Speed-up searching when using the --top_hits_only option about vsearch HOT 3 CLOSED

What do you think @torognes? the idea of using top_hits_only and --maxaccepts 0 is to be sure to find all best matches: if there are 250 target sequences 100% identical to my query sequence, I want to get all of them in the output.

Or did I misunderstand the process? If the target sequences are pre-sorted by k-mer profile similarity first, can I safely set --maxrejects 32 to implement the speed up process I described above?

from vsearch.

In this scenario I think it would make sense to set a reasonable maxrejects value (e.g. 32 or even 1000). I think this will be a good heuristic that would capture all the top hits in almost all cases.

from vsearch.

Yes, I agree with you.

As the initial similarity sorting is done on k-mer profiles, which is a proxy for the actual pairwise similarity, there is always a risk of missing the optimal hit. This is especially true when none of the target sequences is close to the query: the behavior of the k-mer profile pre-sorting gets less predictable for low similarities.

Nonetheless, it is probably safe to use --maxrejects 32 as you suggested.

from vsearch.