Batch Correction in scRNA-seq: How to Preserve Biological Differences

[sunsunb]

I would greatly appreciate your opinions, comments, and advice on the following:

We have scRNA-seq data from two experimental conditions: CTRL and STIM.

Under CTRL, we have two samples (each sample is its own batch).

Under STIM, we have three samples (again, one sample = one batch).

I integrated all these samples into a single Seurat object. However, after applying Harmony for batch correction, according to umap plot, I noticed that the biological differences between the CTRL and STIM groups appear to be removed, which suggests possible overcorrection.

Below is the code I used.

merged_seurat <- merge(singlet_list[[1]], y = singlet_list[2:length(singlet_list)])
merged_seurat <- SCTransform(merged_seurat, vars.to.regress = "percent_mt", verbose = TRUE)

DefaultAssay(merged_seurat) <- "SCT"
df <- FindVariableFeatures(merged_seurat, selection.method = "vst", nfeatures = 2000)
df <- RunPCA(df, verbose = FALSE)
df <- RunUMAP(df, dims = 1:30, verbose = FALSE)
df <- FindNeighbors(df, dims = 1:30, verbose = FALSE)
df <- FindClusters(df, resolution = 0.5, verbose = FALSE)


df.harmony <- RunHarmony(df,group.by.vars = c("orig.ident", "condition"), plot_convergence = TRUE,assay.use="SCT")

df <- RunUMAP(df, dims = 1:30, verbose = FALSE)
df.harmony <- RunUMAP(df.harmony, dims = 1:30, reduction = 'harmony', verbose = FALSE)
# Visualization
before <- DimPlot(df, reduction = "umap", group.by = "orig.ident")
after <- DimPlot(df.harmony, reduction = "umap", group.by = "orig.ident")
by_sample <- DimPlot(df.harmony, reduction = "umap", split.by = "orig.ident")
before|after

before <- DimPlot(df, reduction = "umap", group.by = "condition")
after <- DimPlot(df.harmony, reduction = "umap", group.by = "condition")
by_sample <- DimPlot(df.harmony, reduction = "umap", split.by = "condition")
before|after

Could anyone kindly suggest how I can properly correct for batch effects without eliminating meaningful group differences?
Any suggestions or best practices would be greatly appreciated

[GoulsonWang]

I guess that maybe you can only correct within CTRL and STIM, and don't try to see them as one object