Reminder: Transcript counts are not sufficient to distinguish doublets from
#scRNAseq libraries. You can see this in the original DropSeq barnyard experiment, where libraries mapping to both mouse and human ("mixed") were used to determine the
#doublet rate.
Transcript counts per cell are so broad that using a threshold to predict doublets is impossible. Yes, mean transcript count is higher in doublets, but there's huge overlap in the transcript count range between doublets and singlets.