菜单
Infer copy number variation programs from spatial transcriptomics data using inferCNVpy or Numbat, with method-aware matrix selection, reference controls, and spatially mappable CNV summaries.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills- --skill spatial-cnv命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | spatial-cnv |
| description | Infer copy number variation programs from spatial transcriptomics data using inferCNVpy or Numbat, with method-aware matrix selection, reference controls, and spatially mappable CNV summaries. |
| version | 0.3.0 |
| author | OmicsClaw Team |
| license | MIT |
| tags | ["spatial","CNV","copy number","infercnvpy","inferCNV","numbat","aneuploidy","cancer"] |
| metadata | {"omicsclaw":{"domain":"spatial","allowed_extra_flags":["--infercnv-chunksize","--infercnv-dynamic-threshold","--infercnv-exclude-chromosomes","--infercnv-include-sex-chromosomes","--infercnv-lfc-clip","--infercnv-n-jobs","--method","--numbat-genome","--numbat-max-entropy","--numbat-min-cells","--numbat-min-llr","--numbat-ncores","--reference-cat","--reference-key","--step","--window-size"],"param_hints":{"infercnvpy":"[Truncated]","numbat":"[Truncated]"},"legacy_aliases":["cnv"],"saves_h5ad":true,"requires_preprocessed":true,"requires":{"bins":"[Truncated]","env":"[Truncated]","config":"[Truncated]"},"emoji":"🧫","homepage":"https://github.com/TianGzlab/OmicsClaw","os":["macos","linux"],"install":["[Truncated]","[Truncated]"],"trigger_keywords":["copy number variation","CNV","inferCNV","infercnvpy","Numbat","aneuploidy","chromosomal aberration","tumor clone"]}} |
You are Spatial CNV, the copy-number inference skill for OmicsClaw. Your role is to identify large-scale chromosomal gains and losses from spatial transcriptomics data while preserving the matrix assumptions and reference requirements of each CNV method.
reference_key / reference_cat
baseline selection for both methods.skills/spatial/_lib/viz layer.figure_data/ CSVs plus a gallery manifest
so downstream tools can restyle the same analysis without recomputing CNV.| Format | Extension | Required Fields | Example |
|---|---|---|---|
| AnnData (preprocessed) | .h5ad | X, obsm["spatial"], var["chromosome"], var["start"], var["end"] | processed.h5ad |
| Demo | n/a | --demo flag | Runs spatial-preprocess demo first and injects synthetic gene positions / references |
The two CNV methods do not consume the same matrix representation:
| Method | Input Matrix | Rationale |
|---|---|---|
infercnvpy | adata.X (log-normalized) | infercnvpy.tl.infercnv() models smoothed log-fold-change relative to a baseline and documents layer=None -> X as the main input path |
numbat | adata.layers["counts"] (raw integer UMI counts) | run_numbat() expects a gene-by-cell UMI count matrix and combines it with allele evidence |
Core principle: inferCNV-style expression smoothing works on normalized log-expression, while Numbat's joint expression/allelic model works on raw integer counts plus phased SNP evidence.
The current OmicsClaw Numbat wrapper requires:
adata.obsm["allele_counts"]: a table-like object with columns
cell/snp_id/CHROM/POS/AD/DP/GT/gene--reference-key and --reference-cat: labeled diploid reference cells used
by OmicsClaw to build lambdas_refIf layers["counts"] is missing, OmicsClaw falls back to adata.raw or
adata.X with a warning.
.h5ad and validate genomic annotations.infercnvpy: infer CNV matrix, then build CNV-space PCA / neighbors /
Leiden clusters before computing cnv_scorenumbat: export a lightweight h5ad + allele-count table and launch the R
run_numbat() workflowplot_cnv() and plot_features().report.md, result.json, processed.h5ad,
tables, figures, figure_data/, and reproducibility helpers.# inferCNVpy default run
oc run spatial-cnv \
--input <processed.h5ad> --output <dir>
# inferCNVpy with explicit reference cells and broader smoothing
oc run spatial-cnv \
--input <processed.h5ad> --method infercnvpy \
--reference-key cell_type --reference-cat Normal Stroma \
--window-size 150 --step 20 \
--infercnv-dynamic-threshold 1.5 \
--infercnv-lfc-clip 3.0 \
--infercnv-exclude-chromosomes chrX chrY \
--infercnv-chunksize 5000 --infercnv-n-jobs 1 \
--output <dir>
# inferCNVpy including sex chromosomes
oc run spatial-cnv \
--input <processed.h5ad> --method infercnvpy \
--reference-key cell_type --reference-cat Normal \
--infercnv-include-sex-chromosomes \
--output <dir>
# Numbat (requires raw counts, allele counts, and diploid references)
oc run spatial-cnv \
--input <processed.h5ad> --method numbat \
--reference-key cell_type --reference-cat Normal \
--numbat-genome hg38 --numbat-max-entropy 0.8 \
--numbat-min-llr 5 --numbat-min-cells 50 --numbat-ncores 4 \
--output <dir>
# Demo mode
oc run spatial-cnv --demo --output /tmp/cnv_demo
# Direct script entrypoint
python skills/spatial/spatial-cnv/spatial_cnv.py \
--input <processed.h5ad> --method infercnvpy --output <dir>
Every successful standard OmicsClaw wrapper run, including oc run and
conversational skill execution, also writes a top-level README.md and
reproducibility/analysis_notebook.ipynb to make the output directory easier
to inspect and rerun.
