Integrate several multiome objects

Question

[Sky]

I have scRNA-seq and scATAC-seq for two conditions (a control and stimulus). I have been trying to find a good tutorial but all I can find are tutorials that deal with one condition (ex: https://satijalab.org/seurat/articles/seurat5_atacseq_integration_vignette) . The scRNA-seq and scATAC-sequencing was performed in the same cell. My question is about the pipeline.

1. Do I process all of the control first (process the scRNA-seq object and integrate it with the scATAC-seq object), then process all of the stimulus, and finally integrate the control and stimulus objects so I can directly compare?

or

1. Process both the control and stimulus scRNA-seq objects performing Seurat's integration tutorials, then process both the scATAC-seq control and stimulus together, and finally merge the combined scRNA-seq and scATAC-seq objects at the end.

I am fairly new to processing multiome data so any comments/ suggestions are helpful

Answer 1

[LChart]

I usually integrate the RNA and DNA components of multiomes separately. The outcome of integration is really a good "joint embedding" - ultimately the analyses will (generally) revert to the raw counts (or CPMs) and attempt to control for batch effects using a linear model or some other approach. As such, I'll treat the RNA and DNA components as separate integration tasks, and then create a joint embedding from the RNA and DNA features (typically corrected PCs or LISI) by concatenating them and performing UMAP.

This seems to work fine, but I haven't evaluated alternative approaches, mostly because this makes intuitive sense.

[Sky]

Thank you for your insights! It makes my more comfortable with my decision to process like you have above.

I guess my question would be, once you have the RNA_integrated object and the DNA_integrated object, what is the line of code you use to create one integrated object that contains both the RNA and DNA? I have tried the code below but it does not work.

RNA_integrated[['ATAC']] <- ATAC_integrated[['ATAC']]

RNA_integrated[['integrated_lsi']] <- ATAC_integrated[['integrated_lsi']]

 #says that I can't add this column using the '<-' function. I have tried other alternates like "Getassay"

RNA_integrated <- FindMultiModalNeighbors(RNA_integrated, reduction.list = list("pca", "integrated_lsi"), dims.list = list(1:50, 2:50))

Answer 2

[LChart]

So the RNA and DNA need to be aligned - that is the barcodes have to line up. Usually this is something like

seur <- CreateSeuratObject(
  counts=CreateChromatinAssay(
    counts=counts,
    ranges=peak.bed,
    fragments=frag.file,
    genome='GRCh38',  # or whateverset 
    annotation=annotation,
    validate.fragments=F
  ),
  assay='dna',
  meta.data=obs
)

seur[['rna']] <- CreateAssayObject(counts=rna.counts)

Generating the LISI (dna) and PCA (rna) embeddings will add these objects to 'Reductions'. After combining you can do something like

seur[['joint']] <- CreateDimReducObject(
  embeddings=joint_embeddings,
  key='joint_',
  assay='rna'
)