Segmented inverted alignments

[armintoepfer]

I'm looking at an example where a read has an inversion in the middle. Everything gets aligned properly, but I do not understand why the inverted part has its cigar reversed. The flag only states 2064, for supplements and reverse. Is this a special combination, leading to different record printing? Do you have documentation/specs about this case?

Examples:
Reference https://gist.github.com/armintoepfer/ef6505a5a9db5c1ba9b49ffe910b326e
Seqs: https://gist.github.com/armintoepfer/a24959b2e199784c35fc11651aee84d9

minimap2 -Y -x map-pb -a -z 400,20 ref.fasta seqs.fasta

First example:

The first sequence mid_inv should be aligned in three segments 180 fwd, 540 rev, 1380 fwd.
The longest is picked as primary,

mid_inv 0 chr02__substr__128948_131048 721 60 720S1380M

Makes totally sense.
The first part also gets aligned as forward and supplementary

mid_inv 2048 chr02__substr__128948_131048 1 60 180M1920S

Now the confusion for the middle part

mid_inv 2064 chr02__substr__128948_131048 181 0 1380S540M180S

The cigar is reverse and the actual SEQ is also reverse-complemented.

Second example end_inv

Here, the sequence is split into two segments 960 fwd, 1140 rev.
Minimap2 picks the second segment as the primary alignment, reversed the cigar and rc the SEQ

end_inv 16 chr02__substr__128948_131048 961 60 1140M960S

And the first part is aligned properly as forward

end_inv 2048 chr02__substr__128948_131048 1 60 960M1140S

But if I align a single segment that is reverse-complement, the cigar and sequence are forward strand (genomic) with flag 16 for rev.

Why are segments treated differently? And why are the inversions tagged with MAPQ 0?
Using 2.10-r763-dirty

[lh3]

The 0 mapping quality is a bug, which has just been fixed via 372c90c. I am not sure about the other questions, though. SAM/PAF only works with colinear alignments. If inversion alignment is not reported this way, what would you expect to see?

[armintoepfer]

Thank you for fixing the MAPQ bug; this works now.

I'm not sure how I'd do it. I'll think about it...