Is this the correct Seurat v5 workflow (SCT + Integration (CCA-RPCA-Harmony))?

[OBIYOKA]

I am analyzing a scRNA-seq dataset with two conditions Control and Disease. I am specifically looking for subset that appears in the disease condition. I am concerned that standard integration might "over-correct" and blend this distinct population into the control clusters.

I have set up a Seurat v5 workflow that: Splits layers (to handle V5 requirements). Runs SCTransform (v2) for normalization. Benchmarks CCA, RPCA, and Harmony side by side. Joins layers and log-normalizes the RNA assay at the end for downstream analysis.

My Questions are: Is this order of operations correct for v5? Specifically, the split - SCT - Integrate - Join - Normalize sequence? For downstream analysis (finding markers for this subset), is it standard practice to switch back to the "RNA" assay (LogNormalized) as I have done in step 7? Or should I be using the SCT residuals?

Here is the minimal code I am using. Any feedback on the workflow is appreciated.

1. load 10x

raw_con <- Read10X("path/to/con_matrix")

raw_dis <- Read10X("path/to/dis_matrix")

obj_con <- CreateSeuratObject(counts = raw_con, project = "con")

obj_dis <- CreateSeuratObject(counts = raw_dis, project = "dis")

obj_con$sample <- "con"

obj_dis$sample <- "dis"

Merge into one object 'seu'

seu <- merge(obj_con, y = obj_dis)

seu$sample <- seu$orig.ident

2. QC & Pre-processing

seu <- subset(seu, subset = nFeature_RNA > 200 & nFeature_RNA < 3000 & mt< 10)

3. Split Layers (Critical for V5 integration)

seu[["RNA"]] <- split(seu[["RNA"]], f = seu$sample)

4. SCTransform (Prepares 'SCT' assay for integration)

seu <- SCTransform(

seu,

assay = "RNA",

vst.flavor = "v2",

return.only.var.genes = FALSE,

verbose = FALSE

)

seu <- RunPCA(seu, npcs = 30, verbose = FALSE)

Benchmark Integrations (CCA vs RPCA vs Harmony)

All integrations use the 'SCT' assay but save to different reductions

seu <- IntegrateLayers(

object = seu, method = CCAIntegration,

orig.reduction = "pca", new.reduction = "integrated.cca",

normalization.method = "SCT", verbose = FALSE

)

seu <- IntegrateLayers(

object = seu, method = RPCAIntegration,

orig.reduction = "pca", new.reduction = "integrated.rpca",

normalization.method = "SCT", verbose = FALSE

)

seu <- IntegrateLayers(

object = seu, method = HarmonyIntegration,

orig.reduction = "pca", new.reduction = "integrated.harmony",

normalization.method = "SCT", verbose = FALSE

)

6. Clustering & Visualization

methods <- c("integrated.cca", "integrated.rpca", "integrated.harmony")

for (red in methods) {

seu <- FindNeighbors(seu, reduction = red, dims = 1:30, verbose = FALSE)

seu <- FindClusters(seu, resolution = 0.5, cluster= paste0(red, "_clusters"), verbose = FALSE)

seu <- RunUMAP(seu, reduction = red, dims = 1:30, reduction= paste0("umap.", red), verbose = FALSE)

}

7. Post-Integration Cleanup

Re-join RNA layers for DE analysis and Standard Normalization

seu[["RNA"]] <- JoinLayers(seu[["RNA"]])

seu <- NormalizeData(seu, assay = "RNA", normalization.method = "LogNormalize")

seu <- PrepSCTFindMarkers(seu) # Update SCT models for downstream DE
8. Plot Comparison