domains / families / science / batch-integration

batch-integration

separate

Integrate multiple scRNA-seq samples or batches with Harmony, scVI/scANVI, Seurat (CCA/RPCA), fastMNN, Scanorama, or BBKNN. Resolves which method to use for the dataset size and design, how strongly to correct, when integration is the wrong move (confounded batch/biology), how to score integration with scIB metrics without gaming them, and why corrected expression must not be used for differential expression. Use when integrating batches or datasets, choosing an integration method, diagnosing over-correction, or judging integration quality.

When to use

• integrating batches or datasets
• choosing an integration method
• diagnosing over-correction
• judging integration quality
• scoring integration with scIB metrics
• resolving method for dataset size and design

When not to use

• differential expression on batch-corrected expression
• confounded batch/biology designs
• per-sample analyses like CNV/tumor-clone inference
• technical replicates that already mix well

Skill page

Source