Perform differential expression analysis using DESeq2 in R/Bioconductor. Use for analyzing RNA-seq count data, creating DESeqDataSet objects, running the DESeq workflow, and extracting results with log fold change shrinkage. Use when performing DE analysis with DESeq2.
Before using code patterns, verify installed versions match. If versions differ:
R: packageVersion('<pkg>') then ?function_name to verify parameters
If code throws ImportError, AttributeError, or TypeError, introspect the installed
package and adapt the example to match the actual API rather than retrying.
DESeq2 Basics
Differential expression analysis using DESeq2 for RNA-seq count data.
Required Libraries
library(DESeq2)
library(apeglm) # For lfcShrink with type='apeglm'
Installation
if (!require('BiocManager', quietly = TRUE))
install.packages('BiocManager')
BiocManager::install('DESeq2')
BiocManager::install('apeglm')
Creating DESeqDataSet
Goal: Construct a DESeqDataSet object from various input formats for DE analysis.
Approach: Wrap count data and sample metadata into the DESeq2 container, specifying the experimental design formula.
"Load my RNA-seq counts into DESeq2" → Create a DESeqDataSet from a count matrix, SummarizedExperiment, or tximport object with sample metadata and a design formula.
From Count Matrix
# counts: matrix with genes as rows, samples as columns
# coldata: data frame with sample metadata (rownames must match colnames of counts)
dds <- DESeqDataSetFromMatrix(countData = counts,
colData = coldata,
design = ~ condition)
Goal: Run the complete DESeq2 pipeline from raw counts to shrunken log fold change estimates.
Approach: Create dataset, pre-filter low-count genes, set reference level, run size factor estimation + dispersion estimation + Wald test, then apply LFC shrinkage.
"Find differentially expressed genes between treated and control" → Test for significant expression changes between conditions using negative binomial models with empirical Bayes shrinkage.
Goal: Extract results for specific pairwise or complex comparisons from a fitted DESeq2 model.
Approach: Use coefficient names or contrast vectors to define which groups to compare.
# See available coefficients
resultsNames(dds)
# Results by coefficient name
res <- results(dds, name = 'condition_treated_vs_control')
# Results by contrast (compare specific levels)
res <- results(dds, contrast = c('condition', 'treated', 'control'))
# Contrast with list format (for complex designs)
res <- results(dds, contrast = list('conditionB', 'conditionA'))
Log Fold Change Shrinkage
Goal: Reduce noisy fold change estimates for low-count genes to improve ranking and visualization.
Approach: Apply empirical Bayes shrinkage (apeglm, ashr, or normal) to moderate log fold changes toward zero.
# apeglm method (default, recommended)
resLFC <- lfcShrink(dds, coef = 'condition_treated_vs_control', type = 'apeglm')
# ashr method (alternative)
resLFC <- lfcShrink(dds, coef = 'condition_treated_vs_control', type = 'ashr')
# normal method (original, less recommended)
resLFC <- lfcShrink(dds, coef = 'condition_treated_vs_control', type = 'normal')
Setting Significance Thresholds
Goal: Control the stringency of differential expression calls using adjusted p-value and fold change cutoffs.
Approach: Set alpha for multiple testing correction and optionally apply a minimum log fold change threshold.
# Default: padj < 0.1
res <- results(dds)
# Custom alpha threshold
res <- results(dds, alpha = 0.05)
# With log fold change threshold
res <- results(dds, lfcThreshold = 1) # |log2FC| > 1
Accessing DESeq2 Results
Goal: Retrieve, filter, and sort DE results for downstream use.
Approach: Extract results as a data frame, subset by significance, and order by p-value or fold change.
# Summary of results
summary(res)
# Get significant genes
sig <- subset(res, padj < 0.05)
# Order by adjusted p-value
resOrdered <- res[order(res$padj),]
# Order by log fold change
resOrdered <- res[order(abs(res$log2FoldChange), decreasing = TRUE),]
# Convert to data frame
res_df <- as.data.frame(res)
Result Columns
Column
Description
baseMean
Mean of normalized counts across all samples
log2FoldChange
Log2 fold change (treatment vs control)
lfcSE
Standard error of log2 fold change
stat
Wald statistic
pvalue
Raw p-value
padj
Adjusted p-value (Benjamini-Hochberg)
Normalization and Counts
Goal: Obtain normalized expression values suitable for visualization and cross-sample comparison.
Approach: Extract size-factor-normalized counts or apply variance-stabilizing / rlog transformations.
