Perturb-seq Analysis (bioSkills)

A Claude Code skill for single-cell CRISPR screens (Perturb-seq / CROP-seq) that covers guide assignment, escaper removal, and statistically calibrated differential testing.

   
Type Claude Skill
Supplier GPTomics bioSkills (community OSS, MIT)
Availability GA — part of the bioSkills collection
Pricing Free / OSS (MIT) — Pertpy, scanpy, SCEPTRE, Seurat, scMAGeCK, DESeq2 and edgeR are separately installed OSS
Capabilities Read/Write — Claude runs the skill’s workflow locally (Python/R), not as an MCP tool
Verified works · 2026-08-06
Security cleared · 2026-08-06 — GPTomics/bioSkills MIT, standard OSS deps, no advisories

How to install

bioSkills is not an npm package — skills are plain markdown/code read directly by the agent. Clone the repo, then either run the installer for the whole category or copy the single skill directory.

  • Claude Code — clone and install via the bundled script: 
    git clone https://github.com/GPTomics/bioSkills
cd bioSkills
./install-claude.sh --categories "single-cell"

    The installer copies matching skills into ~/.claude/skills/ (default target). Use ./install-claude.sh --list to preview the skills first.

  • Claude Code / other agents — copy just this one skill: 
    cp -r bioSkills/single-cell/perturb-seq ~/.claude/skills/

    (run from inside your clone — the previous step left you in bioSkills/; otherwise replace bioSkills/ with the absolute path of your clone). Install the Python stack on first use: pip install "pertpy>=0.9" "scanpy>=1.10" "anndata>=0.10"; SCEPTRE 0.10+ and the Seurat/scMAGeCK paths are R installs.

What it does

Walks a CRISPR screen from raw guide counts to defensible effect sizes:

  • Guide assignment — treats guide calling as a mixture problem and uses the mixture-model posterior by default; flat thresholds are reserved for sanity checks.
  • Escaper removal — computes a Mixscape perturbation signature and a two-component knockout / non-perturbed classification, so cells that received a guide but show no transcriptional effect do not dilute the comparison.
  • Calibrated testing — SCEPTRE conditional resampling, or pseudobulk DE with DESeq2/edgeR, instead of naive per-cell differential expression. The replication unit is the transfection, not the cell, so pseudobulk-per-replicate with ≥2–3 replicates is required.
  • Effect quantification — E-distance computed in a fixed PCA embedding, giving a comparable magnitude across perturbations.
  • Compositional analysis — Milo / scCODA differential abundance, separating a shift in cell-state proportions from a change in expression within a state.
  • Foundation-model validation — whole-perturbation holdout plus DE-gene metrics measured against an additive baseline, to test whether a perturbation-prediction model actually beats the trivial predictor.
  • Design guidance — low-MOI designs (~1 guide/cell) discard 70–90% of cells; high-MOI enables combinatorial perturbation designs.

Primary use cases: CRISPR screen differential expression, perturbation effect ranking, benchmarking perturbation-prediction models.

Notes

Distributed as a SKILL.md (plus reference material) in the bioSkills collection — Claude executes the workflow locally rather than as an MCP server. The upstream skill front-matter name is bio-single-cell-perturb-seq (tool_type: python, primary_tool: Pertpy); if invoked as a namespaced plugin command it resolves under the bioSkills plugin, not as a bare /perturb-seq. The skill’s most consequential rule is the replication unit: treating cells as independent replicates inflates significance in essentially every Perturb-seq dataset, which is why SCEPTRE or pseudobulk is the default rather than a scanpy rank_genes_groups call. Guide-library design and sgRNA selection sit upstream in sgRNA Design Guide; Scanpy and AnnData provide the underlying data structures. Upstream directory: single-cell/perturb-seq.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.