Bioconductor Limma

limma — Comprehensive Skill Guide

Domain: Microarray / RNA-seq Bioconductor: limma Paper: Ritchie ME et al. (2015). Nucleic Acids Research, 43:e47.

Linear models for microarray and RNA-seq data. Enables any experimental design expressible as a linear model; uses empirical Bayes shrinkage of standard errors.

Dependencies & Environment

Package-intrinsic requirements from the Bioconductor landing page — reproduce in any R environment.

• Version: 3.68.3 · Bioconductor: 3.23 · R: ≥ 4.6
• Imports: statmod
• Install: BiocManager::install("limma")

When to Use

• Microarray differential expression
• RNA-seq (via voom transformation)
• Complex designs: factorial, time-course, blocking
• Batch effect correction with duplicateCorrelation
• Rotation test for gene set testing (ROAST/MROAST)

Alternatives: DESeq2, edgeR

Do NOT Use When

• Very small sample sizes (n<3 per group) where Bayesian shrinkage cannot be estimated — use edgeR instead.
• RNA-seq without voom() transformation — raw counts violate limma's Gaussian assumption.
• Non-linear effects or complex time-series without manual design matrix specification.

Data Requirements

Installation

if (!require("BiocManager", quietly=TRUE))
    install.packages("BiocManager")
BiocManager::install("limma")
library(limma)

Workflows

RNA-seq with voom

Standard limma-voom workflow for RNA-seq

Setup

library(limma); library(edgeR)
d <- DGEList(counts=counts)
keep <- filterByExpr(d, group=group)
d <- d[keep,]
d <- calcNormFactors(d)
design <- model.matrix(~group)

voom transform + fit

v <- voom(d, design, plot=TRUE)   # visualise mean-variance trend
fit <- lmFit(v, design)
fit <- eBayes(fit, trend=TRUE)     # trend=TRUE for RNA-seq
topTable(fit, coef=2, n=20)        # top DE genes for coefficient 2

Multiple contrasts

contr.matrix <- makeContrasts(AvsB=A-B, AvsC=A-C, BvsC=B-C, levels=design)
vfit <- contrasts.fit(fit, contr.matrix)
vfit <- eBayes(vfit, trend=TRUE)
results <- decideTests(vfit)       # summary across contrasts
summary(results)

Microarray

Agilent or Affymetrix expression arrays

Load and normalise

library(limma)
targets <- readTargets("targets.txt")
RG <- read.maimages(targets, source="agilent", green.only=TRUE)
RG <- backgroundCorrect(RG, method="normexp")
MA <- normalizeBetweenArrays(RG, method="quantile")

DE analysis

design <- model.matrix(~0 + factor(targets$Treat))
colnames(design) <- c("Control","Treated")
fit <- lmFit(MA, design)
contrasts <- makeContrasts(Treated-Control, levels=design)
fit2 <- contrasts.fit(fit, contrasts)
fit2 <- eBayes(fit2)
topTable(fit2, adjust.method="BH")

Key Functions & Parameters

voom()

Transform RNA-seq counts to log-CPM with precision weights for limma

lmFit()

Fit linear model to each gene

makeContrasts()

Build contrast matrix for pairwise or complex comparisons

contrasts.fit()

Re-parametrise MArrayLM for specified contrasts

eBayes()

Empirical Bayes moderation of t-statistics

topTable()

Extract top DE genes

Scientific Assumptions

The method assumes the following about your data and experiment:

• After voom transformation, the data approximately follows a normal distribution.
• Precision weights from voom account for the mean-variance relationship in RNA-seq counts.
• Empirical Bayes shrinkage of variance toward a common value across genes (eBayes).
• The linear model correctly specifies the experimental design.

