Extract, filter, annotate, and export differential expression results from DESeq2 or edgeR. Use for identifying significant genes, applying multiple testing corrections, adding gene annotations, and preparing results for downstream analysis. Use when filtering and exporting DE analysis results.
Reference examples tested with: DESeq2 1.42+, edgeR 4.0+
Before using code patterns, verify installed versions match. If versions differ:
packageVersion('<pkg>') then ?function_name to verify parametersIf code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.
Extract, filter, and export differential expression results.
library(DESeq2) # or library(edgeR)
library(dplyr) # For data manipulation
Goal: Retrieve DE statistics from a fitted DESeq2 model as a usable data frame.
Approach: Call results() with optional shrinkage, then convert to a data frame with gene identifiers.
# Basic results
res <- results(dds)
# With specific alpha (adjusted p-value threshold)
res <- results(dds, alpha = 0.05)
# With log fold change shrinkage
res <- lfcShrink(dds, coef = 'condition_treated_vs_control', type = 'apeglm')
# Convert to data frame
res_df <- as.data.frame(res)
res_df$gene <- rownames(res_df)
Goal: Retrieve DE statistics from a fitted edgeR model as a data frame.
Approach: Use topTags with n=Inf to extract all gene-level results.
# Get all results
results <- topTags(qlf, n = Inf)$table
# Add gene column
results$gene <- rownames(results)
Goal: Identify genes meeting statistical significance and biological effect size criteria.
Approach: Subset results by adjusted p-value, fold change magnitude, and expression level thresholds.
"Get the significant differentially expressed genes" → Filter DE results by adjusted p-value and fold change cutoffs to produce up- and down-regulated gene lists.
# DESeq2
sig_genes <- subset(res, padj < 0.05)
# edgeR
sig_genes <- subset(results, FDR < 0.05)
# Using dplyr
sig_genes <- res_df %>%
filter(padj < 0.05) %>%
arrange(padj)
# Absolute log2 fold change > 1 (2-fold change)
sig_genes <- subset(res, padj < 0.05 & abs(log2FoldChange) > 1)
# Up-regulated only
up_genes <- subset(res, padj < 0.05 & log2FoldChange > 1)
# Down-regulated only
down_genes <- subset(res, padj < 0.05 & log2FoldChange < -1)
# Stringent filtering
sig_genes <- res_df %>%
filter(padj < 0.01,
abs(log2FoldChange) > 1,
baseMean > 10) %>%
arrange(padj)
Goal: Rank DE genes by statistical significance or biological effect size.
Approach: Sort results by adjusted p-value, absolute fold change, or mean expression.
# By adjusted p-value (most significant first)
res_ordered <- res[order(res$padj), ]
# By absolute fold change (largest changes first)
res_ordered <- res[order(abs(res$log2FoldChange), decreasing = TRUE), ]
# By base mean expression
res_ordered <- res[order(res$baseMean, decreasing = TRUE), ]
# Combined: significant genes ordered by fold change
sig_ordered <- res_df %>%
filter(padj < 0.05) %>%
arrange(desc(abs(log2FoldChange)))
Goal: Quantify the number of up- and down-regulated genes at chosen thresholds.
Approach: Count genes passing significance filters and report directional breakdown.
# DESeq2 summary
summary(res)
# Manual counts
n_tested <- sum(!is.na(res$padj))
n_sig <- sum(res$padj < 0.05, na.rm = TRUE)
n_up <- sum(res$padj < 0.05 & res$log2FoldChange > 0, na.rm = TRUE)
n_down <- sum(res$padj < 0.05 & res$log2FoldChange < 0, na.rm = TRUE)
cat(sprintf('Tested: %d genes\n', n_tested))
cat(sprintf('Significant (padj < 0.05): %d genes\n', n_sig))
cat(sprintf('Up-regulated: %d genes\n', n_up))
cat(sprintf('Down-regulated: %d genes\n', n_down))
# edgeR summary
summary(decideTests(qlf))
Goal: Enrich DE results with gene symbols, descriptions, and cross-database identifiers.
Approach: Map Ensembl or Entrez IDs to human-readable annotations using org.db, biomaRt, or custom files.
