Thanks @wangj50!
I agree! I updated that to clarify, "is likely not the case for your demultiplexed data".
With that said, looking back at that log, I made a mistake and must've misread something - I thought that there were the same number of unique sequences in each sample, fwd and reverse (that would likely be an issue, because then everything truly would be identical between fwd and rev). Given that, probably none of what I wrote above is valid, realizing I made a mistake!
Thanks! That is an important distinction (and one I should've made above) - I think most tools should handle making sure read-pairs are binned or dropped together (as a pair), preventing a mismatch from happening, but it really depends on the tool (I suspect there are some homegrown scripts floating around online that are responsible for creating these mismatched data!).
Going back to @hhftang's issue, and your similar issue you just reported, it sounds like this error message (Error in isBimeraDenovoTable(unqs[[i]], …, verbose = verbose) : Input must be a valid sequence table.) is caused when none of the sequences are actually merged. This might be happening because --p-trunc-len-r is set really low here (25). As noted in the docs, there needs to be at least a 20nt overlap between fwd and rev reads. Maybe take another look at your demux summarize viz and make sure that your trunc-len-r value is right. Keep us posted, and thanks for setting me straight!