Result Interpretation

• logFC: log2 fold-change (or contrast coefficient); same direction convention as edgeR.
• adj.P.Val: BH-FDR adjusted p-value; use < 0.05 as standard cutoff.
• t: moderated t-statistic (smaller SE than ordinary t because of eBayes shrinkage).
• B (log-odds): log-odds of DE; B > 0 means more likely DE than not (rarely used directly).
• AveExpr: average log2-expression across all samples; low value → less reliable estimates.
• Volcano: plotMA(fit2) or volcanoplot(fit2) for MA and volcano visualization.
• Use treat() instead of eBayes() to test against a minimum FC threshold.

Best Practices

• For RNA-seq use voom() + lmFit() + eBayes() workflow.
• Set trend=TRUE in eBayes() for RNA-seq data (accommodates mean-variance trend).
• Use duplicateCorrelation() for paired samples or repeated measures.
• Use removeBatchEffect() only for visualisation — not as input for DE; put batch in design.
• For microarray, quantile-normalise with normalizeBetweenArrays() before lmFit.
• Use fry() or camera() for gene set testing — they account for inter-gene correlation.
• voomWithQualityWeights() improves power when sample quality varies.

When to Choose This vs Alternatives

Benchmark Evidence

Key studies that have validated or benchmarked this tool:

• Ritchie ME et al. (2015). Nucleic Acids Res 43:e47. PMID:25605792
  limma-voom competitive with DESeq2/edgeR for RNA-seq; best for microarray and large n.

• Law CW et al. (2014). Genome Biol 15:R29. PMID:24926665
  voom outperforms DESeq2/edgeR at large sample sizes (n≥20); performs comparably for small n.

Common Errors & Troubleshooting

Coefficients not estimable from factor 'X'

Cause: Aliased coefficients — model is not full rank

Fix: Use nonEstimable(design) to identify redundant columns; simplify design

From Community Q&A (Biostars)

• Contrasts for comparing condition within tissue
• Deseq2 Limma - the paired test seems to be incorrect - is it known problem ?
• limma eBayes error in analysis of Illumina HumanHT-12 V4.0 expression beadchip data

Additional Notes from Official Documentation

Extracted from the limma Bioconductor vignette(s)

What is it?

Limma is an R package for the analysis of gene expression data, especially the use of linear models for analysing designed experiments and the assessment of differential expression. Limma provides the ability to analyse comparisons between many RNA targets simultaneously in arbitrary complicated designed experiments. Empirical Bayesian methods are used to provide stable results even when the number of arrays is small. The normalization and background correction functions are provided for microarrays and similar technologies. The linear model and differential expression functions apply to a wide variety of gene expression technologies including microarrays (single-channel or two-color), quantitative PCR, RNA-seq or proteomics.

How to get help

The edgeR User’s Guide is available by or alternatively from the limma landing page . Documentation for specific functions is available through the usual R help system, e.g., ?lmFit . Further questions about the package should be directed to the Bioconductor support site .

> library(limma)
> limmaRUsersGuide()

Further reading

Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, Smyth GK (2015). limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Research 43, e47. doi:10.1093/nar/gkv007 Phipson B, Lee S, Majewski IJ, Alexander WS, and Smyth GK (2016). Robust hyperparameter estimation protects against hypervariable genes and improves power to detect differential expression. Annals of Applied Statistics 10, 946-963. doi:10.1214/16-AOAS920 Law CW, Chen Y, Shi W, Smyth GK (2014). Voom: precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biology 15, R29. doi:10.1186/gb-2014-15-2-r29 . See also the Preprint Version at https://gksmyth.github.io/pubs/VoomPreprint.pdf incorporating some notational corrections. Law CW, Alhamdoosh M, Su S, Do

Resources

• Bioconductor page: https://bioconductor.org/packages/release/bioc/html/limma.html
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf
• GitHub: https://github.com/Bioconductor/limma
• Bioconductor support forum: https://support.bioconductor.org/
• Paper: Ritchie ME et al. (2015). Nucleic Acids Research, 43:e47.

Run this on BioMate

This skill is the knowledge layer — when, why, and how to use limma. To run this analysis on your own data with managed compute, automated QC, and reproducible outputs, use BioMate — free to start.

▶ Open limma on BioMate →