make_fragment_file silently drops marked duplicates, making frac_duplicates always 0 and fragment counts always 1

Hello,

I am passing my BAM file, which has been pre-processed already, to the snap.pp.make_fragment_file and all duplicate-flagged reads are silently dropped before SnapATAC2's own dedup pass. The results that I got are the following:

1. frac_duplicates in the returned stats is always 0.0
2. Column 5 (count) in the output fragments file is always 1 for every row

What can I do in order to retain the duplicates as well, not only the primary reads? Is there a parameter I missed, or a recommended workflow for users whose BAMs come from a pipeline that already runs MarkDuplicates? I'd like column 5 and frac_duplicates to reflect the real per-fragment read counts.

Environment

• SnapATAC2 2.9.0, Python 3.11
• Long-read single-end scATAC BAM file (ONT), Picard MarkDuplicates with --BARCODE_TAG CB

Start at snap.pp.make_fragment_file and trace how duplicate-flagged BAM reads are filtered, along with how frac_duplicates and output column 5 are computed. Done means the behavior for pre-marked duplicates is defined and the returned statistics and fragment counts reflect the retained reads.