scrna-pipeline-conventions

Assay-specific conventions for restartable scRNA-seq objects and milestone packaging. Use when choosing AnnData checkpoints and deliverables, packaging gene identifiers or dual embeddings, cleaning ob

SKILL.md

GitHub

scRNA-seq Pipeline Conventions

Use this skill for the scRNA-specific objects and deliverables shared by cellranger-multi-to-anndata, scrna-cxg-host, and consensus-nmf-multirun. Apply the CRAFT block and analysis-code-conventions for stage naming, layout, helper boundaries, configuration placement, and generic restartability.

Routing decision tree

What scRNA deliverable decision is being made?
│
├─ Is this an intermediate checkpoint or a finalized object?
│     → checkpoints-and-objects.md
│
├─ Which gene identifier should be active at delivery?
│     → gene-identifier-packaging.md
│
├─ Which embeddings must the packaged object carry?
│     → embedding-deliverables.md
│
└─ How should obs metadata and uncertain labels be packaged?
      → metadata-packaging.md

Routing contract

  1. Apply the repository's rendered CRAFT block for standing stage and result mechanics; load analysis-code-conventions for structural judgment.
  2. Select the scRNA deliverable question and read its reference completely.
  3. Load every relevant reference when finalizing an object because identifier, embedding, metadata, and derivation guarantees interact.
  4. Read numeric thresholds and project choices from the project's current configuration and decision records. Preserve their provenance.
  5. Use the adjacent conversion or hosting skill for tool-specific execution.

References (load on demand)

Intent Load
Plan recovery objects and the canonical milestone references/checkpoints-and-objects.md
Package symbols while preserving stable identifiers references/gene-identifier-packaging.md
Carry discovery and integrated embeddings references/embedding-deliverables.md
Clean obs and preserve annotation uncertainty references/metadata-packaging.md

Adjacent skills

  • analysis-code-conventions — structural stages, helpers, configuration, and restartable data flow.
  • cellranger-multi-to-anndata — build the initial pooled AnnData object.
  • single-cell-rna-qc — perform assay QC before milestone packaging.
  • anndatar-seurat-scanpy-conversion — derive a Seurat object from AnnData.
  • louper-seurat-conversion — derive a Loupe object from the packaged data.
  • scrna-cxg-host — validate and host the CellxGene derivative.

When not to use

  • Generic stage sequencing, helper extraction, or project layout decisions.
  • Assay methodology such as QC models, integration choice, or differential tests.
  • Throwaway exploration with no durable scRNA object or collaborator handoff.