scRNA-seq help needed

[Fantastic Mr. Fox]

I'm working on scRNA-seq data from the same tissue type. I have four groups: WT untreated, WT treated, Test untreated, Test treated. And I'm planning to compare across treatment and genotype.

Question is, would you load all samples into one Seurat object with multifactor integration, or keep them separate to avoid messing with clustering and DEGs? I want clustering by cell type, but still need reliable DE results across conditions.

 

[Josef_10]

I'd strongly recommend integrating all your samples into a single Seurat object for your four groups. The multifactor integration in Seurat is specifically designed to correct for technical differences between your samples while preserving the true biological variation. This ensures that your clustering reflects actual cell types across all conditions, rather than being skewed by which sample a cell came from. If you keep them separate, you'll have a much harder time comparing and aligning cell types, making reliable differential expression analysis a nightmare. A key point is to use the integrated data for clustering, but perform your differential expression analysis on the original, uncorrected count data. This approach will give you both unified, accurate clustering and DE results based on your real expression counts, which makes them less prone to being artifacts of the integration process.

 

[Akeem]

If integration risks washing out subtle condition effects, especially in DEGs, isn't stuffing everything into one Seurat object kinda like mixing paint colors and expecting to still see the original shades?