single-cell-rna-qc_skill

Overview

This skill performs automated quality control on single-cell RNA-seq datasets (.h5ad or 10x .h5) following scverse/scanpy best practices. It runs MAD-based outlier detection, gene filtering, and produces comprehensive before/after visualizations. Outputs include filtered and annotated AnnData files plus PNG QC reports for quick inspection.

How this skill works

The recommended complete pipeline auto-detects input format, computes per-cell QC metrics (total counts, detected genes, mitochondrial/ribosomal/hemoglobin percentages), and applies MAD-based outlier detection plus optional hard MT% cutoffs. It filters low-quality cells and lowly-detected genes, writes a filtered .h5ad and an annotated copy with QC flags, and saves multiple diagnostic plots showing thresholds and distributions. Modular functions are available for custom workflows and selective execution of metrics, detection, filtering, or plotting.

When to use it

• You need QC on single-cell RNA-seq data (.h5ad or 10x .h5).
• You want MAD-based outlier filtering or to apply scverse/scanpy best practices.
• You need diagnostic visualizations before and after filtering.
• You want a quick, reproducible pipeline for batch processing datasets.
• You need modular utilities to build a custom QC workflow.

Best practices

• Start with the complete pipeline for standard datasets; it’s a conservative, ‘just works’ default.
• Be permissive initially—retain more cells to avoid losing rare populations; inspect plots before tightening thresholds.
• Always review pre- and post-filtering visualizations to ensure biological plausibility.
• Adjust mitochondrial gene patterns and thresholds for species or tissue-specific differences.
• Use modular functions when you need conditional logic or subset-specific QC rules.

Example use cases

• Run the convenience script on raw 10x output to get filtered .h5ad and PNG QC reports for downstream analysis.
• Calculate QC metrics and only generate visualizations to decide thresholds interactively before filtering.
• Apply only a mitochondrial-percentage hard cutoff while leaving other metrics permissive.
• Build a custom workflow that computes metrics, applies different filters per cell type, and then recombines results.
• Batch-process multiple datasets with consistent MAD thresholds and collect per-dataset QC summaries.