scatac-preprocessing

name	scatac-preprocessing
description	Single-cell ATAC-seq preprocessing with a Signac-style TF-IDF + LSI workflow. Performs cell and peak filtering, top-peak selection, TF-IDF normalization, latent semantic indexing, neighborhood graph construction, UMAP, and Leiden clustering, then exports a downstream-ready AnnData plus a standard OmicsClaw gallery and reproducibility bundle.
version	0.1.0
author	OmicsClaw
license	MIT
tags	["singlecell","scatac","atac","preprocessing","tfidf","lsi","clustering"]
metadata	{"omicsclaw":{"domain":"singlecell","allowed_extra_flags":["--leiden-resolution","--method","--min-cells","--min-peaks","--n-lsi","--n-neighbors","--n-top-peaks","--tfidf-scale-factor"],"param_hints":{"tfidf_lsi":"[Truncated]"},"legacy_aliases":["scatac-preprocess"],"saves_h5ad":true,"requires_preprocessed":false,"requires":{"bins":"[Truncated]","env":"[Truncated]","config":"[Truncated]"},"emoji":"🧬","homepage":"https://github.com/OmicsClaw/OmicsClaw","os":["macos","linux"],"install":["[Truncated]"],"trigger_keywords":["scATAC preprocessing","single-cell ATAC preprocessing","ATAC TF-IDF LSI","chromatin accessibility clustering","scATAC UMAP Leiden"]}}

🧬 scATAC Preprocessing

You are scATAC Preprocessing, the OmicsClaw skill for turning raw-count-like single-cell chromatin accessibility input into a downstream-ready AnnData with QC-filtered cells, selected peaks, TF-IDF-transformed feature space, LSI embedding, neighborhood graph, UMAP, clustering, and a standard OmicsClaw result bundle.

Why This Exists

• Without it: users manually piece together sparse peak filtering, TF-IDF, LSI, graph construction, and clustering with inconsistent export behavior
• With it: one command produces a stable scATAC preprocessing object plus a standard OmicsClaw report, gallery, tables, and reproducibility bundle
• Why OmicsClaw: the wrapper keeps a compact, implementation-aligned contract around a Signac-style TF-IDF + LSI workflow while exposing only the highest-value controls

Scope Boundary

Current OmicsClaw scatac-preprocessing exposes one implemented workflow:

1. tfidf_lsi

This skill does:

1. cell filtering by detected peaks
2. peak filtering by cell support
3. wrapper-level top-peak selection
4. TF-IDF normalization
5. truncated SVD / LSI embedding
6. graph construction, UMAP, and Leiden clustering
7. downstream-ready AnnData export

This skill does not:

1. start from fragment files
2. call peaks from BAM / fragments
3. compute motif enrichments
4. compute gene activity scores
5. perform differential accessibility
6. perform multi-sample integration

Those belong to later OmicsClaw scATAC skills.

Core Capabilities

1. Implementation-aligned scATAC preprocessing: sparse accessibility matrix to TF-IDF + LSI + graph + clustering in one run
2. Compact public controls: a small parameter set tunes QC sparsity thresholds, feature budget, latent dimensionality, graph locality, and clustering granularity
3. Stable gallery contract: UMAP clusters, QC violin, top accessible peaks, and LSI variance figures under figures/
4. Structured figure-data layer: figure_data/ exports summary, cluster, peak, LSI, UMAP, and QC tables for downstream customization
5. Downstream-ready AnnData: processed peak matrix, raw counts layer, raw snapshot over retained peaks, LSI embeddings, graph, and cluster labels