conda and bioconductor version

The idea is that could use conda to setup an environment with different R / BioC packages to avoid situations when a package is update without backwards compatibility (as it happened with scater).

We need to test if in a conda env for R3.5 the BioC version installed is the latest, or that which was released with that R version.

Installing R3.5 in a env went well. However when starting the R console it started to try and install packages in a look, for instance devtools. The issue was my .Rprofile which contained instructions to install some packages if not present. Commenting this out solved the issue.

https://gist.github.com/adefelicibus/2ec97b5f8f12b4d71ed749d4f15ccd73

added 1h of time spent at 2020-01-14

assigned to @herseman

Installing R packages inside the conda env has some problems (possibly depending how many R packages and dependencies are installed upon env creation):

• Rcurl is missing but it can be installed with conda
• Zlib was not found during BioC package installation (see this similar issue).

added 30m of time spent at 2020-01-14

Follow on the zlib issue encountered while installing one the scater dependencies:

BiocManager::install("Rhdf5lib")
checking zlib in /usr... failed
configure: error: either specify a valid zlib installation with --with-zlib=DIR or disable zlib usage with --without-zlib
ERROR: configuration failed for package ‘Rhdf5lib’
* removing ‘/home/domingue/miniconda3/envs/R_3.5.1_essentials/lib/R/library/Rhdf5lib’

This might be because conda changes environment paths / variables and things start missing. devtools, for instance, can't find tar or gzip and this needs to be set: Sys.setenv(TAR = "/bin/tar") (or symlinked). Once that was fixed I managed to install Rhdf5lib from the git repo with devtools::install_github("grimbough/Rhdf5lib")

see also:

• https://github.com/grimbough/Rhdf5lib/issues/21
• https://github.com/r-lib/devtools/issues/379

added 45m of time spent at 2020-01-15

Another dependency failing due to conda was rhdf5 which was solved by installing with conda itself: conda install -c bioconda bioconductor-rhdf5

The issue here was that some library headers were not found by the c++ compiler even though they were in the specified location.

added 30m of time spent at 2020-01-15

As far as I can tell, BiocManager::install() matches the BioC version to the R version. In case of doubt we can also specify the version with BiocManager::install("HDF5Array", version = "3.8").

After all of this I managed to install:

packageVersion("scater")
> [1] ‘1.10.1’

conda can be used to create environments for old / stable packages.

added 15m of time spent at 2020-01-15

closed

mentioned in issue #78 (closed)

Some notes on how I created the env on my computer (Ubuntu 18.04) and dependencies needed:

conda create -n R_3.5.1_essentials -c r r=3.5.1
conda activate R_3.5.1_essentials
conda install -c r r-essentials=3.5.1 r-base=3.5.1
conda install -c anaconda hdf5
conda install -c bioconda bioconductor-rhdf5

@herseman if you want to use the conda env for R.3.5 (with scater 1.10), it's in /projects/conda/envs/sc_BioC_3.8. conda activate /projects/conda/envs/sc_BioC_3.8 should activate it.

Also, and out of curiosity, it takes 1.6G.

@domingue It may be useful to create the ~/.condarc file in order to account for the following to aspects:

(1) prevent conda from activating the base environment by default (https://stackoverflow.com/questions/54429210/how-do-i-prevent-conda-from-activating-the-base-environment-by-default)

conda config --set auto_activate_base false

(2) add channels for package installation

conda config --add channels defaults
conda config --add channels bioconda    #https://bioconda.github.io/user/install.html
conda config --add channels conda-forge #https://conda-forge.org/docs/user/introduction.html