You cannot install your own version of R and use it within RStudio.

All versions are handled with the software framework used and maintained by ITS (Easybuild). New version can be installed by us or ITS providing the install recipe has been released by Easybuild and is available in their GitHub repository.

We also advise against installing your own R version locally or on Seneca with e.g. conda because this will critically limit the reproducibility of your analysis.

An overview of the basic settings to run RStudio is available at https://git.embl.de/-/snippets/94.

Default install location for libraries is your home (~) which disk space is limited by a quota.

By installing to many libraries, you will eventually hit the quota and start experimenting disk space errors. You can circumvent this issue by configuring R to install library somewhere else, for example in your group share. To achieve this, please create a .Renviron file in your home folder (~/.Renviron) and set the R_LIBS_USER variable

R_LIBS_USER="/g/‹your_group›/‹your_username›/R-libs/%p/%V"
# Which resolves to /g/‹your_group›/‹your_username›/R-libs/x86_64-pc-linux-gnu/4.2.1 

Make sure to use the variables %p and %V (resolved respectively as the system architecture name and the R version) so that R adequately maintain version specific library install folders. This is important to not run into dependency conflicts when changing R versions.

Please update a pre-installed library only when needed (i.e. in case you are solving a dependency issue or when you know a newer version has a critical bug fix)

As explained above, R comes with the Bioconductor bundle and therefore has an extensive list of pre-installed libraries. Each is pinned to a specific version (the one listed in the Easybuild recipe). Updating all libraries – as sometimes advised by R – is not recommended here. It will download and install a newer version within your library folder, however the newer version will not be compiled in an optimised way as when we pre-installed it, and will therefore run less efficiently.

We advise to use the latest R-bundle-Bioconductor as it is a stable version of R that comes with a bunch of standard dependencies out-of-the-box (e.g. ggplot2), which you therefore do not need to install on your own (saving us all time and storage and associated CO2 emissions). When using this bundle, we also advise against blindly updating any of R libraries included in the Bioconductor bundle: updating a library in effect install your own version of this library in your library folder (leading to “duplicated” install for not much benefit) and updates many dependencies (potentially causing compatibility issues for other packages relying on older dependencies’ versions). You may still need to update a handful of them when you attempt to install a new R package that depends on a newer version of a dependency.

When you use other R install, please consider:

– R modules (without bioconductor) miss bioconductor dependencies, however they still are bundled with a bunch of standard UNIX dependencies, and therefore should be used in priority compared to R versions compiled from source. Missing system dependencies can usually still be installed upon request.

– R compiled from source is to be used when you want to access the newest versions of R (the module version of R usually comes months after a R release). However, consider that it will lack even some of the most standard system libraries which are not installed on Seneca. We may be able to install them from the CentOS package manager, but we cannot guarantee we will be able to install all of them as we may end up in dependency conflicts, which is exactly what we are trying to avoid using modules.

Important aspects when jumping between the different R installation types of the same R minor version (e.g. 4.3.2)

When installing your own libraries – and assuming you configured R_LIBS_USER in your .Renviron properly (see above) – your libraries install somewhere on your group share at a path like /<my_group_share>/<user>/R/x86_64-pc-linux-gnu-library/4.3.2/ (you may check this by running Sys.getenv("R_LIBS_USER") in the R). As visible on this path, the lib installs are separated by minor R version to avoid conflicts between R versions.

1. R packages installs are separated for different minor versions of R (i.e. a package install for 4.3.1 is separated from the install of the same package for 4.3.2)

2. Two R versions install of the same minor R version (e.g. R/4.3.2-gfbf-2023a and R-bundle-Bioconductor/3.18-foss-2023a-R-4.3.2) do share the same library folder.

3. Your own install of an R package prevails over the module/bundle install.

This means you may encounter dependency conflicts when switching to a version that is not compatible with you own install of an R package. This will happen for example when using the R/4.3.2-gfbf-2023a, installing a bioconductor package X and later switching to R-bundle-Bioconductor/3.18-foss-2023a-R-4.3.2, which relies on a different version of the same package X. You will have to solve such dependency conflict on your own, which typically will mean removing your library folder for this version of R and re-installing what you need.

Problem: RStudio stores sessions information into user’s home directory at /home/<username>. This can lead to issue when hitting disk quotas (50Gb)

Solution: Move the ~/.local/share/rstudio directory to another disk without quota (preferentially on seneca, or to your group share).

mkdir /tmpdata/$USER
mv ~/.local/share/rstudio /tmpdata/$USER/rstudio && ln -s /tmpdata/$USER/rstudio ~/.local/share/rstudio

Cannot convert rmarkdown to pdf/html
If you get errors related to 'X11 display' or 'Invalid argument', set this in your ~/.Rprofile:
options(bitmapType='cairo')
If, specifically, you want to generate PNG images inside HTML output, you can also use following Rmd preamble:
---
output:
  html_document:
    dev: CairoPNG
---

Problem: Conversion of RMarkdown (.rmd) files to HTML or PDF fails with errors related to X11 display or Invalid argument

Solution: Create or update the user ~/.Rprofile to add the following line

options(bitmapType='cairo')

If, specifically, you want to generate PNG images inside HTML output, you can also use following Rmd preamble:

---
output:
  html_document:
    dev: CairoPNG
---

Problem: An R package using a specific/outdated Python version is producing dependencies conflicts with our default Python version.

Solution: Use the reticulate package use_python(), use_virtualenv(), use_condaenv() functions. For example:

library(reticulate)
use_condaenv('my-project')

• Dell Power Edge R7425
• 64 cores capable of 128 concurrent threads (2x AMD EPYC 7601 2,20GHz/2,7GHz, 32C/64T, 64M Cache (180W) DDR4-2666)
• 2Tb RAM (32x 64GB LRDIMM, 2666MT/s)
• 3.2Tb local storage Flash Disk (/tmpdata)

Easybuild builds software modules. Linux comes with the module command-line tool to interact with modules (we use Lmod), and typically load them into your environment, list the existing and/or loaded ones, etc.

List available modules

• module avail lists all modules.
• module avail <string> lists all module with <string> in their name (case insensitive), e.g. module avail python returns Python and IPython modules, etc.
• module spider and module spider <string> do a similar job.

Load a module

• module load <module_name> [<module2_name> ...] loads the given module(s), e.g. module load Python/3.10.8-GCCcore-12.2.0 SciPy-bundle/2023.02-gfbf-2022b loads both Python and SciPy. Find names with the avail or spider commands.

When possible, load matching toolchain versions, i.e. versions that have been compile with the same toolchain.

NB 1: When loading multiple modules and hitting a dependency conflict, the last loaded module wins, i.e. the last module that needs the dependency dictates the loaded version of said dependency.

List loaded modules

• module list lists all the loaded module.

Even after explicitly loading a single module, the list may contain multiple module. This is because loading a module means loading the given one and all the module it depends on. For example loading R-bundle-Bioconductor/3.16-foss-2022b-R-4.2.2 effectively loads R, Bioconductor, as well as 123 other dependencies.

Unload module(s)

module unload <module_name> unloads a given module and all obsolete dependencies .
module purge unloads all the loaded module.

You cannot install your own software with Easybuild^*.

When you identify a piece of software that is not available, you can request its install to us or to IT Services. On our side, installing should not take long, providing that either (1) an official easybuild recipe exists, or (2) that the install procedure is standard & following best practices.

As an alternative, you may also use virtual environments managers (like conda) but we provide only limited support for them.

* Effectively you could maintain your own Easybuild install, but this is advanced usage and out of scope of this document