---

name: preprocessing-qc description: "Use Scanpy preprocessing for filtering, QC metrics, normalization/log transforms, HVG selection, scaling/regression, PCA preparation, sampling/downsampling, basic batch correction wrappers, and Scrublet doublet workflows." disable-model-invocation: true metadata: disco-role: operating license: BSD 3-Clause

Scanpy Preprocessing and QC

Use this sub-skill when a task needs to prepare an AnnData object for graph construction, embedding, clustering, marker testing, or reporting with import scanpy as sc and sc.pp.* APIs.

Route by task

User need	Use	Details
Filter cells/genes or compute QC metrics	sc.pp.filter_cells, sc.pp.filter_genes, sc.pp.calculate_qc_metrics	See QC and filtering.
Preserve raw counts, normalize, or log-transform	adata.layers, sc.pp.normalize_total, sc.pp.log1p	See Normalization, log, and layers.
Select HVGs, including batch-aware or Seurat v3 flavors	sc.pp.highly_variable_genes	See Highly variable genes.
Scale, regress covariates, or compute PCA features	sc.pp.scale, sc.pp.regress_out, sc.pp.pca	See Scaling, regression, and PCA.
Sample cells/counts before expensive preprocessing	sc.pp.sample, sc.pp.downsample_counts, legacy sc.pp.subsample	See Sampling and downsampling.
Apply simple preprocessing batch correction	sc.pp.combat, sc.pp.harmony_integrate	See Basic batch correction.
Detect doublets or simulate doublets	sc.pp.scrublet, sc.pp.scrublet_simulate_doublets	See Doublet detection.
Apply bundled preprocessing recipes	sc.pp.recipe_seurat, sc.pp.recipe_zheng17, sc.pp.recipe_weinreb17	See Recipes.
Check a local install with a tiny fixture	bundled script	Run python sub-skills/preprocessing-qc/scripts/scanpy_preprocess_qc_smoke.py.

Fast rules

• Preserve count data before destructive transforms: adata.layers["counts"] = adata.X.copy() before normalize_total, log1p, scale, or regress_out.
• Make QC and HVG calls explicit about representation: pass layer="counts" or use_raw=True only when that slot contains the intended matrix.
• Keep calculate_qc_metrics(..., percent_top=...) values no larger than adata.n_vars; use percent_top=None or small values for tiny fixtures.
• Use inplace=False to inspect return values for calculate_qc_metrics, filter_*, normalize_total, or highly_variable_genes; use copy=True only where the API documents copied AnnData returns.
• Prefer scale(zero_center=False) for sparse-memory-sensitive workflows; scale(zero_center=True), regress_out, and ComBat can densify data.
• Run pca before harmony_integrate, then route neighbors, UMAP, clustering, marker ranking, graph plotting, and embedding interpretation to graph-embedding-analysis.
• Route IO/data loading to io-data-access, plotting/report figures to plotting-reporting, and optional integration method selection beyond core Scanpy wrappers to external-integrations.

Required references

• Workflows
• API quick reference
• Troubleshooting
• Smoke script