Interpreting QC metrics within clusters

[wudustan]

I have merged 6 scATAC datasets of different lines for comparison. I ran through the relevant vignettes in Signac.

Generating QC violin plots for the merged data I got this:

Based on my results I chose the following parameters for filtering:

peak_region_fragments > 5000
peak_region_fragments < 100000
pct_reads_in_peaks > 15
blacklist_ratio < 0.05
nucleosome_signal < 4
TSS.enrichment > 2

After filtering, I generated UMAP plots of the QC metrics to see how they are distributed across the groups.

1. Looking at peak_region_fragments, I notice that the clusters have 'foci' of high values while the rest of the cluster can be relatively low - how I interpret this?

2. Looking at nucleosome_signal, pct_reads_in_peaks and blacklist_ratio, I'm noting one cluster has a very different / low value compared to the others. Am I correct in assuming this cluster is likely low QC cells and would be best removed?

Any help here would be appreciated

[timoast]

It looks like you have a group of low-quality cells in the center of the plot that have high blacklist-region counts, low percent reads in peaks, low nucleosome banding pattern, and low total counts. I would recommend adjusting some of the QC thresholds so that these cells are filtered out, then re-processing the dataset without those cells.