Skill: Bioinformatics Environment Setup

Use When

• User is starting a new analysis and wants to install the required bioinformatics tools.
• A pipeline step fails because a tool is missing or the wrong version is installed.
• User asks how to install a specific bioinformatics tool (e.g., STAR, Cell Ranger, Scanpy, Seurat).
• User wants a ready-to-use conda environment for a specific pipeline.
• User is setting up a new compute environment (HPC, cloud VM, local workstation).

Inputs

• Required:
  • Target pipeline(s): one or more of ngs-shared, wgs, rnaseq, scrnaseq, metagenomics, or all
• Optional:
  • Installation method: conda (default), docker, or manual
  • Operating system: linux (default), macos
  • Environment name prefix (default: bioinfo)
  • Conda channel priority: strict (default, reproducible) or flexible
  • Whether to verify existing installations before installing (default: true)

Workflow

1. Detect the environment: Check whether conda/mamba is available (which conda), whether Docker is available (docker --version), and whether an HPC module system is present (module avail). Report findings to the user and recommend the best installation path.
2. Check existing installations: For each tool in the requested pipeline group(s), run which <tool> && <tool> --version 2>&1 | head -1 to detect installed versions. Report a summary table: tool name, found/missing, installed version, required minimum version.
3. Select the installation asset: Choose the appropriate conda environment YAML from assets/envs/:
   • ngs-shared.yaml — FastQC, MultiQC, fastp, Trimmomatic, samtools
   • wgs.yaml — BWA-MEM2, GATK, Picard, bcftools, VEP, SnpEff
   • rnaseq.yaml — STAR, HISAT2, featureCounts (subread), salmon, kallisto, DESeq2 (R), edgeR (R)
   • scrnaseq-python.yaml — STARsolo (STAR), Scanpy, scVI-tools, scVelo, Scrublet, CellBender, harmonypy
   • scrnaseq-r.yaml — Seurat v5, SingleR, scDblFinder, SoupX, Monocle3, DoubletFinder
   • metagenomics.yaml — Bowtie2, MetaPhlAn4, HUMAnN3, Kraken2, Bracken, metaSPAdes, MEGAHIT, Prokka
4. Create the conda environment:

   mamba env create -f assets/envs/<pipeline>.yaml
   # or, if mamba is unavailable:
   conda env create -f assets/envs/<pipeline>.yaml

   Prefer mamba over conda for significantly faster dependency resolution.
5. Handle tools that cannot be installed via conda — prompt the user with manual steps:
   • Cell Ranger: Download from the 10x Genomics website (requires registration). Add to PATH: export PATH=/path/to/cellranger:$PATH.
   • CellBender: Install via pip inside the conda environment: pip install cellbender. Requires CUDA for GPU acceleration.
   • DoubletFinder: Install from GitHub inside R: remotes::install_github("chris-mcginnis-ucsf/DoubletFinder").
   • Monocle3: Install from Bioconductor: BiocManager::install("monocle3"). Alternatively: remotes::install_github("cole-trapnell-lab/monocle3").
   • velocyto (for RNA velocity loom generation): pip install velocyto. Requires samtools and a reference GTF.
6. Verify all installations: After installation, re-run the version check for each tool. For R packages, run Rscript -e 'packageVersion("<pkg>")'. Report a final verification table: tool, expected version, installed version, status (PASS/FAIL).
7. Report any failures with the specific error message and the most likely fix (e.g., missing system library, incompatible CUDA version, wrong conda channel).

Output Contract

• Environment detection summary (printed to console)
• Pre-installation version check table (TSV or formatted table): tool, status, version
• Installation commands executed (echo each command before running)
• Post-installation verification table (TSV or formatted table): tool, expected_min_version, installed_version, pass/fail
• List of any tools that require manual installation steps, with exact instructions

Limits

• Cell Ranger requires registration and manual download from 10x Genomics; it cannot be distributed via conda or pip.
• CellBender GPU acceleration requires CUDA ≥ 11.0 and a compatible NVIDIA GPU; CPU-only mode is available but slow (hours per sample).
• Monocle3 has complex Bioconductor dependencies that frequently cause installation conflicts; allocate extra time and follow the official installation guide.
• Conda environments can be large (2–5 GB each); ensure sufficient disk space before creating them.
• On HPC clusters, use module load for pre-installed tools before attempting conda installation; conda may conflict with system modules.
• macOS (Apple Silicon) requires CONDA_SUBDIR=osx-64 for some bioinformatics tools that lack native ARM builds.
• Common failure cases:
  • conda solve timing out on complex environments; switch to mamba for faster resolution.
  • GATK or Picard failing due to wrong Java version; ensure Java 17+ is available (java --version).
  • R package installation failing due to missing system libraries (e.g., libgdal, libcurl); install the system library first with apt-get or yum.
  • STARsolo requiring a STAR version ≥ 2.7.9a for Velocyto output; check STAR version with STAR --version.