Peak call? or Just convert to bedfile by bedtools bamtobed? to generate the accessibility matrix of single-end single cell ATAC-seq

Question

[soya]

I'm wondering about the "single-end" single cell ATAC-seq (scATAC-seq) analysis. The reads obtained from each cell is at most 1 read at each region, by nature.

Some paired-end scATAC-seq pipelines execute the peak-calling by MACS2 etc., but it seems to be hard to detect peaks from single-end single cell data. Actually, the sum of obtained peak region was about only 3% of the read coverage simply calculated by bedtools bamtobed when I tried with my sample data.

Then, how should I treat the "open chromatin region" of each single cell?

Calculate the peak region with the aggregate of single cell, and then consider whether there is a read in each cell by bedtools intersect is the best?

Answer 1

[Devon Ryan]

Then, how should I treat the "open chromatin region" of each single cell?

Don't, there's no reasonable way to measure it.

The single-cell level should only be used to cluster cells into what are hopefully meaningful cell types or cell states. After that it only makes sense to perform analyses on the aggregate population level (that is, on each cell type or state as a bulk sample). Since you'll be getting at most 2 reads in a peak from a single cell (assuming a diploid organism) and drop-out is always considerable with single-cell experiments I would encourage you to not do filtering of open regions according to whether there's at least one read in each cell.