Which metric is best to represent cell-level counts/features in Seurat: raw/log-normalized (nCount_RNA/nFeature_RNA) or SCTransform (nCount_SCT/nFeature_SCT)?

Question

[solo.albif]

Hi all,

I am analyzing CosMx spatial transcriptomics data using Seurat. My Seurat object includes both the RNA assay (nCount_RNA and nFeature_RNA) and the SCTransform assay (nCount_SCT and nFeature_SCT).

I want to visualize and summarize cell-level total counts and detected features for quality control and interpretation. I am unsure which metric better represents the data:

The raw or log-normalized counts/features from the RNA assay?

Or the counts/features associated with the SCTransform assay?

Which approach is recommended or more meaningful for downstream analysis and visualization (e.g., with FeaturePlot)?

Thanks in advance for your guidance!

Answer 1

[ATpoint]

Usually for these basic QCs one uses raw counts as one wants to ensure remotely comparable sequencing depth and detected genes per cell. Normalization plays no role here as "detected genes" is simply the sum of genes with non-zero counts, and then per-cell depth is the sum of the raw read / UMI counts.