Unable to integrate separately SCTransformed scRNAseq Seurat Objects

Question

[awsk]

Hello and thank you for anyone who can offer some help.

I have five separate scRNAseq experiments I am trying to merge and integrate. As part of my quality control pipeline I am performing ambient RNA removal with DecontX, manual thresholding based on %mt and counts, followed by doublet removal with DoubletFinder, on each sample individually.

As part of the DoubletFinder processing, the objects are SCTransformed. From my reading of the vignettes I understand this to be supported, but when I merge and try to integrate the sets I run into many errors.

First, I merge the separate seurat objects..

dX.seur <- merge(x = dX.seur[[1]],y = dX.seur[2:length(dX.seur)], add.cell.ids = names(dX.seur), project = "nb_merged", merge.data = TRUE)

Then I split into layers...

dX.seur[["SCT"]] <- split(dX.seur[["SCT"]], f = dX.seur$orig.ident)

And here is where I run into many issues. I have no variable features or reductions after the merge, so I've tried to either set the variable features equal to the scaledata features before the split, or to run FindVariableFeatures again on the split object. Either way when I try to run the PCA, it fails to consider the multiple layers...

In LayerData.Assay5(object = object, layer = layer) : multiple layers are identified by scale.data.bard1_count scale.data.brca2july2024_count scale.data.brca2older_count scale.data.ddr_wt_count scale.data.palb2_count only the first layer is used

Anyways, how am I actually supposed to do this?

[Bastien Hervé]

I think the problem lays when you are trying to merge objects that have been SCTransformed after DoubletFinder, SCTransform will generate scale.data matrix for each of your experiments. When you are trying to merge your objects it does not make sense to merge scale.data because each experiment scaling is different, therefore Seurat create a different scale.data matrix for each experiment when you are tyring to merge your objects (that is what you see in the error message : scale.data.bard1_count scale.data.brca2july2024_count scale.data.brca2older_count scale.data.ddr_wt_count scale.data.palb2_count)

After DoubletFinder, flag all your cells that are doublets for your, keep the prediction score, and reload all your datasets, merge them, split them by layers and start from there.

[awsk]

After doublet removal I should go back to the raw RNA assay, split, and re-run the SCT? Thank you for replying!

[Bastien Hervé]

yes, that is my guess, because if you merge scale.data matrices it does not make sense to me.

[awsk]

I agree and I think what you are proposing makes sense... It's just frustrating since in their workflow vignettes they make it clear you can merge and work with independently SCTransformed data. I'm not sure if they're just tweaking the scaledata in these cases... Anyways, thanks again.