Still confused about SCT

[ChenyuZhang97]

Hi Seurat team and community,

After reviewing the vignette and GitHub discussions, I still have some uncertainties regarding the SCT workflow for integrating multiple samples. Would you kindly clarify these points?

• Is this workflow correct? NormalizeData() → CellCycleScoring() → SCTransform(..., vars.to.regress = "CellCycle"). I understand SCTransform replaces normalization, but CellCycleScoring requires normalized data. Should I run normalization only for cell cycle scoring before SCT?
• After IntegrateLayers(..., method = HarmonyIntegration, assay = "SCT"), is JoinLayers() necessary before downstream analysis (e.g., clustering)? It seems unneccessary and does not work on SCT, but I'm quite unsure.
• Do I need to set the parameter (assay = 'SCT' or 'RNA) when running FindAllMarkers()? I thought I should let it be 'RNA'?

Below is my current code. I’d be grateful if you could spot any additional issues:

After loading multiple datasets of many samples:

scRNA_merged <- merge(DataList[[1]], y = DataList[-1], add.cell.ids = RawDataInfo$Sample)
scRNA_merged <- NormalizeData(scRNA_merged)

s.genes <-  read.csv('Mouse_s.genes.csv', row.names = 1)[,1]
g2m.genes <-  read.csv('Mouse_g2m.genes.csv', row.names = 1)[,1]
scRNA_merged <- NormalizeData(scRNA_merged) 
scRNA_merged <- CellCycleScoring(scRNA_merged, s.features = s.genes, g2m.features = g2m.genes, set.ident = T)

scRNA_merged <- SCTransform(object = scRNA_merged, vars.to.regress = c('S.Score', 'G2M.Score', 'percent.mt'), vst.flavor = 'v2',  do.scale = T, do.center = T)

scRNA_merged <- RunPCA(scRNA_merged, verbose=TRUE)
scRNA_harmony <- IntegrateLayers(object = scRNA_merged, method = HarmonyIntegration, orig.reduction = "pca", assay = "SCT", new.reduction = "harmony")

scRNA_harmony <- FindNeighbors(scRNA_harmony, reduction = 'harmony', dims = 1:30)
scRNA_harmony <- FindClusters(scRNA_harmony, resolution = 0.5)
scRNA_harmony <- RunUMAP(scRNA_harmony, reduction = 'harmony',  dims = 1:30)

scRNA_harmony <- PrepSCTFindMarkers(scRNA_harmony,  assay = 'SCT', verbose = T)
AllClusterMarkers<-FindAllMarkers(scRNA_harmony, assay = 'SCT', only.pos = TRUE,  min.pct = 0.1, logfc.threshold = 1)

[Tripfantasy]

Hello, I've been looking around some previous issue threads involving SCT and some of the things you mentioned here. I'm not sure if these are 100% correct, but seem to have been accepted as a standard (at least at the time of their original posting)

• The process of normalizing the data prior to SCT in order to obtain cell cylce scores is valid and others seem to be doing the same.
• In previous versions of SCT it seems people were doing DEA directly on the RNA assay. Perhaps within the last few years, they've made it where you can run on SCT, so long as PrepSCTFindMarkers() is run prior. Which you seem do do above. So all checks out.

Note: One thing I would be sure of with your current workflow is that your active.ident is the variable you want to be your basis for DEA. By default it will probably be seurat_clusters.

[ChenyuZhang97]

Hi, thanks for your help. I became more confident in my code. But I always get DEGs amoung different seurat clusters using these commands, so I'm not very sure about why the active.ident is the variable to be the basis for DEA. Did I miss anything?