Checklist-style reference for OmicVerse downstream tutorials covering AUCell scoring, metacell DEG, and related exports.
The single-cell-downstream-analysis skill provides a structured, checklist-style reference for executing downstream analyses in OmicVerse single-cell workflows. It consolidates multiple tutorial notebooks into a practical guide that outlines prerequisites, core API entry points, validation steps, and resource considerations. The skill focuses on pathway activity scoring with AUCell, differential expression analysis using bulk-style metacell aggregation, and cell-type–specific and composition-based differential testing. By organizing these workflows into clear execution stages, it helps users avoid common setup mistakes, ensure correct data structures (such as properly prepared AnnData objects), and validate results through recommended visualization and diagnostic checkpoints.
Core capabilities include AUCell-based pathway scoring for individual or multiple gene sets, enrichment scoring across full pathway libraries, and downstream visualization using embedding plots and ranked pathway dotplots. For differential expression, the skill covers both bulk-style meta cell analysis using pyDEG and single-cell–level DEG approaches with methods such as Wilcoxon, t-test, and memento-de. It also supports differential composition testing through approaches like scCODA and Milo, including required preprocessing steps such as batch correction, neighbor graph construction, and UMAP embedding. Each module documents computational considerations, including CPU parallelism, GPU acceleration, memory planning for dense matrices, and optional export steps such as writing AnnData objects or saving figures.
This skill is intended for bioinformaticians, computational biologists, and data scientists working with single-cell RNA-seq data in Python using OmicVerse and Scanpy-compatible data structures. It is particularly useful for researchers performing pathway activity analysis, identifying cluster-enriched gene programs, detecting differential expression across conditions or cell types, and assessing differential abundance. The structured checklist format makes it suitable both for reproducible research workflows and for training users who need a concise but technically grounded reference for downstream single-cell analysis tasks.
An AnnData object is required, including clustering or embedding information such as adata.obsm['X_umap']. For DEG workflows, raw counts should be retained in adata.raw before highly variable gene filtering.
Pathway activities are computed using AUCell-based functions such as ov.single.geneset_aucell, ov.single.pathway_aucell, and ov.single.pathway_aucell_enrichment. Scores range from 0 to 1 and can be visualized on embeddings or ranked across clusters.
GPU usage is optional but recommended for certain steps. Metacell construction can use GPU acceleration, and batch correction methods such as Harmony or scVI require GPU memory when enabled. Most steps can fall back to CPU execution.
Library-wide AUCell scoring and methods like memento or Milo may require multiple CPUs and significant memory, especially when working with dense transposed matrices or high k values in graph-based analyses. Users should allocate sufficient compute time and memory.
Yes. AUCell score matrices can be saved using write_h5ad, and metacell embeddings or other visual outputs can be saved as matplotlib figures with adjustable legend settings for publication-ready formatting.
Quick Setup:
.claude/skills/Repository
Starlitnightly/omicverse