Single-cell clustering and batch correction with omicverse

Batch correction removes technical variations across datasets using methods like Harmony, ComBat, Scanorama, scVI, or CellANOVA. This Skill automates preprocessing, integration, and evaluation to harmonize multi-batch single-cell data while preserving biological signal in AnnData objects.

Single-cell clustering supports multiple algorithms: Leiden, Louvain, scICE, Gaussian mixture models, topic modeling, and cNMF. This Skill applies and evaluates each method on preprocessed data, enabling comparison across clustering strategies to identify robust cell populations.

Yes. This Skill integrates omicverse with scanpy to filter low-quality cells, normalize, scale, and reduce dimensionality via PCA. It builds neighborhood graphs and enforces reproducible workflows through dependency checks and checkpointing for resumable analysis.

This Skill evaluates batch correction using scib-metrics to assess integration quality across cohorts. It quantifies whether corrected batches preserve biological structure while removing technical effects, ensuring robust insights from multi-batch single-cell experiments.

Single-cell clustering operates on omicverse-enabled AnnData objects containing expression matrices. Preprocessing steps—filtering, normalization, scaling—prepare raw count data for neighborhood graph construction and clustering method application.

Different clustering algorithms capture distinct cell populations and topologies. Applying Leiden, scICE, GMM, and cNMF reveals consensus clusters and identifies algorithm-specific insights, improving confidence in cell type annotation and biological interpretation.