bio-atac-seq-single-cell-atac

What it does

It instructs an AI agent to process and analyze 10X scATAC-seq data (and paired scRNA-seq for Multiome) by building per-cell fragment matrices, performing per-cell QC, and choosing among Signac, ArchR, SnapATAC2 ecosystems for downstream analyses. It covers steps for per-cluster pseudobulk peak calling, integrating with RNA data, and various doublet-detection strategies, plus guidance on per-cluster differential accessibility and cell-type annotation.

How it works

The skill enumerates multiple concrete workflows across ecosystems:

• R/Signac path: CreateChromatinAssay, Seurat-based pipeline with TF-IDF, SVD, UMAP, Leiden, and a gene-activity assay for annotation.
• R/ArchR path: createArrowFiles, doublet scoring, ArchRProject, iterative LSI, UMAP, clustering, then per-cluster reproducible peak sets.
• Python/SnapATAC2 path: import_fragments, QC metrics, tile matrix, spectral embedding, UMAP, Leiden, and per-cluster MACS3 peaks, plus gene activity.
• CLI/Cell Ranger ATAC path: preprocessing via cellranger-atac count or chromap at the preprocessing layer.
• Reconciliation rules across tools to compare UMAP shapes, doublet rates, and cluster boundaries, plus an operational rule to require cross-ecosystem agreement for cell-type annotation.
• Per-cluster pseudobulk peak calling via MACS3 with guidance on cluster size.
• Per-tool failure modes with fixes for known pitfalls (depth-related first component in Signac, TileMatrix vs PeakMatrix, memory quirks in SnapATAC2, ATAC-specific empty droplets in Cell Ranger ATAC, WNN weighting in Multiome).

When to use it

Use when processing 10X scATAC data from cellranger output or Multiome, needing per-cell QC, ecosystem choice guidance, cluster-level peak calling, scRNA-seq integration, or doublet detection. Triggered by workflows that require per-cell QC, cross-ecosystem comparisons, or differential accessibility analyses per cluster.

What it can touch

It mentions tools and commands such as:

• R: Signac, Seurat, ArchR, and associated functions like CreateChromatinAssay, RunTFIDF, RunSVD, RunUMAP, FindNeighbors, FindClusters, GeneActivity, CallPeaks, etc.
• Python: snapatac2 with functions like import_fragments, add_tile_matrix, spectral, UMAP, Leiden, macs3, make_gene_matrix.
• CLI: cellranger-atac and chromap.
• External tools: MACS3, AMULET, scDblFinder, scater, MOFA+, SCENIC.

Caveats

Lists known limitations and failure modes for different tools, including depth confounding in Signac components, TileMatrix vs PeakMatrix usage, memory layout requirements in SnapATAC2, and specific caveats for Multiome WNN weighting. Also notes to verify chemistry type (10X Multiome vs single-modality scATAC) before analysis and to standardize clustering parameters when reconciling across tools.