elbow plot function

[mitchTwoTimes]

hello,

Does archR support a way to generate elbow plots as a heuristic to determine the number of principal components? If not, are there any hacks to accomplish this?

Thank you!

[rcorces]

Are you asking whether ArchR provides a way to plot this information or are you asking whether ArchR gives you the information to be plotted? And by principal components, are you talking about LSI dimensions?

[mitchTwoTimes]

Are you asking whether ArchR provides a way to plot this information or are you asking whether ArchR gives you the information to be plotted?

I suppose I'm asking both

And by principal components, are you talking about LSI dimensions?

Oh, right. I'm referring to the dimsToUse arg in addIterativeLSI.

To rephrase, I would like to know if there is a method to settle on the number of dimensions to include when adding iterative LSI.

[rcorces]

This is definitely not a topic that I consider myself super well read on. That being said, I dont think you will find many reasons to exclude specific LSI dims. Certainly this is not very important during that addIterativeLSI() step but I also dont think there are many good reasons to change this from the default in downstream steps like addUMAP().
ArchR automatically handles LSI dimensions that are correlated to sequencing depth (via the corCutOff parameter) so you dont need to worry about this like you would with other scATAC tools.

[mitchTwoTimes]

Thanks for the thoughtful response. In the context of multiome (paired) data, is it appropriate to include all dimensions for the gene expression matrix as well?

[rcorces]

sorry for the delay in replying. Yes - I would include all dims for RNA as well.

[AnjaliC4]

Hi @mitchTwoTimes, here is a simple way to make elbowPlots (as Signac does):
rd <- getReducedDims(proj, reducedDims = "IterativeLSI",dimsToUse = 1:50)
sd <- colSds(rd)
ggplot(data.frame(dims = 1:50, stdev = sd[1:50]))+geom_point(mapping = aes_string(x = 'dims', y = 'stdev'))+labs(x = "LSI",y="Standard Deviation")

Hope it helps!

[DaianeH]

Hi,

I’m sending attached the elbow plot of LSI dimensions generated with the above code on my archR object, and it seems that the first dimension has the lowest SD - is it because it was correlated to depth so it was removed by archR?

Thank you in advance,

elbow_estimatedLSI.pdf

[qdong2023]

Hi,

Is there a way to plot the correction of sequencing depth with each reduced dimension component in Archr? Like to function of DepthCor() in Signac?

Thanks