diff --git a/.tests/config/simple_config.yaml b/.tests/config/simple_config.yaml index cdac8ed2..0d737022 100644 --- a/.tests/config/simple_config.yaml +++ b/.tests/config/simple_config.yaml @@ -3,10 +3,10 @@ # -------------------------------------------------------- # MosaiCatcher version -version: 2.2.0 +version: 2.2.1 # Ashleys-QC pipeline version -ashleys_pipeline_version: 2.2.0 +ashleys_pipeline_version: 2.2.1 # Email for notifications about the pipeline's status email: "" diff --git a/config/config.yaml b/config/config.yaml index 9571162c..78dc3d51 100644 --- a/config/config.yaml +++ b/config/config.yaml @@ -3,10 +3,10 @@ # -------------------------------------------------------- # MosaiCatcher version -version: 2.2.0 +version: 2.2.1 # Ashleys-QC pipeline version -ashleys_pipeline_version: 2.2.0 +ashleys_pipeline_version: 2.2.1 # Email for notifications about the pipeline's status email: "" diff --git a/github-actions-runner/Dockerfile-2.2.1.dockerfile b/github-actions-runner/Dockerfile-2.2.1.dockerfile new file mode 100644 index 00000000..d5f5feb1 --- /dev/null +++ b/github-actions-runner/Dockerfile-2.2.1.dockerfile @@ -0,0 +1,227 @@ +FROM condaforge/mambaforge:latest +LABEL io.github.snakemake.containerized="true" +LABEL io.github.snakemake.conda_env_hash="77eaa388d65d5205b87324fb0adb89561bc0e532a328995990a1d580aeb894ae" + +# Step 1: Retrieve conda environments + +# Conda environment: +# source: https://github.com/snakemake/snakemake-wrappers/raw/v1.7.0/bio/bwa/index/environment.yaml +# prefix: /conda-envs/5681728a49bd83ceed09ba194330c858 +# channels: +# - bioconda +# - conda-forge +# - defaults +# dependencies: +# - bwa ==0.7.17 +RUN mkdir -p /conda-envs/5681728a49bd83ceed09ba194330c858 +ADD https://github.com/snakemake/snakemake-wrappers/raw/v1.7.0/bio/bwa/index/environment.yaml /conda-envs/5681728a49bd83ceed09ba194330c858/environment.yaml + +# Conda environment: +# source: https://github.com/snakemake/snakemake-wrappers/raw/v1.7.0/bio/fastqc/environment.yaml +# prefix: /conda-envs/08d4368302a4bdf7eda6b536495efe7d +# channels: +# - bioconda +# - conda-forge +# - defaults +# dependencies: +# - fastqc ==0.11.9 +RUN mkdir -p /conda-envs/08d4368302a4bdf7eda6b536495efe7d +ADD https://github.com/snakemake/snakemake-wrappers/raw/v1.7.0/bio/fastqc/environment.yaml /conda-envs/08d4368302a4bdf7eda6b536495efe7d/environment.yaml + +# Conda environment: +# source: https://raw.githubusercontent.com/friendsofstrandseq/ashleys-qc-pipeline/dev/workflow/envs/ashleys_base.yaml +# prefix: /conda-envs/87c04f5d115eff742eca84455513deba +# name: ashleys_base +# channels: +# - conda-forge +# - bioconda +# dependencies: +# - samtools +# - tabix +# - bwa +# - sambamba +# - mosaicatcher +# # - alfred +# - ashleys-qc +# - pandas +# # PUBLISHDIR +# - rsync +# # MULTIQC +# - multiqc +# # Fix sklearn update +# - scikit-learn=1.2.2 +RUN mkdir -p /conda-envs/87c04f5d115eff742eca84455513deba +ADD https://raw.githubusercontent.com/friendsofstrandseq/ashleys-qc-pipeline/dev/workflow/envs/ashleys_base.yaml /conda-envs/87c04f5d115eff742eca84455513deba/environment.yaml + +# Conda environment: +# source: https://raw.githubusercontent.com/friendsofstrandseq/ashleys-qc-pipeline/dev/workflow/envs/ashleys_rtools.yaml +# prefix: /conda-envs/9b847fc31baae8e01dfb7ce438a56b71 +# name: rtools +# channels: +# - conda-forge +# - bioconda +# - r +# - anaconda +# dependencies: +# # - bioconductor-biocparallel +# # - bioconductor-bsgenome +# # - bioconductor-bsgenome.hsapiens.ucsc.hg19 +# # - bioconductor-bsgenome.hsapiens.ucsc.hg38 +# # - bioconductor-fastseg +# # - bioconductor-genomicalignments +# - bioconductor-genomicranges +# # - bioconductor-rsamtools +# # - bioconductor-s4vectors +# - r-assertthat +# - r-base +# # - r-biocmanager +# - r-cowplot +# - r-data.table +# # - r-devtools +# # - r-doparallel +# # - r-foreach +# - r-ggplot2 +# # - r-gtools +# - r-reshape2 +# # - r-zoo +# # - r-dplyr +# # - r-mc2d +# # - r-pheatmap +# # - bioconductor-complexheatmap +# # - r-gplots +# - r-scales +# - r-rcolorbrewer +# # - r-stringr +# - r-cairo +# - fonts-anaconda +# # NEW +# - bioconductor-edger +# - r-r.utils +# # PLATE PLOT +# - r-dplyr +# - r-platetools +# - r-viridis +# # GC_correction +# - r-tidyr +# - r-ggpubr +# # SOLVE R lib issue +# - r-stringi=1.7.12 +RUN mkdir -p /conda-envs/9b847fc31baae8e01dfb7ce438a56b71 +ADD https://raw.githubusercontent.com/friendsofstrandseq/ashleys-qc-pipeline/dev/workflow/envs/ashleys_rtools.yaml /conda-envs/9b847fc31baae8e01dfb7ce438a56b71/environment.yaml + +# Conda environment: +# source: workflow/envs/mc_base.yaml +# prefix: /conda-envs/c80307395eddf442c2fb6870f40d822b +# name: mc-base +# channels: +# - conda-forge +# - bioconda +# dependencies: +# - pandas +# - intervaltree +# - scipy +# - pysam +# - tqdm +# - perl +# - pypdf2 +# - parmap +# # NEW +# - pyyaml +# - seaborn +# - matplotlib +# # SOLVE se-pe detection +# - samtools +# # ArbiGent Hufsah deps +# - pytables +# - xopen +RUN mkdir -p /conda-envs/c80307395eddf442c2fb6870f40d822b +COPY workflow/envs/mc_base.yaml /conda-envs/c80307395eddf442c2fb6870f40d822b/environment.yaml + +# Conda environment: +# source: workflow/envs/mc_bioinfo_tools.yaml +# prefix: /conda-envs/f251d84cdc9f25d0e14b48e780261d66 +# name: mc-bioinfo-tools +# channels: +# - conda-forge +# - bioconda +# dependencies: +# - bcftools +# - freebayes +# - mosaicatcher +# - samtools +# - tabix +# - whatshap +RUN mkdir -p /conda-envs/f251d84cdc9f25d0e14b48e780261d66 +COPY workflow/envs/mc_bioinfo_tools.yaml /conda-envs/f251d84cdc9f25d0e14b48e780261d66/environment.yaml + +# Conda environment: +# source: workflow/envs/rtools.yaml +# prefix: /conda-envs/598c87b6c764d05e0c66953cc67f2931 +# name: rtools +# channels: +# - bioconda +# - conda-forge +# - r +# - anaconda +# dependencies: +# # # NEW +# - strandphaser +# # ############### +# - bioconductor-biocparallel +# - bioconductor-bsgenome +# - bioconductor-bsgenome.hsapiens.ucsc.hg38 +# - bioconductor-complexheatmap +# # - bioconductor-fastseg +# - bioconductor-genomicalignments +# - bioconductor-genomicranges +# - bioconductor-rsamtools +# # - bioconductor-s4vectors +# - fonts-anaconda +# - r-assertthat +# - r-base +# - r-biocmanager +# - r-cairo +# - r-cowplot +# - r-data.table +# - r-devtools +# - r-doparallel +# - r-dplyr +# - r-foreach +# - r-ggplot2 +# - r-gplots +# - r-gtools +# - r-mc2d +# - r-rcolorbrewer +# - r-reshape2 +# - r-scales +# - r-stringr +# # SV_CALLS_DEV +# # - r-zoo +# - r-r.utils +# - r-ggnewscale +# # HEATMAP +# - r-tidyr +# # ARBIGENT +# - r-reshape +# - r-optparse +# - r-tidyr +# - r-ggbeeswarm +# - r-pheatmap +# # GC_correction +# - r-ggpubr +# - bioconductor-edger +# # SOLVE R lib issue +# - r-stringi=1.7.12 +RUN mkdir -p /conda-envs/598c87b6c764d05e0c66953cc67f2931 +COPY workflow/envs/rtools.yaml /conda-envs/598c87b6c764d05e0c66953cc67f2931/environment.yaml + +# Step 2: Generate conda environments + +RUN