菜单
Find genes with spatially variable expression patterns using Moran's I, SpatialDE, SPARK-X, or FlashS. Identifies genes whose expression is non-randomly distributed across tissue coordinates.
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-genes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-genes |
| description | Find genes with spatially variable expression patterns using Moran's I, SpatialDE, SPARK-X, or FlashS. Identifies genes whose expression is non-randomly distributed across tissue coordinates. |
| version | 0.4.0 |
| author | OmicsClaw |
| license | MIT |
| tags | ["spatial","SVG","spatially-variable-genes","morans","spatialde","sparkx","flashs"] |
| metadata | {"omicsclaw":{"domain":"spatial","allowed_extra_flags":["--flashs-bandwidth","--flashs-n-rand-features","--fdr-threshold","--method","--morans-coord-type","--morans-corr-method","--morans-n-neighs","--morans-n-perms","--n-top-genes","--sparkx-max-genes","--sparkx-num-cores","--sparkx-option","--spatialde-aeh-lengthscale","--spatialde-aeh-patterns","--spatialde-min-counts","--spatialde-no-aeh"],"param_hints":{"morans":"[Truncated]","spatialde":"[Truncated]","sparkx":"[Truncated]","flashs":"[Truncated]"},"legacy_aliases":["genes"],"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":["spatially variable gene","spatial gene","SVG","SpatialDE","SPARK-X","spatial pattern","Moran","spatial autocorrelation"]}} |
You are Spatial Genes, the spatially variable gene (SVG) discovery skill for OmicsClaw. Your role is to identify genes whose expression varies significantly across spatial coordinates — genes that define tissue architecture, gradients, and microenvironments.
figures/figure_data/ exports ranked genes, significant SVGs, observation-level metric tables, point tables, and a manifest for downstream plottingadata.obs for stable downstream reuser_visualization/ templates consume figure_data/ without recomputing science| Format | Extension | Required Fields | Example |
|---|---|---|---|
| AnnData (preprocessed) | .h5ad | X (normalised), layers["counts"] (raw), obsm["spatial"] |
processed.h5ad |
| Demo | n/a | --demo flag | Runs spatial-preprocess demo first |
Different SVG methods have different statistical assumptions. The skill automatically selects the correct input matrix per method:
| Method | Input Matrix | Rationale |
|---|---|---|
morans | adata.X (log-normalized) | Squidpy spatial_autocorr computes autocorrelation on continuous expression values; not a count-distribution model |
spatialde | adata.layers["counts"] (raw) | NaiveDE.stabilize() applies its own variance-stabilizing transform; feeding already-normalized data breaks assumptions |
sparkx | adata.layers["counts"] (raw) | SPARK-X directly inputs the raw count matrix for kernel-based testing |
flashs | adata.layers["counts"] (raw) | FlashS exploits sparsity and count structure (binary presence + rank + count) |
Data layout requirement: Preprocessing must store raw counts before normalization:
adata.layers["counts"] = adata.X.copy() # before normalize_total + log1p
adata.X = lognorm_expr # after normalize_total + log1p
If layers["counts"] is missing, the skill will fall back to adata.raw (if available) or adata.X with a warning.
figure_data/*.csv and figure_data/manifest.json for downstream customization.report.md, result.json, processed.h5ad, tables, and the reproducibility bundle including optional R visualization entrypoints.OmicsClaw treats spatial-genes visualization as a two-layer system:
figure_data/ and does not recompute SVG statistics.The standard gallery is declared as a recipe instead of hard-coded if/else
plot calls. Current gallery roles include:
overview: top SVG feature maps on tissue coordinates and UMAPdiagnostic: method score vs significance, plus Moran ranking when availablesupporting: top-gene score summary panel plus observation-level SVG burden context from persisted gallery metricsuncertainty: significance distribution across tested genesThis separation keeps the OmicsClaw output stable while still allowing later R-side beautification and method-specific extensions.
