data normalization strategies

[grst]

We have three kinds of data

• raw, unnormalized, non-log transformed (UMI) counts.
  • should be kept somewhere until the very end, because in principle everything can be done with it.
  • possibly store in adata.obsm.
• per-cell normalized, log-transformed counts
  • useful for visualization of gene expression. Export for single cell explorer. Should be stored in adata.raw.
• filtered, normalized, scaled data
  • useful for clustering, PCA, UMAP, and most trajectory inference tools
  • keep in main adata.X.