domains / families / science / clustering

clustering

separate

Dimensionality reduction and graph-based clustering for single-cell RNA-seq with Scanpy (Python) and Seurat (R). Resolves which algorithm to use (Leiden vs Louvain), how many PCs and neighbors to set, how to sweep and validate resolution, when a split is over-clustering, and why post-clustering marker p-values are not valid inference. Use when clustering cells, choosing a clustering resolution, deciding whether two clusters are one population, building a UMAP/tSNE, or judging whether clusters are real.

When to use

• clustering cells
• choosing a clustering resolution
• deciding whether two clusters are one population
• building a UMAP/tSNE
• judging whether clusters are real
• sweeping and validating resolution
• selecting Leiden vs Louvain algorithms
• setting PC and neighbor parameters

When not to use

• using post-clustering marker p-values as valid inference
• clustering on embedding coordinates
• treating resolution as a truth knob

Skill page

Source