菜单
Multi-sample integration and batch correction for spatial transcriptomics data using Harmony, BBKNN, or Scanorama.
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-integrate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。
正在显示 SKILL.md
| name | spatial-integrate |
| description | Multi-sample integration and batch correction for spatial transcriptomics data using Harmony, BBKNN, or Scanorama. |
| version | 0.5.0 |
| author | OmicsClaw |
| license | MIT |
| tags | ["spatial","integration","batch correction","Harmony","BBKNN","Scanorama"] |
| metadata | {"omicsclaw":{"domain":"spatial","allowed_extra_flags":["--batch-key","--method","--harmony-theta","--harmony-lambda","--harmony-max-iter","--bbknn-neighbors-within-batch","--bbknn-n-pcs","--bbknn-trim","--scanorama-knn","--scanorama-sigma","--scanorama-alpha","--scanorama-batch-size"],"param_hints":{"harmony":"[Truncated]","bbknn":"[Truncated]","scanorama":"[Truncated]"},"legacy_aliases":["integrate"],"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]"],"trigger_keywords":["multi-sample integration","batch correction","Harmony","BBKNN","Scanorama","merge samples"]}} |
You are Spatial Integrate, a specialised OmicsClaw agent for multi-sample integration and batch effect correction. Your role is to align multiple spatial transcriptomics samples into a shared embedding while preserving biological variation.
figure_data/*.csv and figure_data/manifest.json for downstream customization.report.md, result.json, processed.h5ad, figures, tables, and the reproducibility bundle.theta, lambda, and iteration controlsneighbors_within_batch, n_pcs, trim)figures/figure_data/ contains integration plotting contracts for downstream customizationr_visualization/ templates consume figure_data/ without recomputing integration resultsOmicsClaw treats spatial-integrate visualization as a two-layer system:
figure_data/ and does not recompute integration results.The standard gallery is declared as a recipe instead of hard-coded if/else
plot branches. Current gallery roles include:
overview: before/after UMAP by batchdiagnostic: UMAP by cluster and per-batch highlight panelssupporting: batch-size summaryuncertainty: batch-mixing entropy bars plus local entropy visualizations| Format | Extension | Required Fields | Example |
|---|---|---|---|
| AnnData (multi-sample, preprocessed) | .h5ad | X (log-norm), obsm["X_pca"], obs[batch_key] | merged_samples.h5ad |
After spatial-preprocess, the AnnData contains multiple representations.
All three integration methods require PCA embeddings from upstream
preprocessing (log-normalized expression → HVG selection → PCA):
adata.X # log-normalized expression
adata.obsm["X_pca"] # PCA embeddings (from scaled HVGs) — used by all methods
adata.obs[batch_key] # batch/sample labels (e.g., "batch", "sample_id")
adata.obsp["connectivities"] # k-NN neighbor graph
Each method consumes the preprocessed data differently:
| Method | Primary input consumed | What it operates on | Output embedding |
|---|---|---|---|
| Harmony | obsm["X_pca"] (PCA embeddings) | Iteratively adjusts PCs to remove batch effects | X_pca_harmony |
| BBKNN | obsm["X_pca"] (PCA embeddings) | Builds batch-balanced k-NN graph from PCA space | Modified connectivities graph |
| Scanorama | obsm["X_pca"] (via scanpy wrapper) | MNN stitching on PCA basis; original method works on expression matrices | X_scanorama |
Harmony / BBKNN: Both operate directly on PCA embeddings. The upstream pipeline must have computed
normalize_total→log1p→ HVG → PCA before running integration. Neither method touchesadata.X.
Scanorama: The original Scanorama algorithm (Hie et al., 2019) was designed to stitch preprocessed expression matrices via mutual nearest neighbors. The scanpy wrapper
sc.external.pp.scanorama_integrateapplies the same algorithm to the PCA basis for computational efficiency. For expression-level integration closer to the original paper, users can callscanorama.integrate()directly on per-batch HVG matrices.Current OmicsClaw behavior: if batches are interleaved in
adata.obs[batch_key], OmicsClaw temporarily sorts cells by batch before calling the Scanpy wrapper and restores the corrected embedding to the original cell order afterward.
