair of samples.”

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

All between-arrays normalization methods are accessible for RNA-seq data from within the voom function.

“All the between-arrays normalization methods are accessible for RNA-seq data from within the voom function.”

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

More than 120 other Bioconductor packages make use of limma as of March 2014.

“More than 120 other Bioconductor packages make use of limma (as of March 2014).”

May combine two claims.

limma includes enhancements: ability to replace the loess curve with a spline curve with high robustness breakdown properties, and ability to apply empirical Bayes moderation to spline curves for multiple regions within the same array.

“Other enhancements include the ability to replace the loess curve with a spline curve that has high robustness breakdown properties, and the ability to apply empirical Bayes moderation to the spline curves for multiple regions within the same array (robust-spline normalization).”

May combine two claims.

The camera function computes a variance inflation factor from the inter-gene correlation and uses this to adjust the variance of the summary statistics, avoiding optimistic P-values but reducing statistical power.

“Camera computes a variance inflation factor from the inter-gene correlation and uses this to adjust the variance of the summary statistics. This avoids optimistic P-values in the test results, but also reduces statistical power.”

• cites [56] Wu, 2012, Camera: a competitive gene set test accounting for inter-gene correlation, Nucleic Acids Res. sourceNot checked yet

  Not checked in the free run.

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

The limma package is implemented primarily in R and includes some C code to speed up computationally intensive steps.

“The limma package is implemented primarily in R (65) and includes some C code to speed up computationally intensive steps.”

• cites [65] R Development Core Team, 2014Not checked yet

  Not checked in the free run.

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

The limma package provides three levels of documentation.

“The limma package provides three levels of documentation.”

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

limma defines a number of classes that have been tailored to handle both microarray and RNA-seq data.

“limma defines a number of classes that have been tailored to handle both microarray and RNA-seq data.”

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

Applications of limma's linear modelling strategy beyond the intended analysis of gene expression data have been made in a variety of applications, including the analysis of data from Nuclear Magnetic Resonance spectroscopy, PCR (including Nanostring), quantitative proteomics, DNA methylation arrays and comparative ChIP-seq.

“Applications of limma's linear modelling strategy beyond the intended analysis of gene expression data have been made in a variety of applications, including the analysis of data from Nuclear Magnetic Resonance spectroscopy, PCR (including Nanostring), quantitative proteomics (80), DNA methylation arrays and comparative ChIP-seq (81).”

• cites [80] Brusniak, 2008, Corra: computational framework and tools for LC-MS discovery and targeted mass spectrometry-based proteomics, BMC Bioinformatics sourceNot checked yet

  Not checked in the free run.

• cites [81] Lun, 2014, De novo detection of differentially bound regions for ChIP-seq data using peaks and windows: controlling error rates correctly, Nucleic Acids Res. sourceNot checked yet

  Not checked in the free run.

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

Each data class in limma has a documentation page explaining all the required and optional components of objects of that class.

“Similarly, each data class has a documentation page explaining all the required and optional components of objects of that class.”

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

The limma package comes with an extensive user's guide of over 120 pages.

“Third, the package comes with an extensive user's guide of over 120 pages, available from the drop-down menu in Windows or alternatively launched by the limmaUsersGuide command.”

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

The limma user's guide includes 10 fully worked case studies for which full data and code are provided.

“It also includes 10 fully worked case studies for which full data and code are provided.”

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

The Bioconductor support site archives answers to many common questions, including many queries about experimental design and setting up appropriate design matrices.

“The support site archives answers to many common questions, including many queries about experimental design and setting up appropriate design matrices.”

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

A series of more general subject help pages in limma link together functions and classes that are used for related purposes.

“Second, a series of more general subject help pages serve to link together functions and classes that are used for related purposes.”

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

The limma user's guide gives detailed advice on how to analyse a variety of common study designs.

“The user's guide gives detailed advice on how to analyse a variety of common study designs.”

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

limma is a core component of Bioconductor.

“The limma package is a core component of Bioconductor, an R-based open-source software development project in statistical genomics (1,2).”

• cites [1] Gentleman, 2004, Bioconductor: open software development for computational biology and bioinformatics, Genome Biol. sourceNot checked yet

  Not checked in the free run.

