Scanpy

Scanpy: Single-Cell Analysis

Overview

Scanpy is a scalable Python toolkit for analyzing single-cell RNA-seq data, built on AnnData. Apply this skill for complete single-cell workflows including quality control, normalization, dimensionality reduction, clustering, marker gene identification, visualization, and trajectory analysis. Targets Scanpy 1.12.4 (released 2026-08-27), reviewed 2026-10-01. Native synthetic tests establish data/API contracts, not biological validity.

Installation

Requires Python 3.12+ (scanpy 1.12 dropped Python ≤3.11) and anndata ≥0.10.

uv pip install "scanpy[leiden]"

The [leiden] extra installs igraph and leidenalg; the scripts explicitly select flavor="igraph". For reproducible environments, pin a version: uv pip install "scanpy[leiden]==1.12.4".

For large or out-of-core datasets, many functions support Dask arrays (experimental):

uv pip install "scanpy[leiden]" dask

See the Using dask with Scanpy tutorial. For GPU-accelerated scanpy-like operations, use rapids-singlecell as a separate package.

If the input is an R-native single-cell object (.rds, .RData, Seurat, or SingleCellExperiment), first convert it to .h5ad with R tooling, then load it with Scanpy. Read references/r_interop.md for agent-run installation and conversion instructions across macOS, Linux, and Windows.

For AnnData structure and I/O details, use the anndata skill. For probabilistic models and batch correction, use scvi-tools.

Representation and inference contracts

• Input to QC/preprocessing must be identified raw counts. The scripts reject negative, fractional, nonfinite or zero-total cells. Use --counts-layer counts when X is already normalized; never infer provenance from integer-looking values alone.
• layers["counts"] remains unnormalized; .raw is an independent full-gene log-normalized snapshot. Gene subsetting also subsets every layer, but does not subset .raw genes. Retain a full-gene object for pseudobulk; the full pipeline does so.
• seurat/cell_ranger HVGs use log-normalized values; seurat_v3 and seurat_v3_paper use counts and require scanpy[skmisc]. Scaling/regression are optional, may densify sparse data, and can remove biology along with covariates.
• Marker ranking defaults to .raw when present; --no-use-raw selects X and --layer selects an explicit log-normalized layer. --groups/--reference control contrasts. Wilcoxon/t-test adjusted p-values use BH within each comparison; logreg returns ranking scores without p-values. Clusters chosen from the same data yield exploratory markers.
• Condition DE requires raw-count sums per sample and cell type, independent biological replicates, aligned condition/donor metadata and a full-rank design. Analyze each cell type separately; repeated donor samples require a suitable paired design. Aggregation cannot fix confounding, missing replication, doublets or incorrect annotations.
• Harmony (harmonypy) changes PCA coordinates; BBKNN (bbknn) changes the graph; ComBat changes X. Keep original expression/counts for DE, and assess biological conservation and batch mixing together. Requested integration/doublet failures stop.
• The toolkit loads files into memory. For backed data, explicitly call .to_memory() on a chosen subset and close the file; backed mode is not a general out-of-core pipeline. Copy views before mutation. Dask support varies by function/array layout.
• .rds needs an explicit R conversion stage. Cluster IDs do not determine cell types: all mappings/signature assets are illustrative human marker examples, requiring review.

Harmony 2 returns cells x PCs; the toolkit calls its native API because the Scanpy 1.12.4 wrapper still transposes that output and fails with current Harmony.

Optional packages: scikit-image for Scrublet automatic thresholding, loompy for Loom, scikit-misc for v3 HVGs, harmonypy==2.0.2/bbknn==1.6.0 for their integration branches, and louvain for the deprecated Louvain option (separate environment: louvain 0.8.2 requires igraph <0.12, conflicting with the tested igraph 1.0). Install only needed branches. R and optional integration execution boundaries are recorded in references/upstream-review.md.

When to Use This Skill

This skill should be used when:

• Analyzing single-cell RNA-seq data (.h5ad, 10X, CSV formats)
• Working with R-friendly single-cell datasets (.rds, .RData, Seurat, SingleCellExperiment) that need conversion to .h5ad
• Performing quality control on scRNA-seq datasets
• Creating UMAP, t-SNE, or PCA visualizations
• Identifying cell clusters and finding marker genes
• Annotating cell types based on gene expression
• Conducting trajectory inference or pseudotime analysis
• Generating publication-quality single-cell plots

Script Toolkit (prefer these over writing code from scratch)

This skill bundles ready-to-run CLI scripts in scripts/ for every common step. Run these instead of hand-writing scanpy code — they handle file loading by extension, figure setup, sensible defaults, raw-count preservation, and progress logging. Each reads and writes .h5ad, so they chain together, and each has its own --help. Only drop down to writing scanpy code when a task isn't covered by a script or needs unusual customization.