# Standard usage (Harmony, default)
oc run spatial-integration \
--input <merged.h5ad> --output <dir> --batch-key sample_id
# Harmony tuning
oc run spatial-integration \
--input <data.h5ad> --method harmony --batch-key batch \
--harmony-theta 3.0 --harmony-lambda -1 --harmony-max-iter 15 --output <dir>
# BBKNN tuning
oc run spatial-integration \
--input <data.h5ad> --method bbknn --batch-key sample_id \
--bbknn-neighbors-within-batch 5 --bbknn-n-pcs 30 --bbknn-trim 60 --output <dir>
# Scanorama tuning
oc run spatial-integration \
--input <data.h5ad> --method scanorama --batch-key batch \
--scanorama-knn 30 --scanorama-sigma 12 --scanorama-alpha 0.15 \
--scanorama-batch-size 4000 --output <dir>
# Demo mode
oc run spatial-integration --demo
# Direct script entrypoint
python skills/spatial/spatial-integrate/spatial_integrate.py \
--input <merged.h5ad> --method harmony --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.obs with ≥2 batchesX_pca is present (from upstream preprocessing: log-normalized → HVG → PCA)adata.obsm["X_pca"] + adata.obs[batch_key]adata.obsm["X_pca_harmony"]--harmony-theta: diversity penalty--harmony-lambda: ridge penalty (-1 enables Harmony auto-lambda estimation)--harmony-max-iter: maximum Harmony outer iterationsadata.obsm["X_pca"] + adata.obs[batch_key]adata.obsp["connectivities"] and adata.obsp["distances"]--bbknn-neighbors-within-batch: neighbors per batch--bbknn-n-pcs: PCA dimensionality used to build the graph--bbknn-trim: optional graph trimmingadata.obsm["X_pca"] (via scanpy wrapper) + adata.obs[batch_key]adata.obsm["X_scanorama"]--scanorama-knn: neighbors used for batch matching--scanorama-sigma: Gaussian kernel width--scanorama-alpha: minimum alignment score cutoff--scanorama-batch-size: incremental alignment batch sizeShared key parameters:
--batch-key: obs column identifying batches (default: batch)--method: harmony, bbknn, or scanorama (default: harmony)output_directory/
├── README.md
├── report.md
├── result.json
├── processed.h5ad
├── figures/
│ ├── umap_before_by_batch.png
│ ├── umap_by_batch.png
│ ├── umap_by_cluster.png
│ ├── batch_highlight.png
│ ├── batch_sizes.png
│ ├── batch_mixing.png
│ ├── batch_entropy_after_umap.png
│ ├── batch_entropy_distribution.png
│ └── manifest.json
├── figure_data/
│ ├── batch_sizes.csv
│ ├── integration_metrics.csv
│ ├── umap_before_points.csv
│ ├── umap_after_points.csv
│ ├── corrected_embedding_points.csv # for embedding-returning methods
│ └── manifest.json
├── tables/
│ ├── integration_metrics.csv
│ ├── batch_sizes.csv
│ └── integration_observations.csv
└── 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 usually produces the
skill-native outputs plus reproducibility/commands.sh.
The bundled repo-side R helpers live under:
skills/spatial/spatial-integrate/r_visualization/
├── README.md
└── integration_publication_template.R
Required:
scanpy >= 1.9Optional:
harmonypy — Harmony integration (recommended, lightweight)bbknn — batch-balanced KNNscanorama — panoramic stitchingTrigger conditions:
Chaining partners:
spatial-preprocess — QC before integrationspatial-annotate — Label transfer post-integration基于 SOC 职业分类
