Rationale behind HTODemux

[XinLi-0419]

Hi Developers, I am using HTODemux right now and I understand it is based on the bimodal distribution and a positive quantile. However, I found always there are some of the HTO A singlet that would have more HTO B hashtags than the calculated cutoff for HTO B, not by a lot of course. I wonder if there is any rationale for that? And I also wonder is it okay to process it like a strict bimodal expression and use the valley as the cutoff? (This is easier and what we do with most bimodal distribution). Thanks a lot for answering my question!