Install DESeq2 and edgeR via conda (R/Bioconductor, Apple Silicon)

DESeq2 and edgeR are the two workhorse R/Bioconductor packages for calling differentially expressed genes from RNA-seq count data, and nearly every published differential-expression pipeline leans on one of them. Both ship as prebuilt bioconda packages, so you can get a working Bioconductor stack without compiling anything from source or wrestling with install.packages() and missing system libraries. This guide builds a single conda environment with r-base, DESeq2, and edgeR pinned together, completely isolated from whatever R you already have on your Mac.

Why install through conda instead of install.packages()

The usual route to DESeq2 is BiocManager::install("DESeq2") from inside R. On macOS that often means Xcode command line tools, a matching gfortran, and compiling a long chain of C/C++ and Fortran dependencies from source — well over an hour, breaking in creative ways whenever your compiler toolchain drifts from the R build.

Conda sidesteps all of it. The bioconda packages bioconductor-deseq2 and bioconductor-edger are prebuilt binaries that pull in a matching r-base plus every Bioconductor and CRAN dependency, also prebuilt. Nothing compiles locally, and the whole stack lives inside one environment you can delete and rebuild at will.

Prerequisites

You need a working conda install first. If you do not have one yet, follow our Miniconda on Apple Silicon guide and come back here. This assumes an Apple Silicon Mac (osx-arm64) with a recent conda defaulting to the fast libmamba solver (conda 23.10 and later).

As with every bioconda install on this site, we install only from conda-forge and bioconda and pass --override-channels so conda never touches the Anaconda defaults channel, which is gated behind a Terms-of-Service prompt on recent conda releases.

TL;DR: copy-paste install

conda create -n bu-deseq2 --override-channels -c conda-forge -c bioconda bioconductor-deseq2 bioconductor-edger
conda activate bu-deseq2
R -e 'library(DESeq2); library(edgeR)'

That is the whole install. Expect it to take a while to solve — more on that below — and the rest of this guide walks through each step in detail.

Step 1: Create the environment

Ask for both packages in the same conda create call rather than installing them separately. Solving them together lets conda pick one consistent r-base version and one consistent set of shared dependencies for both packages at once.

conda create -n bu-deseq2 --override-channels -c conda-forge -c bioconda bioconductor-deseq2 bioconductor-edger

You should see something like this shape of output, trimmed down: conda resolves a large "NEW packages will be INSTALLED" list headed by r-base and dozens of r-* and bioconductor-* packages, downloads several hundred megabytes, and finishes with Executing transaction: done. Do not worry about exact package counts or version numbers — bioconda updates these builds regularly.

Solving environment: done

The following NEW packages will be INSTALLED:

  r-base                bioconda/osx-arm64::r-base-...
  bioconductor-deseq2   bioconda/osx-arm64::bioconductor-deseq2-...
  bioconductor-edger    bioconda/osx-arm64::bioconductor-edger-...
  ... (dozens more r-* and bioconductor-* dependencies)

Preparing transaction: done
Verifying transaction: done
Executing transaction: done

Step 2: Verify the install in R

Activate the environment — you need to do this in every new shell — and load both libraries directly:

conda activate bu-deseq2
R -e 'library(DESeq2); library(edgeR); sessionInfo()'

You should see something like this: each library() call prints a few "Loading required package" messages as dependencies attach, no errors, and sessionInfo() lists both packages under "other attached packages" with version strings such as DESeq2_x.y and edgeR_x.y (exact numbers depend on when you installed).

Loading required package: S4Vectors
Loading required package: BiocGenerics
...

other attached packages:
[1] edgeR_x.y          limma_x.y          DESeq2_x.y
[4] SummarizedExperiment_x.y   GenomicRanges_x.y   ...

Also confirm this R is the one inside your environment, not a system Homebrew or CRAN install:

which R

It should resolve inside .../envs/bu-deseq2/bin/R. If you get a different path, or R: command not found, activate the environment again.

