After using subsetCells(), should AddTileMatrix be rerun?

[macpheke]

If I subset my project with subsetCells specifically, do I need to rerun AddTileMatrix and AddGeneScoreMatrix? That is, prior to redoing downstream processes such as iterative LSI, UMAP, MAGIC, etc.

I know it reduces the size of the ArrowFile - presumably to keep info from just the subsetted cells - but I'm wondering if that matters for LSI and the variable features it uses, etc.

The goal here is to subcluster a group of cells to improve resolution.

[rcorces]

I think there are a few points of confusion here. Make sure you understand the difference between an ArchRProject and the associated Arrow files.

subsetCells() doesnt touch the arrow files, just the ArchRProject.
You are most likely looking for subsetArchRProject() - https://www.archrproject.com/reference/subsetArchRProject.html

After subsetArchRProject(), you do not need to rerun addTileMatrix() or addGeneScoreMatrix()

[macpheke]

What I mean is, if I use subsetCells() on an existing ArchRProject object, then I re-run addTileMatrix() and addGeneScoreMatrix(), the associated Arrow file is reduced in size. So I am wondering:

• does subsetCells() + addTileMatrix() + addGeneScoreMatrix() perform a similar operation as subsetArchRProject() (when dropCells = TRUE,) in that it removes inactive cells from the Arrow file?
• would using subsetCells() without re-running addTileMatrix() and addGeneScoreMatrix() adversely affect the accuracy of any subsequent iterative LSI and UMAP?

I ask because subsetArchRProject() is not working for me when dropCells = TRUE (I'm trying to save space), I get the same error as #663 and a nearly identical logfile but I am using genome hg38. While I try to figure out why I was hoping to use this as a workaround

[rcorces]

This isnt the designed use for this function and I would not consider the proposed use as equivalent to subsetArchRProject(). Arrow files also contain the fragments and other information that isnt accounted for. I have no idea if you will run into downstream problems using your approach but I would imagine so. Hopefully we have a solution to #663 soon.

[macpheke]

Ahh, thank you, I will certainly avoid doing that then.

So in the meanwhile, if I perform subsetCells() on an ArchRProject and then re-run addIterativeLSI(), addUMAP(), addImputeWeights() etc., the results are still valid even though the associated Arrow file contains unwanted cells/cells not in the desired subset list?

[rcorces]

should be! I would think that any downstream functions that run without error are valid.