All scripts use a shared scripts/_common.py helper (loading, saving, figure config) — keep it alongside the others. Run from the skill directory or pass full paths; figures default to ./figures/.

One-shot end-to-end run

# Counts → clustered object and exploratory marker ranks + figures + marker CSVs
python scripts/run_pipeline.py raw.h5ad -o processed.h5ad \
    --resolution 0.5 --n-top-genes 2000 --scrublet
# With multi-sample integration:
python scripts/run_pipeline.py raw.h5ad -o processed.h5ad --batch-key sample --batch-method harmony
# Reproducible parameters via JSON (keys mirror flag names with underscores):
python scripts/run_pipeline.py raw.h5ad -o processed.h5ad --config params.json

Step-by-step chain (when you need to inspect/iterate between stages)

python scripts/qc_analysis.py        raw.h5ad  -o qc.h5ad   --scrublet
python scripts/preprocess.py         qc.h5ad   -o norm.h5ad --n-top-genes 2000
python scripts/reduce_dimensions.py  norm.h5ad -o red.h5ad  --n-pcs 40
python scripts/cluster.py            red.h5ad  -o clu.h5ad  --resolution 0.3 0.5 0.8
python scripts/find_markers.py       clu.h5ad  -o clu.h5ad  --groupby leiden_0.5 --use-raw
# inspect results/markers/*.csv, decide labels, write a mapping JSON, then:
python scripts/annotate.py           clu.h5ad  -o ann.h5ad  --mapping celltypes.json --cluster-key leiden_0.5

The sections below document the underlying scanpy calls each script performs — read them when customizing beyond the script flags.

Quick Start

Basic Import and Setup

import scanpy as sc
import pandas as pd
import numpy as np

# Configure settings
sc.settings.verbosity = 3
sc.set_figure_params(dpi=80, facecolor='white')
sc.settings.figdir = './figures/'
sc.settings.autosave = True  # Preferred over per-plot save= (deprecated in scanpy 1.12)

Loading Data

# From 10X Genomics
adata = sc.read_10x_mtx('path/to/data/')
adata = sc.read_10x_h5('path/to/data.h5')

# From h5ad (AnnData format)
adata = sc.read_h5ad('path/to/data.h5ad')

# From CSV
adata = sc.read_csv('path/to/data.csv')

For R-native files, do not try to parse Seurat .rds directly in Python. Convert first:

# See references/r_interop.md for installing R and conversion packages.
Rscript convert_rds_to_h5ad.R input.rds output.h5ad

adata = sc.read_h5ad('output.h5ad')

Understanding AnnData Structure

The AnnData object is the core data structure in scanpy:

adata.X          # Expression matrix (cells × genes)
adata.obs        # Cell metadata (DataFrame)
adata.var        # Gene metadata (DataFrame)
adata.uns        # Unstructured annotations (dict)
adata.obsm       # Multi-dimensional cell data (PCA, UMAP)
adata.raw        # Snapshot of X/var; here full log-normalized expression, NOT counts

# Access cell and gene names
adata.obs_names  # Cell barcodes
adata.var_names  # Gene names

Standard Analysis Workflow

The seven steps, with code and the parameters that matter at each, are in references/analysis_workflow.md:

1. Quality control — filter cells and genes; inspect mitochondrial fraction and counts before choosing thresholds rather than copying defaults.
2. Normalization and preprocessing — normalize, log-transform, select highly variable genes, and keep .raw for later plotting.
3. Dimensionality reduction — PCA, then the neighbour graph, then UMAP.
4. Clustering — Leiden at a resolution chosen for the question, not the default.
5. Marker gene identification — ranked genes per cluster.
6. Cell type annotation — mapping clusters to types from markers.
7. Save results — writing the annotated AnnData.

Common follow-on tasks — publication plots, trajectory inference, pseudobulk differential expression between conditions, gene set scoring, and batch correction — are in the same file. See also references/standard_workflow.md and references/plotting_guide.md.

Key Parameters to Adjust

Quality Control

• min_genes: Minimum genes per cell (typically 200-500)
• min_cells: Minimum cells per gene (typically 3-10)
• pct_counts_mt: Mitochondrial threshold (typically 5-20%)

Normalization

• target_sum: Target counts per cell (scripts use 1e4; Scanpy default None uses a median)

Feature Selection

• n_top_genes: Number of HVGs (typically 2000-3000)
• min_mean, max_mean, min_disp: HVG selection parameters