MDU-PHL / bohra

A pipeline for bioinformatics analysis of bacterial genomes
GNU General Public License v3.0
18 stars 4 forks source link

Containerized Bohra installation #72

Open tetedange13 opened 1 year ago

tetedange13 commented 1 year ago

Hi,

Context

I did not want to mess up my good old Conda install and Biocontainers' Docker image for Bohra only has bohra itself (and conda is absent so impossible to install all tools through bohra init) => So I tried to set up everything for Bohra inside a Singularity image => It took me ages but it eventually worked, so I thought it would be useful to share it here

Disclaimer: I am not an expert in Singularity (neither in "Conda inside Singularity"), so surely some parts can be improved

Singularity definition file

Bootstrap: docker
From: condaforge/mambaforge:22.11.1-4
# Don't know why but above never worked for me...
# (instead had to create a local "mambaforge" Sing img and "bootstrap" from it)

%post
    ## 1) Enable 'activation' of Conda envs (also add '$CONDA_PREFIX/bin' to $PATH)
    . "/opt/conda/etc/profile.d/conda.sh"

    ## 2) Configure Conda channels that will be used later by "bohra init"
    # (order matters: "conda-forge" should be added last to be prior used channel)
    # (good to also add 'defaults' channel as "mambaforge" image does not have it configured natively)
    conda config --add channels defaults && \
        conda config --add channels bioconda && \
        conda config --add channels conda-forge

    ## 3) Install "bohra" itself (alone)
    # (had to pin Python version, otherwise env cannot be solved)
    mamba install --quiet --name base --yes python=3.8 bohra=2.3.2
    conda activate base

    ## 4) Fix packages issues 
    # * "altair" package missing from bohra's Bioconda recipe ?
    pip install altair==4.2.2
    # * Manually change installed version of Moba-suite, to avoid write of ".lock" file by "mob_recon" process
    #   (incompatible with Sing img which are read-only ; see: https://github.com/phac-nml/mob-suite/issues/82)
    sed --in-place 's/mob_suite=3.0.2/mob_suite=3.0.3/' "$CONDA_PREFIX"/lib/python3.8/site-packages/bohra/bohra_install.sh

    ## 5) Let bohra install everything
    bohra init

    ## 6) Clean-up to make final image smaller (actually saves 3 GB)
    mamba clean --all --yes

    ## 7) Fix activation of local Conda env by Nextflow
    # Cmd used: "source $(conda info --json | awk '/conda_prefix/ { gsub(/"|,/, "", $2); print $2 }')/bin/activate /opt/conda/envs/bohra-${TOOL}"
    # Which results (don't know why) in "source /bin/activate /opt/conda/envs/bohra-{TOOL}"
    # -> Dirty solution is to create a symlink
    ln -s /opt/conda/bin/activate/opt/conda/bin/activate /bin/activate

%environment
    # Required to have "base" env with "bohra" inside activated at runtime
    . "/opt/conda/etc/profile.d/conda.sh"
    conda activate base

%runscript
    exec "$@"


Misc

Hope this helps ! Have a nice day, Felix.

kristyhoran commented 10 months ago

@tetedange13 thank you so much for taking the time!! Impressive effort getting all the dependencies to play nicely together!! I will attempt to add this as a recipe and some instructions to the repo for others to use!!