"Add gene names to my DE results" → Map gene identifiers to symbols and descriptions using annotation databases, then merge with the results table.
library(org.Hs.eg.db) # Human; use org.Mm.eg.db for mouse
# If gene IDs are Ensembl
res_df$symbol <- mapIds(org.Hs.eg.db,
keys = rownames(res_df),
column = 'SYMBOL',
keytype = 'ENSEMBL',
multiVals = 'first')
res_df$entrez <- mapIds(org.Hs.eg.db,
keys = rownames(res_df),
column = 'ENTREZID',
keytype = 'ENSEMBL',
multiVals = 'first')
res_df$description <- mapIds(org.Hs.eg.db,
keys = rownames(res_df),
column = 'GENENAME',
keytype = 'ENSEMBL',
multiVals = 'first')
library(biomaRt)
mart <- useMart('ensembl', dataset = 'hsapiens_gene_ensembl')
annotations <- getBM(
attributes = c('ensembl_gene_id', 'external_gene_name', 'description'),
filters = 'ensembl_gene_id',
values = rownames(res_df),
mart = mart
)
# Merge with results
res_annotated <- merge(res_df, annotations,
by.x = 'row.names', by.y = 'ensembl_gene_id',
all.x = TRUE)
# Load annotation file
gene_info <- read.csv('gene_annotations.csv')
# Merge with results
res_annotated <- merge(res_df, gene_info, by = 'gene', all.x = TRUE)
Goal: Save DE results in formats suitable for sharing, publication, or downstream tools.
Approach: Write filtered and annotated results to CSV, Excel workbooks, or ranked gene lists for pathway analysis.
# All results
write.csv(res_df, file = 'deseq2_all_results.csv', row.names = FALSE)
# Significant only
sig_genes <- res_df %>% filter(padj < 0.05)
write.csv(sig_genes, file = 'deseq2_significant.csv', row.names = FALSE)
library(openxlsx)
# Create workbook with multiple sheets
wb <- createWorkbook()
addWorksheet(wb, 'All Results')
writeData(wb, 'All Results', res_df)
addWorksheet(wb, 'Significant')
writeData(wb, 'Significant', sig_genes)
addWorksheet(wb, 'Up-regulated')
writeData(wb, 'Up-regulated', up_genes)
addWorksheet(wb, 'Down-regulated')
writeData(wb, 'Down-regulated', down_genes)
saveWorkbook(wb, 'de_results.xlsx', overwrite = TRUE)
# Just gene IDs for GO/KEGG analysis
sig_gene_list <- rownames(subset(res, padj < 0.05))
write.table(sig_gene_list, file = 'significant_genes.txt',
quote = FALSE, row.names = FALSE, col.names = FALSE)
# With fold changes for GSEA
gsea_input <- res_df %>%
filter(!is.na(log2FoldChange)) %>%
select(gene, log2FoldChange) %>%
arrange(desc(log2FoldChange))
write.table(gsea_input, file = 'gsea_input.rnk',
sep = '\t', quote = FALSE, row.names = FALSE, col.names = FALSE)
Goal: Assess concordance between DESeq2 and edgeR results to identify robust DE genes.
Approach: Compute set overlaps and visualize with a Venn diagram.
# Get significant genes from both methods
deseq2_sig <- rownames(subset(deseq2_res, padj < 0.05))
edger_sig <- rownames(subset(edger_results, FDR < 0.05))
# Overlap
common <- intersect(deseq2_sig, edger_sig)
deseq2_only <- setdiff(deseq2_sig, edger_sig)
edger_only <- setdiff(edger_sig, deseq2_sig)
cat(sprintf('DESeq2 significant: %d\n', length(deseq2_sig)))
cat(sprintf('edgeR significant: %d\n', length(edger_sig)))
cat(sprintf('Common: %d\n', length(common)))
cat(sprintf('DESeq2 only: %d\n', length(deseq2_only)))
cat(sprintf('edgeR only: %d\n', length(edger_only)))
# Venn diagram
library(VennDiagram)
venn.diagram(
x = list(DESeq2 = deseq2_sig, edgeR = edger_sig),
filename = 'de_overlap.png',
fill = c('steelblue', 'coral')
)
Goal: Apply or compare multiple testing correction methods for DE p-values.
Approach: Use Benjamini-Hochberg (default), Bonferroni, or IHW for adjusted p-values.
# DESeq2 uses Benjamini-Hochberg by default
# To use different methods:
# Independent Hypothesis Weighting (more powerful)
library(IHW)
res_ihw <- results(dds, filterFun = ihw)
# Manual p-value adjustment
res_df$padj_bonferroni <- p.adjust(res_df$pvalue, method = 'bonferroni')
res_df$padj_bh <- p.adjust(res_df$pvalue, method = 'BH')
res_df$padj_fdr <- p.adjust(res_df$pvalue, method = 'fdr')
Goal: Understand and handle missing values in DE results caused by filtering or outlier detection.
Approach: Identify the source of NAs (zero counts, independent filtering, outliers) and remove or investigate them.
