---

name: core-analysis description: "Core OmicVerse AnnData IO, datasets, preprocessing, plotting, reporting, reproducibility, and registry discovery workflows." disable-model-invocation: true metadata: disco-role: operating license: GPL 3.0

Core Analysis

Use this sub-skill for OmicVerse tasks centered on an AnnData object: loading data, checking QC, preprocessing, dimensionality reduction, neighbor graph construction, lightweight clustering prerequisites, plotting, reports, reproducibility, and package registry discovery.

For root routing and install strategy, return to ../../SKILL.md.

When to Use

• Read or write common matrix objects with ov.read, ov.io.read, ov.io.read_h5ad, ov.io.read_10x_mtx, ov.io.read_10x_h5, ov.io.read_csv, ov.io.save, or ov.io.load.
• Load or synthesize example data with ov.datasets.*, especially ov.datasets.create_mock_dataset(...) for no-network testing.
• Run core QC and preprocessing with ov.pp.qc_metrics, ov.pl.qc, ov.pp.qc, ov.pp.preprocess, ov.pp.scale, ov.pp.pca, ov.pp.neighbors, ov.pp.umap, ov.pp.leiden, or ov.pp.louvain.
• Plot embeddings or QC distributions with ov.pl.embedding, ov.pl.umap, ov.pl.pca, ov.pl.qc, palettes, and plot style helpers.
• Create a one-file HTML pipeline report with ov.report.from_anndata(...) or inspect provenance with ov.report.get_provenance(...).
• Discover registered OmicVerse functions through ov.list_functions, ov.get_function_help, ov.find_function, ov.recommend_function, or ov.export_registry.

Route Elsewhere

• Use ../single-cell-workflows/SKILL.md for annotation, marker ranking, trajectory, pseudotime, velocity/fate, batch integration, communication, pseudobulk, and single-cell biological interpretation.
• Use ../multiomics-statistics/SKILL.md for bulk RNA-seq, enrichment/signature scoring, metabolomics, proteomics, microbiome, and table-based omics statistics.
• Use ../spatial-integration/SKILL.md for Visium/Xenium/Nanostring spatial workflows, histology, deconvolution, tissue zones, and spatial mapping.
• Use ../agentic-and-mcp/SKILL.md for CLI, MCP, JARVIS, registry manifests, skill seeker, and agent runtime configuration.

Fast Start

import omicverse as ov

ov.set_seed(0)
adata = ov.io.read_10x_mtx("filtered_feature_bc_matrix", var_names="gene_symbols")
adata.var_names_make_unique()

ov.pp.qc_metrics(adata)
fig = ov.pl.qc(adata, tresh={"mito_perc": 0.2, "nUMIs": 500, "detected_genes": 250})
fig.savefig("qc.png", dpi=150, bbox_inches="tight")

adata = ov.pp.qc(
    adata,
    tresh={"mito_perc": 0.2, "nUMIs": 500, "detected_genes": 250},
    doublets=False,
)
ov.pp.preprocess(adata, mode="shiftlog|pearson", n_HVGs=2000, target_sum=1e4)
ov.pp.scale(adata)
ov.pp.pca(adata, n_pcs=50)
ov.pp.neighbors(adata, n_neighbors=15, n_pcs=50)
ov.pp.umap(adata)

ov.pl.embedding(adata, basis="X_umap", color=["nUMIs", "mito_perc"], show=False)
ov.report.from_anndata(adata, output="core_report.html", title="Core preprocessing")
adata.write_h5ad("processed.h5ad", compression="gzip")

Expected core slots after this workflow:

• adata.layers['counts']: raw counts preserved by ov.pp.preprocess.
• adata.var['highly_variable'] and adata.var['highly_variable_features']: selected HVGs.
• adata.layers['scaled']: scaled expression from ov.pp.scale.
• adata.obsm['X_pca'], adata.varm['PCs'], adata.uns['pca']: PCA outputs.
• adata.uns['neighbors'], adata.obsp['distances'], adata.obsp['connectivities']: neighbor graph.
• adata.obsm['X_umap']: UMAP coordinates.
• adata.uns['_ov_provenance']: best-effort report provenance for tracked calls.

References

• references/core-workflows.md: end-to-end AnnData workflows, validation checkpoints, and plotting/reporting patterns.
• references/api-reference.md: concrete API names, signatures, slot effects, and registry helpers.
• references/data-formats.md: input formats, 10x layouts, tabular safety, AnnData slot assumptions, and output guidance.
• references/troubleshooting.md: lazy import failures, workflow-order errors, plotting backends, downloads, warnings, and optional dependency issues.

Bundled Check

Run the packaged introspection script before deeper debugging:

python sub-skills/core-analysis/scripts/inspect_core.py --help
python sub-skills/core-analysis/scripts/inspect_core.py --json
python sub-skills/core-analysis/scripts/inspect_core.py --smoke-mock --mock-cells 40 --mock-genes 80

The script reports OmicVerse version, lazy root attributes, selected module imports, registry size, and optionally a small synthetic AnnData smoke. It does not require the original repository checkout and does not download data.

Guardrails

• Preserve raw counts before filtering or normalization; keep them in .layers['counts'] or .raw when possible.
• Run QC metrics before threshold plots: ov.pl.qc raises when standard QC metrics are missing.
• Follow the order qc_metrics or qc → preprocess → scale → pca → neighbors → umap → plotting/reporting.
• Treat dataset loaders as potentially networked unless the function is explicitly synthetic, such as create_mock_dataset.
• If a lazy import says Failed to import omicverse.<module>, isolate the missing optional dependency by importing that module directly and checking its traceback.