omicverse-single-cell-preprocessing

What problem does it solve?

Convert raw or partially processed single-cell AnnData into a validated, analysis-ready object by automating QC, normalization, HVG selection, scaling, dimensionality reduction, neighbor graph construction, clustering, and marker discovery so users avoid repetitive notebook rewrites and fragile ad-hoc steps.

Core Features & Use Cases

• End-to-end preprocessing spine covering QC, multiple preprocessing modes, HVG selection, scaling, PCA, neighbor graph construction, and alternative embeddings (UMAP, t-SNE, MDE).
• Clustering and marker discovery with Leiden clustering and marker extraction branches (cosg for raw counts, wilcoxon and other statistical tests for log-normalized data), plus marker plotting helpers.
• Validation and reproducibility checks for required obs/var/obsm/uns keys, preservation of raw counts when needed, explicit mode and method selection, and smoke-testable acceptance criteria for CI.
• Use Case: Run a small synthetic smoke test to verify CPU preprocessing, then run the full pipeline on experiment data to obtain clusters and top markers for annotation.

Quick Start

Run the OmicVerse preprocessing pipeline on an AnnData object to produce normalized counts, PCA and UMAP embeddings, a neighbor graph, Leiden cluster labels, and optional marker tables.