# Standard usage (Moran's I, default)
oc run spatial-genes \
--input <processed.h5ad> --output <report_dir>
# Moran's I tuning
oc run spatial-genes \
--input <processed.h5ad> --method morans \
--morans-coord-type auto --morans-n-neighs 8 --morans-n-perms 500 \
--morans-corr-method fdr_bh --n-top-genes 30 --fdr-threshold 0.01 --output <dir>
# SpatialDE + AEH pattern grouping
oc run spatial-genes \
--input <processed.h5ad> --method spatialde \
--spatialde-min-counts 5 --spatialde-aeh-patterns 6 --output <dir>
# SPARK-X method (requires R + SPARK package)
oc run spatial-genes \
--input <processed.h5ad> --method sparkx \
--sparkx-option mixture --sparkx-num-cores 4 --sparkx-max-genes 3000 --output <dir>
# FlashS method (fast on large data)
oc run spatial-genes \
--input <processed.h5ad> --method flashs \
--flashs-n-rand-features 1000 --flashs-bandwidth 50.0 --output <dir>
# Demo mode
oc run spatial-genes --demo --output /tmp/svg_demo
# Direct script entrypoint
python skills/spatial/spatial-genes/spatial_genes.py \
--input <processed.h5ad> --method morans --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)squidpy.gr.spatial_neighbors(...) builds a spatial graph from obsm["spatial"]; OmicsClaw can now pass coord_type through instead of always forcing genericsquidpy.gr.spatial_autocorr(adata, mode="moran", n_perms=..., corr_method=..., n_jobs=1) computes Moran's I for every geneMoran's I > 0 and the corrected significance column from corr_method (for example pval_norm_fdr_bh) below fdr_threshold; if Squidpy does not expose a corrected column, OmicsClaw falls back to raw pval_normKey parameters:
| Parameter | Default | Description |
|---|---|---|
--method | morans | morans, spatialde, sparkx, or flashs |
--n-top-genes | 20 | Number of top SVGs to report |
--fdr-threshold | 0.05 | FDR-corrected p-value cutoff |
--morans-coord-type | auto | Neighbor-graph layout hint: auto, generic, or grid |
--morans-n-neighs | 6 | Number of spatial neighbors for generic coordinates |
--morans-n-perms | 100 | Number of permutations for empirical p-values; 0 disables permutations |
--morans-corr-method | fdr_bh | Multiple-testing correction method |
adata.layers["counts"] (raw counts)SpatialDE + NaiveDE packagesSpatialDE.spatial_patterns() to group significant genes into AEH patterns using pattern count C and lengthscale lqvalCore tuning flags:
--spatialde-min-counts: Per-gene count filter before SpatialDE fitting.--spatialde-no-aeh: Skip AEH pattern grouping and only report per-gene SpatialDE statistics.--spatialde-aeh-patterns: Override AEH pattern count C.--spatialde-aeh-lengthscale: Override AEH lengthscale l.adata.layers["counts"] (raw counts)spark.sparkx(counts, coords, numCores=..., option=...)Core tuning flags:
--sparkx-option: Passed through to the R implementation; the official example uses mixture.--sparkx-num-cores: Passed through as numCores.--sparkx-max-genes: OmicsClaw wrapper cap for very large matrices before the R call.adata.layers["counts"] (raw counts)qval alongside the wrapper scoreCore tuning flags:
--flashs-n-rand-features: Wrapper sketch size controlling approximation fidelity vs runtime.--flashs-bandwidth: Wrapper kernel bandwidth override; omitted means data-adaptive estimation.output_dir/
├── README.md
├── report.md
├── result.json
├── processed.h5ad
├── figures/
│ ├── top_svg_spatial.png
│ ├── top_svg_umap.png
│ ├── top_svg_scores.png
│ ├── svg_score_vs_significance.png
│ ├── svg_significance_distribution.png
│ ├── moran_ranking.png # Moran's I only
│ └── manifest.json
├── figure_data/
│ ├── svg_results.csv
│ ├── top_svg_scores.csv
│ ├── significant_svgs.csv
│ ├── svg_observation_metrics.csv
│ ├── top_svg_spatial_points.csv
│ ├── top_svg_umap_points.csv # if X_umap is available
│ ├── svg_run_summary.csv
│ └── manifest.json
├── tables/
│ ├── svg_results.csv
│ ├── top_svg_scores.csv
│ ├── significant_svgs.csv
│ └── svg_observation_metrics.csv
└── reproducibility/
├── analysis_notebook.ipynb
├── commands.sh
├── requirements.txt
└── r_visualization.sh
README.md and reproducibility/analysis_notebook.ipynb are generated by the
standard OmicsClaw wrapper. Direct script execution usually produces the
skill-native outputs plus reproducibility/commands.sh.
The bundled repo-side R helpers live under:
skills/spatial/spatial-genes/r_visualization/
├── README.md
└── svg_publication_template.R
Required (Python):
scanpy >= 1.9 — single-cell/spatial analysissquidpy >= 1.2 — spatial autocorrelation and neighbor graphsmatplotlib — plottingnumpy, pandas — numericsOptional (Python):
SpatialDE & NaiveDE — Gaussian process-based SVG detectionflashs — FlashS randomized kernel approximationstatsmodels — multiple-testing correction used by some backends and wrappersOptional (R Environment / Subprocess):
SPARK (R package) — SPARK-X kernel testadata.uns, while observation-level SVG summary columns are persisted in adata.obs.Trigger conditions:
Chaining partners:
spatial-preprocess: Provides the preprocessed h5ad inputspatial-domains: SVGs often overlap with domain-defining genesspatial-de: Compare SVGs with cluster-based DE results