Long overhang in POA causes bad global alignment

[rlorigro]

Some bicliques have dramatically different size overlaps:

L   s14.utg028499l +  s14.utg050706l +  20122M
L  s14.utg002821l_L   -  s14.utg028499l -  9027M172D

And create bad global alignments:

Because abpoa fails to allow large indels on one end. Need hemi-global alignment? @jeizenga

[jeizenga]

It looks like abPOA has a prefix-to-prefix option, but no suffix-to-suffix option. Also, it doesn't allow you to mix between global/local/prefix for different sequences in the alignment. Still, I think this might be something we could accomplish with a fork, or maybe a feature request.

[rlorigro]

Just updating this: it seems that a local alignment may improve the outcome in situations like this. However, @Sebastien-Raguideau it does look like there are large portions of the node that probably shouldn't be in the overlap.

In the provided example:

L   s14.utg028499l +  s14.utg050706l +  20122M
L  s14.utg002821l_L   -  s14.utg028499l -  9027M172D

It says that you have a 172bp deletion, which indicates that your exact aligner has found the hifiasm approximate overlap to be too long for one of the sequences. When we run local alignment it produces this graph:

Showing a portion of 172bp that cant align at all to the other sequence.

A dotplot of the overlaps shows a small gap (which appears as many bubbles in the GFA POA graph)

[Sebastien-Raguideau]

Hi @rlorigro, nice plot :) I saw that a lot when generating the cigars. It is pretty weird because the non-overlaping bit is at the edge of the unitig. So rather than the overlap length being too long, it is that the overlap doesn't start at the edge and instead start 172 nucleotide from it. It feels as if some strange tips were not removed from the assembly graph.

[rlorigro]

Yeah, I looked a bit longer and realized that as well, so i guess there is no easy solution for these. I agree its odd that 9,000bp would be a perfect match and then deviate into random for only 172bp.