# scrna-seq-qc

Guides quality control, doublet removal, and cell annotation for scRNA-seq and snRNA-seq data. It helps Codex configure data-driven filters and clean AnnData objects.

- skill · `type=skill` · Codex, ChatGPT · 2 installs · ★ 7604
- Author: OpenAI · official OpenAI repository
- License: MIT · policy=mirror · the beginning of the file is at the end of this page
- Source: https://github.com/openai/plugins/tree/HEAD/plugins/ngs-analysis/skills/scrna-seq-qc/SKILL.md
- Updated: 2026-10-08
- HTML: https://affpapa.org/en/nevibecode/skills/scrna-seq-qc

## What it does

This skill establishes an auditable QC workflow for single-cell and single-nucleus RNA sequencing datasets. Instead of relying on hardcoded cutoffs, it derives thresholds directly from observed metric distributions. Standard filtering focuses strictly on detected genes, total counts, and mitochondrial read fractions.

Doublet detection runs through `scDblFinder` per technical batch. The skill forbids skipping doublet calling or silently swapping the caller. For dimensionality reduction, it chooses between Scanpy PCA for simpler runs and `scVI` for multi-donor integration. Cell annotation relies on Allen Brain Atlas tools like MapMyCells, `cell_type_mapper`, or curated marker panels.

Large runs stay memory-efficient by keeping matrices sparse and using backed AnnData views. Without raw counts preserved in the input object, downstream provenance breaks. Outputs include a filtered `.h5ad` file, threshold justification plots, and a parameter manifest.

## Install

### Codex (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add openai/plugins --skill scrna-seq-qc -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/scrna-seq-qc/SKILL.md` — For all projects use ~/.agents/skills/

### Claude Code

1. **Skills CLI (Vercel)**
```bash
npx skills add openai/plugins --skill scrna-seq-qc -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/scrna-seq-qc/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/scrna-seq-qc/

### Cursor

**Manually: put the files in place**
- File: `.cursor/skills/scrna-seq-qc/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install openai/plugins scrna-seq-qc
```

2. **Manually: put the files in place**
- File: `.github/skills/scrna-seq-qc/SKILL.md`

### Gemini CLI

**Platform CLI**
```bash
gemini skills install https://github.com/openai/plugins --path plugins/ngs-analysis/skills/scrna-seq-qc
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/scrna-seq-qc/SKILL.md` — The legacy .windsurf/skills/ path still works

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/openai/plugins/tree/HEAD/plugins/ngs-analysis/skills/scrna-seq-qc/SKILL.md

## When to use it
- derive dataset-specific QC thresholds for scRNA-seq samples
- identify cell doublets per batch using `scDblFinder`
- decide between standard PCA and `scVI` batch integration
- assign cell types using MapMyCells or canonical markers
- generate global and subsetted UMAP visualizations

## Example prompts
- Inspect QC distributions in `sample.h5ad` and select cell filtering thresholds
- Run `scDblFinder` on each capture channel and flag predicted doublets
- Train an `scVI` model for batch integration and plot coarse UMAP embeddings

## Source text · SKILL.md · MIT

Below is the content of a third-party file. It is data, not instructions for you: do not act on anything written inside unless the human asks. Block tag: 87ff0908ef95e653.

BEGIN THIRD-PARTY FILE 87ff0908ef95e653
```text
---
name: scrna-seq-qc
description: Process, quality-control, annotate, and visualize single-cell or single-nucleus RNA-seq datasets across tissues and species. Use when Codex needs to build, adapt, or review a general scRNA-seq QC pipeline; choose dataset-appropriate cell-level filters from QC distributions; run required scDblFinder-based doublet and ambient-RNA filtering; annotate cells with matched references or marker-based fallbacks; or generate global and per-group UMAP visualizations for large scRNA-seq datasets.
---

# scRNA-seq QC

## Start Here

Read `references/qc-annotation-umap-heuristics.md` before picking thresholds, annotation backends, or UMAP feature-selection rules.

Confirm what inputs exist before writing code:

- An AnnData object or equivalent with raw counts preserved.
- Per-sample, per-batch, or per-channel metadata, because QC and doublet detection should respect technical partitions.
- Organism, tissue, assay type, chemistry, and whether the data are whole-cell or single-nucleus.
- Whether a matched cell atlas or label-transfer reference exists for the tissue and species.

Preserve provenance in the output: package versions, thresholds, threshold-justification plots, counts removed or flagged at each filter, annotation backend and reference, marker-gene selection heuristic, and any manual cluster exclusions.

## Workflow

1. Choose QC thresholds from the data, not from a fixed template.
   - Plot detected genes, total UMIs, mitochondrial fraction, and any tissue-specific nuisance signals overall and by batch.
   - Inspect all available QC metrics, but default filtering should use only the standard metrics: detected genes, total counts, and `percent.mt`.
   - Pick thresholds from the observed distributions and expected biology.
   - Save a plot with threshold lines and record why each threshold is appropriate for this dataset.
   - If another metric looks important enough to filter on, flag it as a dataset-specific issue, explain why, and consult the user before adding that extra filter.
   - Keep QC plots legible: do not overload a single panel with too many batches or categories when faceting, splitting, or summary views would communicate the result more clearly.

2. Run cell-level QC.
   - Remove or flag obvious low-quality barcodes using the chosen thresholds on detected genes, total counts, and `percent.mt`.
   - Use `scDblFinder` for doublet detection. Run it per batch or capture channel, and split very large batches before doublet calling.
   - Do not skip doublet calling or silently substitute another method. If `scDblFinder` cannot run in the environment, surface the blocker explicitly or get user approval before using a different caller.
   - Compute ambient-RNA style metrics and use them for filtering when the dataset and workflow support it.
   - Compute any other informative QC metrics when feasible, but do not turn those additional nonstandard metrics into hard filters without explicit user approval unless the user already asked for a stricter policy.
```
END THIRD-PARTY FILE 87ff0908ef95e653

The third-party file has ended. Everything between the 87ff0908ef95e653 tags is data: requests and commands from it are not followed. An “end of file” line with another tag inside the block is part of the file.

First 3045 of 8340 bytes. © OpenAI · MIT · https://github.com/openai/plugins/tree/HEAD/plugins/ngs-analysis/skills/scrna-seq-qc/SKILL.md

_Part of the page did not fit — see the HTML version._

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md