Indeed does happen, but more often they are duplex reads that are not detected as such by Dorado. Then the sequences on either side of the mid strand adapter are reverse complement (example pic)

Hey Misha, on a 10K read sample around 60% is the expected dual-end (top), but often a single-end is enough to assign already. That puts it at around 90% (top 3 in image). Double barcode ligations do happen (bottom) but not sure about the exact stats of bleeding because of that.

In Barbell we solve this by first annotating all the reads, and then detecting all patterns, which looks like this:

Why is this a problem? Remaining adapters/barcodes not only contaminated assemblies but also created "artificial" links in taxonomic annotation, to Enterobacteriaceae, and to contaminated assemblies in NCBI.

Many tools, including the widely used Dorado, almost always only detected the *first* occurrence, leaving the rest untrimmed. In the figure black lines indicate matches to adapters + barcodes in trimmed reads.

For rapid barcoding only ~83% of the reads contain the expected single barcode on the left. The rest? Barcodes on both sides (6.1%), two barcodes on the left side (3.5%), and so on.