Difficulty removing depth correlation in clustering

[APuchkina]

Hi,

Thank you for the great package!

I am trying to use this package to analyse single cell methylation sequencing data, focussing on the methylation in gene bodies and TSS. In my case more methylation means higher gene expression, so in that sense the data is very similar to ATAC seq data (also in terms of sparcity). The main difference however is that in my case all fragments are the same size (5 bp) and the couse matrix is based on 100kb genomic bins and not peaks.

For the analysis I largely followed the PBMC vignette except for the TSS score and NucleosomeSignal filtering, as these do not apply in my case. Additionally, I need to perform integration following the https://stuartlab.org/signac/articles/integrate_atac vignette, since I have some batch effects due to the plate based sequencing performed.

It has been working quite well, but I have noticed something related to issue #122.

This is what my DepthCor plot looks like prior to integration for the merged object:

With component 1 and 2 both correlating with depth.
When I look at these plots for the different plates separately, depth correlates either with component 1 only or with 1 and 2 but not as clearly as for the merged object:

So my question is: how is it possible that both component 1 and 2 correlate equally and what would be the recommended components to use for integration?

I have tried different things but overall my UMAP still seems to be influenced by sequencing depth:

Thank you in advance!

Arina