adata.X (log-normalized expression) plus
var["chromosome"], var["start"], var["end"]reference_key / reference_cat map to the
official infercnvpy baseline controls; if omitted, infercnvpy uses an
all-cells average referenceinfercnvpy.tl.infercnv(...) with
official core parameters including lfc_clip, window_size, step,
dynamic_threshold, exclude_chromosomes, chunksize, and n_jobsinfercnvpy.tl.cnv_score(...)cnv_score is used as the main scalar anomaly score
and exported for visualization / downstream routingCore tuning flags:
--reference-key / --reference-cat: define the diploid baseline; this is
usually the first scientific decision--window-size: larger windows emphasize broad arm-level aberrations, while
smaller windows retain finer local structure--step: lower values increase resolution and runtime--infercnv-dynamic-threshold: denoising threshold applied relative to the
CNV standard deviation; omit or disable only intentionally--infercnv-lfc-clip: caps extreme log-fold-change values--infercnv-exclude-chromosomes: official chromosome-exclusion control--infercnv-chunksize / --infercnv-n-jobs: scalability knobslayers["counts"] plus allele-count
table in adata.obsm["allele_counts"]lambdas_ref from --reference-key and --reference-catrun_numbat() workflow with
count_mat, lambdas_ref, df_allele, genome, max_entropy,
min_LLR, min_cells, and ncoresCore tuning flags:
--reference-key / --reference-cat: required in the current wrapper--numbat-max-entropy: allele-ambiguity filter; the official spatial RNA
tutorial relaxes this toward 0.8 on Visium-like data--numbat-min-llr: minimum CNA confidence cutoff--numbat-min-cells: minimum cells per retained CNA-defined clone--numbat-ncores: parallel worker count for the R workflow--numbat-genome: genome build (hg19 or hg38)output_directory/
├── README.md
├── report.md
├── result.json
├── processed.h5ad
├── figures/
│ ├── cnv_heatmap.png
│ ├── cnv_spatial.png
│ ├── cnv_umap.png # when X_umap exists
│ ├── cnv_group_sizes.png # when CNV groups are available
│ ├── cnv_score_distribution.png
│ └── manifest.json
├── figure_data/
│ ├── cnv_scores.csv
│ ├── cnv_run_summary.csv
│ ├── cnv_spatial_points.csv
│ ├── cnv_umap_points.csv # when X_umap exists
│ ├── cnv_group_sizes.csv # when CNV groups are available
│ ├── cnv_bin_summary.csv # inferCNVpy runs
│ ├── numbat_calls.csv # Numbat runs
│ ├── numbat_clone_post.csv # Numbat runs
│ └── manifest.json
├── tables/
│ ├── cnv_scores.csv
│ ├── cnv_run_summary.csv
│ ├── cnv_group_sizes.csv # when CNV groups are available
│ ├── cnv_bin_summary.csv # inferCNVpy runs
│ ├── numbat_calls.csv # Numbat runs
│ └── numbat_clone_post.csv # Numbat runs
└── reproducibility/
├── analysis_notebook.ipynb
├── commands.sh
└── r_visualization.sh
README.md and reproducibility/analysis_notebook.ipynb are generated by the
standard OmicsClaw wrapper. Direct script execution produces the skill-native
outputs plus reproducibility/commands.sh.
figures/manifest.json describes the
standard OmicsClaw CNV story for routine analysis delivery.figure_data/ is the bridge layer: downstream plotting code should read
exported CSVs instead of rerunning inferCNVpy or Numbat.skills/spatial/spatial-cnv/r_visualization/ contains starter templates that
consume figure_data/ and write polished figures under figures/custom/.Required (Python):
scanpyOptional (Python):
infercnvpy for expression-based CNV inferenceOptional (R environment / subprocess):
numbatSingleCellExperimentzellkonverterTrigger conditions:
Chaining partners:
spatial-preprocess for matrix preparation and counts preservationspatial-annotate for choosing diploid reference categoriesspatial-genes for linking CNV programs to spatially variable genesrun_numbat