Description
Hi,
Looks like with the recent changes in the chaining algorithm (?) minimap2 does not align some of the difficult-to-detect inversions, which are large (several kb) and have flanking inverted repeats. The 2.18 version produces 3 alignments for each read (with a middle part inverted) as expected, although the alignments overlap by the repetitive part which could be an issue for the downstream analysis. On the other hand, starting the version 2.20 (2.22 is also affected), minimap2 would only produce two alignments for such inversion without the middle part.
"Regular" inversions with no flanking repeats are unaffected.
I attach a couple examples of IGV visualizations. Those are HG002 ONT reads, aligned to a HG002 assembly. Top panel is 2.18 and the bottom one is 2.20; "link supplementary alignments" mode enabled. As you can see, all inversions are heterozygous (making the pattern somewhat hard to see), but I suspect homozugous inversions will be no different. I was not able to detect inversion patterns from a bam file with 2.20 alignment, but the pattern was clear in 2.18.
What do you think? I will be happy to provide a few examples of actual bams if you need any.
Thanks,
Mikhail
