I've come across an peculiarity when aligning a 19bp sequence (AGCATGGGGAGCTCCCGGG) to the human reference genome (hg38) using bwa aln.

When aligning with 6 mismatches allowed (-n 6), I obtained a smaller number of hits compared to an alignment with 5 mismatches allowed (-n 5).

I used the following configurations for the alignments:

bwa aln -n 6 -k 0 -l 100 -o 0 -N hg38Index toyExample.fastq > out_n5.sai
bwa aln -n 6 -k 0 -l 100 -o 0 -N hg38Index toyExample.fastq > out_n6.sai

where toyExample.fastq is a dummy fastq file containing my sequence:

@AGCATGGGGAGCTCCCGGG
AGCATGGGGAGCTCCCGGG
+AGCATGGGGAGCTCCCGGG
~~~~~~~~~~~~~~~~~~~

I set the seed length to a large number to ignore seed-specific constraints. Anyone has any idea what's going on? I would expect all the hits stored in out_n5.sai to be a subset of the hits stored in out_n6.sai, but that's not the case. This phenomenon does not happen for smaller number of mismatches (that is all hits found for -n 4 mismatches are also found with -n 5 mismatches, for instance).