Longer sequences can be handled by breaking them up into overlapping segments. I could add a feature to USEARCH to do this internally, which would simplify the task for the user by handling the bookkeeping details (re-mapping coordinates and joining alignments that span multiple segments). However, other tools such as LASTZ and BLAT already provide good speed/sensitivity alternatives to BLASTN for this type of task.

Let me know if you think a USEARCH implementation for long sequences would be useful.