• cites [2] Smyth, 2005, Limma: linear models for microarray data sourceNot checked yet

  Not checked in the free run.

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

Simply subtracting background from foreground intensities is too heavy-handed.

“Simply subtracting background from foreground intensities is too heavy-handed (23).”

• cites [23] Ritchie, 2007, A comparison of background correction methods for two-colour microarrays, Bioinformatics sourceNot checked yet

  Not checked in the free run.

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

limma's backgroundCorrect function offers a range of more sophisticated alternatives to simple background subtraction, most unique to the package.

“The limma backgroundCorrect function offers a range of more sophisticated alternatives, most unique to the package.”

Classified as probably not a substantial claim.

Traditionally, RNA-seq data require specialized software based on the negative binomial or similar distributions.

“Traditionally, RNA-seq data require specialized software based on the negative binomial or similar distributions (20).”

• cites [20] Robinson, 2010, edgeR: a Bioconductor package for differential expression analysis of digital gene expression data, Bioinformatics sourceNot checked yet

  Not checked in the free run.

May combine two claims.

If permutation is applied to all samples in a multi-factor experiment, the composite null hypothesis being tested is uninteresting and the power to reject it may be highly dependent on the existence of differential expression between treatment conditions that are not of primary interest.

“If permutation is applied to all the samples in a multi-factor experiment, then the composite null hypothesis being testing is an uninteresting one and the power to reject it may be highly dependent on the existence of DE between treatment conditions that are not of primary interest.”

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

The plotMDS plot provides unsupervised clustering of samples and is useful for identifying unexpected patterns such as batch effects that should be adjusted for.

“This provides a type of unsupervised clustering of the samples. It is useful for examining how different are the profiles produced by different experimental factors and for identifying unexpected patterns, such as batch effects, that should be adjusted for during the linear model analysis.”

May combine two claims.

The ability to use quantitative weights in loess normalization allows downweighing less reliable probes or giving higher priority to control probes or house-keeping genes.

“giving it the ability to downweigh less reliable probes or to give higher priority to control probes or house-keeping genes.”

May combine two claims.

The flexibility of the linear modelling approach allows almost any experimental design to be handled, including experiments with two or more groups, factorial and time-course designs, and internal controls such as dye-swaps.

“The flexibility of the linear modelling approach allows almost any experimental design to be handled. Experiments with two or more groups, factorial and time-course designs, and internal controls such as dye-swaps can all be modelled and summarized using the lmFit function.”

May combine two claims.

The robust strategy offers the benefits of shrinkage to the majority of genes while negating the effects of outliers.

“This robust strategy offers the benefits of shrinkage to the majority of the genes, whilst negating the effects of outliers.”

May combine two claims.

The empirical Bayes framework reduces the number of false positives for genes with very small variances and improves power to detect DE for genes with larger variances.

“which reduces the number of false positives for genes with very small variances and improves power to detect DE for genes with larger variances.”

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

The differential splicing approach is considerably faster than alternative approaches, making large-scale surveys of differential exon usage feasible.

“This approach is considerably faster than alternative approaches to differential splicing, making large-scale surveys of differential exon usage feasible.”

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

Different technology platforms introduce different biases and so require different normalization methods.

“Different technology platforms introduce different biases and so require different normalization methods.”

May combine two claims.

These rank-based tests implicitly assume that the expression level of each gene is conditionally independent of other genes and hence give optimistic P-values.

“but they implicitly assume that the expression level of each gene is conditionally independent of other genes and hence give optimistic P-values (55).”

• cites [55] Wu, 2010, ROAST: rotation gene set tests for complex microarray experiments, Bioinformatics sourceNot checked yet

  Not checked in the free run.

• AssumptionThe authors state this as an assumption: The geneSetTest and wilcoxGST rank-based tests assume that the expression level of each gene is conditionally independent of other genes.

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

The development of the voom methodology unlocks the majority of analysis methods for use on RNA-seq data, such as random effects modelling and gene set testing.

“Although originally developed with microarray data in mind, the development of the voom methodology unlocks the majority of analysis methods for use on RNA-seq data, such as random effects modelling and gene set testing.”