菜单
Load when extracting gene programs (NMF / cNMF factorisation) and per-cell program usage scores from a non-negative scRNA AnnData. Skip when ranking marker genes per cluster (use sc-markers) or for inferring TF → target regulons (use sc-grn).
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 sc-gene-programs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。
Operate MedSAM2 for promptable segmentation of 3D medical images and medical videos, including CT lesion propagation, MRI volumes, RECIST-guided prompts, efficient CPU-oriented variants, training, and 3D Slicer integration. Use when generating or validating volumetric masks from sparse prompts or propagating masks through image slices or video frames.
Build reproducible healthcare imaging pipelines with Project MONAI for DICOM, NIfTI, pathology, and multidimensional imaging tasks including preprocessing, augmentation, training, sliding-window inference, evaluation, model bundles, labeling, and deployment. Use when implementing medical image classification, segmentation, registration, detection, generative, or foundation-model workflows in PyTorch.
Operate Google TxGemma prediction and chat models for therapeutic property prediction across small molecules, proteins, nucleic acids, diseases, targets, and cell lines. Use when formatting Therapeutics Data Commons tasks, choosing TxGemma model size or variant, running local or Model Garden inference, fine-tuning on private therapeutic data, or evaluating TxGemma in drug-discovery workflows.
基于 SOC 职业分类
正在显示 SKILL.md
| name | sc-gene-programs |
| description | Load when extracting gene programs (NMF / cNMF factorisation) and per-cell program usage scores from a non-negative scRNA AnnData. Skip when ranking marker genes per cluster (use sc-markers) or for inferring TF → target regulons (use sc-grn). |
| version | 0.2.0 |
| author | OmicsClaw |
| license | MIT |
| tags | ["singlecell","scrna","gene-programs","nmf","cnmf","factorisation"] |
| requires | ["anndata","scanpy","numpy","pandas","scikit-learn"] |
The user has a non-negative scRNA AnnData (raw counts or log-normalised expression) and wants to decompose it into K gene programs (latent factors) plus a per-cell usage matrix. Two methods:
cnmf (default) — consensus NMF (multiple runs + clustering of
factors) for stable programs. Auto-falls back to nmf if the cnmf
package isn't installed.nmf — sklearn NMF, single run.Output: tables/program_usage.csv (cells × K), tables/program_weights.csv
(genes × K), tables/top_program_genes.csv (top-N genes per program).
For per-cluster marker discovery use sc-markers; for TF → target
regulons use sc-grn; for per-cell pathway scores against curated
gene sets use sc-pathway-scoring.
| Input | Format | Required |
|---|---|---|
| Non-negative AnnData | .h5ad (raw counts in layers["counts"] or log-normalised in .X) | yes (unless --demo) |
| Output | Path | Notes |
|---|---|---|
| Annotated AnnData | processed.h5ad | adds obsm["X_gene_programs"] (cells × K usage matrix; sc_gene_programs.py:469). The program_usage name only applies to the tables/program_usage.csv file, NOT to the obsm key. |
| Per-cell usage | tables/program_usage.csv | cells × K matrix |
| Gene weights | tables/program_weights.csv | genes × K matrix |
| Top genes | tables/top_program_genes.csv | top---top-genes per program |
| Figures | figures/mean_program_usage.png, figures/program_correlation.png | always |
| Report | report.md + result.json | always |
import cnmf; if it fails, silently switch --method to nmf.--input) or build a demo.--layer (auto-prefer layers["counts"] for cnmf when --layer is unset); reject negative values; warn if n_genes < 50 or running NMF on raw counts without --layer counts.--n-iter runs) or sklearn NMF (single run, --seed).processed.h5ad, report.md, result.json.cnmf silently auto-falls back to nmf if cnmf is not installed. sc_gene_programs.py:407-409 catches ImportError from import cnmf, logs a warning, sets args.method = "nmf" (so summary["method"] at :527 already reflects the post-fallback value). summary["backend"] at :528 records the same. Inspect either before quoting "we used cNMF".sc_gene_programs.py:132 raises SystemExit("NMF/cNMF requires non-negative input, but the data matrix contains negative values. This usually means the data has been z-score scaled. ...") with a multi-option fix message. Most common cause: feeding a sc.pp.scale-d AnnData where .X is mean-centred. Pass --layer counts or re-run sc-preprocessing without scaling.cnmf auto-prefers layers["counts"]; nmf doesn't. sc_gene_programs.py:111-112 switches to layers["counts"] for cnmf if --layer is unset and the layer exists. nmf without --layer uses .X directly. If your raw counts live elsewhere, pass --layer <name> explicitly to avoid silent fallback to .X.--layer value also SystemExits. sc_gene_programs.py:118 raises SystemExit("Layer '<name>' not found in adata.layers. Available layers: <list>. ...") when an explicit --layer doesn't resolve. Wrappers expecting ValueError need to catch SystemExit.--input mandatory unless --demo. sc_gene_programs.py:418 raises SystemExit("Provide --input or use --demo").--n-programs, sc_gene_programs.py:533-534 records summary["degenerate_output"] = True and lists degenerate_issues — but the script returns 0. Always inspect (line 529, the count) before chaining downstream.# Demo (cNMF on synthetic data, falls back to NMF if cnmf missing)
python omicsclaw.py run sc-gene-programs --demo --output /tmp/sc_gp_demo
# cNMF with 8 programs on raw counts
python omicsclaw.py run sc-gene-programs \
--input clustered.h5ad --output results/ \
--method cnmf --n-programs 8 --n-iter 200 --layer counts
# NMF on log-normalised .X (faster, less stable)
python omicsclaw.py run sc-gene-programs \
--input normalized.h5ad --output results/ \
--method nmf --n-programs 10 --top-genes 50
references/parameters.md — every CLI flag, NMF / cNMF tunablesreferences/methodology.md — when consensus NMF wins; layer-selection guidereferences/output_contract.md — obsm["X_gene_programs"] / tables/program_*.csv schemassc-preprocessing (upstream — produces a non-negative .X or layers["counts"]), sc-markers (parallel — cluster markers, NOT latent factors), sc-pathway-scoring (parallel — supervised program scoring against curated gene sets), sc-grn (parallel — TF → target regulons; complementary to gene programs)result.json["n_programs"]