scRNA Preprocessing And Clustering

Version Compatibility

Reference examples assume:

• scanpy 1.10+
• anndata 0.10+
• pandas 2.2+
• matplotlib 3.8+

Before using code patterns, verify installed versions match the environment:

• Python: python -c "import scanpy, anndata; print(scanpy.__version__, anndata.__version__)"
• If signatures differ, inspect the installed API and adapt the pattern instead of retrying unchanged.

Overview

Use this skill to turn raw or minimally processed scRNA-seq data into an analysis-ready object with:

• QC-filtered cells and genes
• normalized expression values
• highly variable genes
• PCA and UMAP embeddings
• Leiden clusters
• saved h5ad artifact for annotation, DE, integration, or trajectory analysis

When To Use This Skill

• raw 10x matrices, filtered count matrices, or h5ad inputs need standard preprocessing
• the user wants UMAP, clustering, or marker discovery
• downstream tasks depend on a stable single-cell object rather than ad hoc plots

Quick Route

• If the input is already a processed h5ad, inspect adata.raw, embeddings, cluster columns, and QC columns before rerunning preprocessing.
• If the input is raw counts, do QC first and only normalize after filtering obvious low-quality cells.
• If multiple batches are present, preprocess cleanly first, then consider integration instead of hiding batch effects with aggressive filtering.

Progressive Disclosure

• Read technical_reference.md for QC decision rules, assay caveats, and integration branching.
• Read commands_and_thresholds.md for concrete Scanpy code, default thresholds, and output conventions.

Default Rules

• Keep raw counts recoverable. Prefer adata.raw = adata.copy() before regression or scaling.
• Report thresholds explicitly. Do not silently drop cells or genes.
• Show QC distributions before applying hard filters.
• Use vector outputs such as .pdf or .svg for final figures when possible.

Expected Inputs

• 10x directory, .h5, .h5ad, or count matrix
• cell metadata if available
• species context for mitochondrial or ribosomal gene detection

Expected Outputs

• results/processed.h5ad
• qc/cell_qc_metrics.tsv
• qc/gene_qc_metrics.tsv
• figures/qc_violin.pdf
• figures/pca_variance_ratio.pdf
• figures/umap_leiden.pdf

Preferred Tools

• scanpy
• anndata
• pandas
• matplotlib
• seaborn

Starter Pattern

import scanpy as sc

adata = sc.read_10x_mtx("counts/")
adata.var_names_make_unique()
adata.var["mt"] = adata.var_names.str.upper().str.startswith("MT-")
sc.pp.calculate_qc_metrics(adata, qc_vars=["mt"], inplace=True)

adata = adata[
    (adata.obs["n_genes_by_counts"] >= 200)
    & (adata.obs["n_genes_by_counts"] <= 6000)
    & (adata.obs["pct_counts_mt"] < 15),
    :
].copy()

sc.pp.filter_genes(adata, min_cells=3)
sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
adata.raw = adata.copy()

sc.pp.highly_variable_genes(adata, n_top_genes=3000, flavor="seurat_v3")
adata = adata[:, adata.var["highly_variable"]].copy()
sc.pp.scale(adata, max_value=10)
sc.tl.pca(adata, svd_solver="arpack")
sc.pp.neighbors(adata, n_neighbors=15, n_pcs=30)
sc.tl.umap(adata)
sc.tl.leiden(adata, resolution=0.5, key_added="leiden_r05")
adata.write("results/processed.h5ad")

Workflow

1. Load and validate the object

• confirm orientation is cells by genes
• make gene names unique
• record sample IDs and batch labels before merging or filtering

2. Compute QC metrics and inspect distributions

• n_genes_by_counts
• total_counts
• pct_counts_mt
• optional ribosomal or hemoglobin fractions

Plot distributions before filtering. Thresholds vary by chemistry, tissue, and nucleus versus whole-cell assay.

3. Filter cells and genes

Use dataset-aware thresholds. Good first-pass defaults:

• min_genes >= 200
• max_genes <= 5000-8000 to remove likely doublets in many droplet datasets
• pct_counts_mt < 10-20 depending on tissue stress
• min_cells >= 3 for genes

4. Normalize, log-transform, and select HVGs

• normalize with target_sum=1e4
• log1p
• select 2000-4000 HVGs
• save raw counts before heavy transformations

5. Reduce dimensions and cluster

• PCA on HVGs
• neighbor graph using 10-30 PCs and 10-30 neighbors as a starting range
• UMAP for visualization
• Leiden across a small resolution grid such as 0.2, 0.5, 0.8, 1.0

6. Export analysis-ready artifacts

Always save:

• processed h5ad
• QC tables
• cluster assignments
• publication-ready QC and UMAP figures

Output Artifacts

• results/processed.h5ad: main reusable AnnData object
• results/cluster_assignments.tsv: barcode plus cluster labels
• qc/filter_summary.tsv: counts before and after filtering
• figures/umap_leiden.pdf: main embedding figure

Quality Review

• Median genes per cell should be plausible for the chemistry and tissue.
• Mitochondrial fraction should not dominate retained cells.
• PCA variance should decay smoothly rather than showing obvious technical axes only.
• UMAP should be reviewed together with QC metrics and batch labels, not alone.
• Cluster labels should not be finalized before marker inspection.

Anti-Patterns

• reprocessing an already integrated object as if it were raw counts
• using a single universal mitochondrial threshold for every tissue
• interpreting UMAP separation as biology before checking batch and QC covariates
• discarding raw counts needed later for DE or pseudobulk

Related Skills

• Cell Annotation
• Cell Communication
• Trajectory And Lineage
• Multiome And scATAC

Optional Supplements

• anndata
• scanpy