Recommendation to regress out confounding variables in scatac dataset?

[ejc043]

Hello,
Thank you very much for this wonderful tool and all the great documentation and resources for the community.
For scRNA-seq preprocessing, there is an option to regress out confounding variables in the ScaleData vars.to.regress parameter. I understand that RunTFIDF should account for confounding variables like sequencing depth.
Looking at a scATAC dataset analyzed in this vignette, I would expect to see greater euchromatin (more open chromatin) in pluripotent stem cells (HSC) compared to background cells.
I ran two different types of DAR analysis: 1. typical wilcoxon test FindMarkers(human.atac, ident.1 = 'HSC') and 2. LR while regressing out nCount_ATAC FindMarkers(human.atac, ident.1 = 'HSC', test.use = 'LR',latent.vars = 'nCount_ATAC' )
Indeed, the lattern identifies more genomic features with increased accessibility in HSC suggesting sequencing-depth related confounding effects are still present in the normalized data. I have below two histograms which show the distribution of log2FC between HSC vs background cells.

histogram of avg log2FC using wilcoxon

histogram of avg log2FC using LR ~ nCount_ATAC

I need normalized counts that have removed confounding effects like from nCount_ATAC. I'm wondering if it is valid to use the typical NormalizeData ScaleData preprocessing workflow on scATAC dataset to leverage vars.to.regress option? Or would the team have any alternatives?

Thank you so much for all your support!