easy-search splits perfect full-length nucleotide alignments into two contiguous HSPs unlike BLASTN

[mlarjim]

Hi,

I’m observing a difference between BLASTN and MMseqs2 (easy-search, version 18.8cc5c) when aligning full-length nucleotide alleles against a bacterial genome.
In multiple cases, BLASTN produces a single perfect alignment (100% identity, 100% coverage), while MMseqs2 splits the same alignment into two perfectly contiguous HSPs with 100% identity.

This happens even when both fragments together exactly reconstruct the full-length BLAST alignment.

Example 1
Blastn result

qaccver	saccver	pident	length	qcovhsp	mismatch	gapopen	qlen	qstart	qend	sstart	send	evalue	bitscore
lmo0027_10	NODE_4_length_460766_cov_107.215521	100	1905	100	0	0	1905	1	1905	151470	149566	0	3518

MMseqs2 result

qaccver	saccver	pident	length	qcovhsp	mismatch	gapopen	qlen	qstart	qend	sstart	send	evalue	bitscore
lmo0027_10	NODE_4_length_460766_cov_107.215521	100	1470	77.2	0	0	1905	1470	1	150001	151470	0	2635
lmo0027_10	NODE_4_length_460766_cov_107.215521	100	435	22.8	0	0	1905	1905	1471	149566	150000	3.28E-226	781

Example 2
Blastn result

qaccver	saccver	pident	length	qcovhsp	mismatch	gapopen	qlen	qstart	qend	sstart	send	evalue	bitscore
lmo0052_8	NODE_4_length_460766_cov_107.215521	100	1974	100	0	0	1974	1	1974	121630	119657	0	3646

MMseqs2 result

qaccver	saccver	pident	length	qcovhsp	mismatch	gapopen	qlen	qstart	qend	sstart	send	evalue	bitscore
lmo0052_8	NODE_4_length_460766_cov_107.215521	100	1630	82.6	0	0	1974	1630	1	120001	121630	0	2922
lmo0052_8	NODE_4_length_460766_cov_107.215521	100	344	17.4	0	0	1974	1974	1631	119657	120000	4.57E-177	618

In both cases, MMseqs2 splits the alignment exactly at:

• 150000/150001
• 120000/120001

The two fragments are:

• perfectly contiguous in query and target
• 100% identical
• sum exactly equals the BLASTN alignment length

No sequence discontinuity (no Ns, no gaps, no rearrangements)

Command used:

mmseqs easy-search --search-type 3 --cov-mode 2 -c 0.1 --mask 0 --format-mode 0 -s 7 --split-memory-limit 32G --min-seq-id 0.9 -k 13 --spaced-kmer-mode 0 --max-accept 20 --threads 52 --remove-tmp-files --strand 2 --format-output query,target,pident,alnlen,qcov,mismatch,gapopen,qlen,qstart,qend,tstart,tend,evalue,bits,taln

Also tested with --threads 1, --split-mode 1 and --merge-query 1, and without --split-memory-limit, same behavior

Is this expected behavior of MMseqs2 nucleotide chaining/alignment (i.e. lack of HSP merging across adjacent chains), or is there a parameter that controls merging of adjacent alignments into a single HSP similar to BLASTN?

Thank you in advance