# Count NAs
sum(is.na(res$padj))
# Remove genes with NA padj
res_complete <- res[!is.na(res$padj), ]
# Understand why NAs occur
# - baseMean = 0: No counts
# - NA only in padj: Outlier or low count filtered by independent filtering
# Check outliers
res[which(is.na(res$pvalue) & res$baseMean > 0), ]
| Column | Description |
|---|---|
baseMean | Mean normalized counts |
log2FoldChange | Log2 fold change |
lfcSE | Standard error of LFC |
stat | Wald statistic |
pvalue | Raw p-value |
padj | Adjusted p-value (BH) |
| Column | Description |
|---|---|
logFC | Log2 fold change |
logCPM | Average log2 CPM |
F | Quasi-likelihood F-statistic |
PValue | Raw p-value |
FDR | False discovery rate |
| Experiment Type | Expected % DE (padj < 0.05, |LFC| > 1) |
|---|---|
| Subtle perturbation (low-dose drug, mild stress) | 0.5-3% |
| Standard treatment vs control | 3-10% |
| Different tissues or cell types | 15-40% |
| Cancer vs normal | 10-30% |
| Prokaryotic stress response | 10-50%+ |
If >50% of genes are DE in a standard comparison, suspect a technical issue (batch effect, normalization failure, sample swap). Prokaryotic stress experiments are the exception — bacteria can rewire large portions of their transcriptome.
| Cutoff | When to Use | Rationale |
|---|---|---|
| |LFC| > 0 (padj only) | Exploratory; generating ranked lists for GSEA | Captures all statistically significant changes |
| |LFC| > 0.5 (~1.4-fold) | Default for most experiments | Filters trivially small but statistically significant changes |
| |LFC| > 1 (~2-fold) | Standard stringent cutoff | Conventional in literature; good for large-effect studies |
| |LFC| > 2 (~4-fold) | Drug screens, very high-signal comparisons | May miss biologically important small changes (e.g., transcription factors) |
Prefer formal threshold testing (lfcThreshold in DESeq2, glmTreat in edgeR) over post-hoc filtering. Formal tests control the false positive rate at the threshold boundary; post-hoc filtering does not.
Check the raw p-value distribution before trusting DE results:
| Shape | Interpretation | Action |
|---|---|---|
| Uniform + spike near 0 | Correct: null genes uniform, true DE near 0 | Proceed normally |
| Anti-conservative (U-shape or spike at both ends) | Inflated significance; unmodeled batch or violated assumptions | Check for batch effects, verify model |
| Conservative (spike near 1, depleted near 0) | Over-correction; too many covariates or wrong dispersion | Simplify model, check dispersion plot |
| Spike at p = 1 only | Discrete artifact from low-count genes | Pre-filter more aggressively |
| Task | Use |
|---|---|
| Significance calls (which genes are DE) | Un-shrunken p-values (padj/FDR) |
| Ranking genes by effect size | Shrunken LFCs (apeglm/ashr) |
| GSEA input (ranked gene list) | Shrunken LFCs or Wald statistic |
| Volcano plot x-axis | Shrunken LFCs |
| Post-hoc LFC filtering | Apply to shrunken LFCs for more stable gene lists |
| Method | Input Required | How to Prepare |
|---|---|---|
| ORA (enrichGO, enrichKEGG) | Significant gene list + background | sig_genes <- subset(res, padj < 0.05); background = all tested genes |
| GSEA (fgsea, clusterProfiler::GSEA) | ALL genes ranked, no cutoff | Rank by stat (DESeq2 Wald) or sign(logFC) * -log10(PValue) (edgeR) |
Never use ORA on a ranked list or GSEA on a filtered list. For ORA, always supply the background (all genes that were tested), not just the genome — pre-filtering and independent filtering reduce the tested set.
# GSEA ranking from DESeq2
gsea_ranks <- res_df$stat
names(gsea_ranks) <- res_df$gene
gsea_ranks <- sort(gsea_ranks[!is.na(gsea_ranks)], decreasing = TRUE)
# GSEA ranking from edgeR
gsea_ranks <- sign(results$logFC) * -log10(results$PValue)
names(gsea_ranks) <- rownames(results)
gsea_ranks <- sort(gsea_ranks[is.finite(gsea_ranks)], decreasing = TRUE)
For bacterial/archaeal organisms, Ensembl and org.db packages are unavailable. Use:
# Load annotation from Prokka/Bakta GFF
library(rtracklayer)
gff <- import('annotation.gff3')
gene_info <- as.data.frame(gff[gff$type == 'gene', c('locus_tag', 'Name', 'product')])
# Merge with DE results
res_annotated <- merge(res_df, gene_info, by.x = 'gene', by.y = 'locus_tag', all.x = TRUE)
# KEGG enrichment with bacterial organism code
library(clusterProfiler)
# Find strain-specific KEGG code
search_kegg_organism('Pseudomonas aeruginosa', by = 'scientific_name')
# Use the code (e.g., 'pae' for PAO1)
kegg_res <- enrichKEGG(gene = sig_gene_ids, organism = 'pae', keyType = 'kegg')