scanpy

scanpy

Standard single-cell RNA-seq analysis pipeline — QC, normalization, dimensionality reduction, clustering, differential expression, and visualization.

What it does

Scanpy is the standard Python toolkit for scRNA-seq analysis built on AnnData. This skill provides two CLI scripts plus comprehensive reference docs covering the full workflow: load data → QC → normalize → select features → PCA → neighborhood graph → UMAP/t-SNE → cluster (Leiden) → find marker genes → annotate cell types → save.

Setup

cd scanpy
python3 -m venv .venv && source .venv/bin/activate && pip install scanpy -q

Environment variables

None. Pure local library.

Usage

Input — QC analysis (main script)

python3 scripts/qc_analysis.py input_data.h5ad --mt-threshold 5 --min-genes 200

Output

Produces qc_filtered.h5ad with filtered cells and genes, plus QC diagnostic plots (unless --skip-plots).

Help output

usage: qc_analysis.py [-h] [--output OUTPUT] [--mt-threshold MT_THRESHOLD]
                      [--min-genes MIN_GENES] [--min-cells MIN_CELLS]
                      [--skip-plots]
                      input

positional arguments:
  input                 Input file (h5ad, 10X mtx, csv, etc.)

optional arguments:
  --output OUTPUT       Output file name (default: qc_filtered.h5ad)
  --mt-threshold        Max mitochondrial percentage (default: 5)
  --min-genes           Min genes per cell (default: 200)
  --min-cells           Min cells per gene (default: 3)
  --skip-plots          Skip generating QC plots

Scripts

Script	Purpose
qc_analysis.py	Quality control: calculates metrics, generates diagnostic plots, filters low-quality cells/genes
demo.py	Placeholder for basic workflow demonstration (returns scanpy version info)

CLI flags (qc_analysis.py)

Flag	Description	Default
input	Input file: .h5ad, 10X .mtx, .csv, etc.	required
--output	Output filtered file path	qc_filtered.h5ad
--mt-threshold	Max mitochondrial gene %	5
--min-genes	Min genes per cell	200
--min-cells	Min cells per gene	3
--skip-plots	Skip generating QC plots	false

Dependencies

• scanpy (includes anndata, numpy, scipy, matplotlib, pandas, seaborn as transitive deps)

Tested with

• demo.py --format json: ✅ Returns clean JSON (scanpy imported successfully)
• qc_analysis.py --help: ✅ All 5 flags documented with defaults
• Agno agent (Claude Haiku 4.5 via OpenRouter): ✅ Agent loaded instructions, ran qc_analysis.py --help, described the 9-step standard workflow (load → QC → normalize → features → PCA → UMAP → cluster → markers → annotate), both scripts' roles, and QC parameters

Agno agent verdict (excerpt)

Standard scRNA-seq workflow: load data (10X/h5ad/CSV) → QC (filter low-quality cells by gene count, total counts, mitochondrial %) → normalize (10K counts/cell, log-transform) → feature selection (~2K HVGs) → PCA → neighborhood graph → UMAP/t-SNE → Leiden clustering → marker gene identification → cell type annotation → save. The qc_analysis.py script automates the QC step with configurable thresholds (mt-threshold 5%, min-genes 200, min-cells 3) and diagnostic plots.

See also

• references/standard_workflow.md — complete Scanpy workflow code walkthrough
• references/api_reference.md — key function signatures and parameters
• references/plotting_guide.md — visualization recipes

Fix notes

• Fixed skill_name undefined bug in demo.py (replaced with "scanpy")
• Removed stray __pycache__/
• qc_analysis.py unchanged (200 lines, already functional)