Hi @Peter,
I think dada2 and vsearch fundamentally differ enough that a comparison wouldn't really be fair to either one.
I think dada2 doesn't allow for mismatches because it assumes that by the time merging is occuring those nts are error-free, either by inferring the correct sequences, or by strictly discarding those that couldn't be resolved. So working under that assumption it probably makes sense to be strict and not allow mismatches. Vsearch doesn't do any inferring so of course it relies on other methods for quality control. I would agree that in this scenario sharing info between the merging region makes total sense.
Only within the overlapping sequences/regions though, that is what you meant right?
Yes and no? I think with PE reads the emphasis is on making sure they can be confidently joined, otherwise joining PEs with low quality overlap or lack therefore completely can lead to so much spurious reads.