Step 3: A tiny real usage example

A minimal, runnable example that exercises both packages end to end on a toy count matrix — two genes, three control and three treatment replicates. Far too small to draw any biological conclusion from, but it confirms the install works and shows each package's API.

counts <- matrix(
  c(20, 25, 18, 200, 210, 195,
    30, 28, 32,  15,  12,  18),
  nrow = 2, byrow = TRUE,
  dimnames = list(
    c("geneA", "geneB"),
    c("ctrl1", "ctrl2", "ctrl3", "trt1", "trt2", "trt3")
  )
)
condition <- factor(c("ctrl", "ctrl", "ctrl", "trt", "trt", "trt"))

DESeq2 wraps the counts and metadata in a DESeqDataSet, then runs its full normalization, dispersion-estimation, and Wald-test pipeline with a single DESeq() call:

library(DESeq2)
coldata <- data.frame(condition = condition, row.names = colnames(counts))
dds <- DESeqDataSetFromMatrix(countData = counts, colData = coldata, design = ~ condition)
dds <- DESeq(dds)
results(dds)

edgeR takes a slightly more manual route: build a DGEList, normalize with calcNormFactors(), estimate dispersion, then test:

library(edgeR)
y <- DGEList(counts = counts, group = condition)
y <- calcNormFactors(y)
y <- estimateDisp(y)
et <- exactTest(y)
topTags(et)

Both should run without error and print a results table with a log2 fold change and a p-value for geneA and geneB. For a real experiment, feed in a full count matrix from featureCounts or Salmon plus tximport, with thousands of genes so dispersion estimates have enough data to shrink sensibly. See our RNA-seq differential expression walkthrough for the full statistical picture.

DESeq2 vs edgeR: which one to reach for

Both model counts with a negative binomial distribution and both are well validated, but they differ in defaults and workflow feel:

In practice both give similar gene lists on the same dataset. Many labs run both as a cross-check, especially in a course setting.

Apple Silicon and long-solve gotchas

A few things specific to this stack, beyond the generic bioconda advice:

• The solve really is slow — that is normal. R/Bioconductor environments have one of the largest dependency graphs in bioconda. A Solving environment step hanging for minutes is real work, not a stall — avoid interrupting it.
• Install both packages in one conda create, not two. Installing them separately can force the solver to downgrade r-base or shared dependencies later, triggering a bigger, slower re-solve.
• Native osx-arm64 builds exist for this stack. Both packages carry compiled C/C++ code (DESeq2 via Rcpp, edgeR via its own C routines), so bioconda ships genuine osx-arm64 builds rather than relying on Rosetta 2. Confirm with conda list -n bu-deseq2 r-base.
• If a stray dependency ever lacks an arm64 build, fall back to CONDA_SUBDIR=osx-64 conda create -n bu-deseq2-x86 --override-channels -c conda-forge -c bioconda bioconductor-deseq2 bioconductor-edger, then conda config --env --set subdir osx-64. Rarely needed for this stack today.
• Do not mix this with system R or RStudio's CRAN R. Keep any BiocManager-installed system R separate, and point RStudio at .../envs/bu-deseq2/bin/R if you want to use this environment there.

Common errors and fixes

Managing the environment

Update both packages in place:

conda update -n bu-deseq2 --override-channels -c conda-forge -c bioconda bioconductor-deseq2 bioconductor-edger

Snapshot the exact environment so a collaborator (or a future you) can reproduce it precisely:

conda env export -n bu-deseq2 > deseq2-edger-env.yml

And when you no longer need it, remove it cleanly. Because everything — r-base included — lived inside the environment, this leaves nothing behind on your system:

conda remove -n bu-deseq2 --all

Next steps

With a working DESeq2/edgeR environment in hand, a couple of natural follow-ons:

• Work through the RNA-seq differential expression walkthrough to see how counts, normalization, and the DESeq2 model fit into a full raw-reads-to-results pipeline.
• If you still need conda itself, start with Miniconda on Apple Silicon before coming back to this guide.