mamba env create --prefix /conda-envs/5681728a49bd83ceed09ba194330c858 --file /conda-envs/5681728a49bd83ceed09ba194330c858/environment.yaml && \ + mamba env create --prefix /conda-envs/08d4368302a4bdf7eda6b536495efe7d --file /conda-envs/08d4368302a4bdf7eda6b536495efe7d/environment.yaml && \ + mamba env create --prefix /conda-envs/87c04f5d115eff742eca84455513deba --file /conda-envs/87c04f5d115eff742eca84455513deba/environment.yaml && \ + mamba env create --prefix /conda-envs/9b847fc31baae8e01dfb7ce438a56b71 --file /conda-envs/9b847fc31baae8e01dfb7ce438a56b71/environment.yaml && \ + mamba env create --prefix /conda-envs/c80307395eddf442c2fb6870f40d822b --file /conda-envs/c80307395eddf442c2fb6870f40d822b/environment.yaml && \ + mamba env create --prefix /conda-envs/f251d84cdc9f25d0e14b48e780261d66 --file /conda-envs/f251d84cdc9f25d0e14b48e780261d66/environment.yaml && \ + mamba env create --prefix /conda-envs/598c87b6c764d05e0c66953cc67f2931 --file /conda-envs/598c87b6c764d05e0c66953cc67f2931/environment.yaml && \ + mamba clean --all -y \ No newline at end of file diff --git a/github-actions-runner/docker_procedure.md b/github-actions-runner/docker_procedure.md index c2661b4b..f3ea5c61 100644 --- a/github-actions-runner/docker_procedure.md +++ b/github-actions-runner/docker_procedure.md @@ -34,7 +34,16 @@ conda activate snakemake snakemake --configfile .tests/config/simple_config.yaml --config ashleys_pipeline=True MultiQC=True --profile workflow/snakemake_profiles/local/conda/ -c 8 --forceall -n snakemake --configfile .tests/config/simple_config.yaml --config ashleys_pipeline=True MultiQC=True --profile workflow/snakemake_profiles/local/conda_singularity/ -c 8 --forceall +# Solved in -# Solved in +https://github.com/weber8thomas/workspace-images/tree/docker-apptainer-mamba -https://github.com/weber8thomas/workspace-images/tree/docker-apptainer-mamba \ No newline at end of file +# Dockerfile creation + +snakemake --configfile .tests/config/simple_config.yaml --config ashleys_pipeline=True MultiQC=True --profile workflow/snakemake_profiles/local/conda/ -c 8 --containerize > Dockerfile + +# Docker commands + +docker login -u weber8thomas +docker build --platform=linux/amd64 -t weber8thomas/mosaicatcher-pipeline:VERSION . +docker push -t weber8thomas/mosaicatcher-pipeline:VERSION diff --git a/workflow/Snakefile b/workflow/Snakefile index e8e15c2b..980f9ca0 100644 --- a/workflow/Snakefile +++ b/workflow/Snakefile @@ -25,9 +25,9 @@ if config["ashleys_pipeline"] is True: github( "friendsofstrandseq/ashleys-qc-pipeline", path="workflow/Snakefile", - # tag=str(config["ashleys_pipeline_version"]) + tag=str(config["ashleys_pipeline_version"]) # branch="main", - branch="dev", + # branch="dev", ) config: config diff --git a/workflow/rules/common.smk b/workflow/rules/common.smk index fbbb1e4d..999140ba 100644 --- a/workflow/rules/common.smk +++ b/workflow/rules/common.smk @@ -285,7 +285,7 @@ class HandleInput: # List of folders/files to not consider (restrict to samples only) l_to_process = [e for e in os.listdir(thisdir) if e not in exclude and e.endswith(".zip") is False] - print(l_to_process) + # print(l_to_process) if config["samples_to_process"]: l_to_process = [e for e in l_to_process if e in config["samples_to_process"]] @@ -327,7 +327,7 @@ class HandleInput: complete_df = complete_df.sort_values(by=["Cell", "File"]).reset_index( drop=True ) - print(complete_df) + # print(complete_df) return complete_df