Small alignment block in between large gaps observed in CIGAR

[SHuang-Broad]

Hello Heng,

After you fixed #392 via the latest release (thanks!), I've noticed another edge case that is—I gues—related issue/feature.

I am observing some alignments that are showing CIGAR with small alignment block in between large gaps, e.g.

340D9M340I
341I1M339I
339D1M680D
341I1M340D

Now a side effect of this is that SV callers might be fooled by these CIGARs and emit more than one calls (yes, I've actually observed this behavior).

What do you think should be the correct behavior here? Do you think breaking it into split alignments is better?

I've attached a BAM with such alignment records, where a custom attribute XN:Z is added to emphasize the CIGAR.

Thanks! strange_cigars.zip

[lh3]

Completely fixing this will be very hard. I will have a look at some point. Thanks very much on the example.

[SHuang-Broad]

I think this is somewhat tied to the characteristics of PacBio read errors (haven't yet looked into ONT reads yet); so yes, I can imagine this being very hard. Thanks!