msyd keeps SNPS/Indels from trans/dup alignments overlapping syntenic alignments

[mparker2]

Hi Leon,

I'm having some more issues with msyd VCFs. I found that when merging VCFs, msyd keeps all SNPs and Indels overlapping called core syntenic regions. However in syri output, syntenic alignments can actually overlap with non-syntenic ones (e.g. when a region is syntenic between two genomes but also duplicated elsewhere). I think msyd doesn't filter out these regions of the syntenic alignments (perhaps it should btw, or at least have the option to?), and also doesn't check the parents records of SNPs/Indels that overlap them during vcf filtering. The result is VCFs with conflicting variants like these:

Here minimap2 outputs three alignments between Col-0 and Ler. syri considers one syntenic, one as a duplication and one as a translocation. The syntenic and translocation alignments have corresponding but different SNPs and Indels. Instead of just keeping the variants that originate from the syntenic alignment however, msyd copies all of the variants that overlap the syntenic region across to the output vcf. The consequence (as well as not making much biological sense) is a broken VCF, because a SNP adjacent to an Indel upsets a lot of tools (e.g. bcftools consensus).

Something similar also happens when two syntenic alignments overlap, but this is more of a syri problem...

[lrauschning]

Hi Matt,

thanks for reporting! This seems like quite an annoying issue...
Yes, currently msyd outputs whatever is in the SyRI/VCF output at positions it deems core/sufficiently multisyntenic.

I guess one solution could be to pre-filter the small variant calls to those derived from syntenic alignments, which should be possible just from the VCF Parent record/the first three letters of the ID in col. 9 of the syri.out file.
This wouldn't work in overlapping SYNAL regions, but those should hopefully not have (too) different variant calls.
Filtering from these would be a lot more annoying, as msyd doesn't preserve the SyRI SYNAL IDs.

Do you think this would work/would it be possible to try this manually at first in your setup? I think this functionality should be in msyd, but I'd prefer to know it works before implementing it to avoid making a breaking change for no reason.

[mparker2]

you are right, I could filter like that. My current workaround is to use bedtools genomecov to identify regions in the original bam file where alignments (of any kind, syntenic/nonsyntenic) overlap, and then create a bed file of blacklisted regions which I then use to filter the vcf file after msyd:

https://github.com/schneebergerlab/snco_mapping_pipeline/blob/08878c22a4ea2ec57646ba5017b82160b2166d94/snco_pipeline/pipeline/rules/variants.snakefile#L180C1-L205C13

This works for my purposes but I kind of think that the issue needs solving upstream too. I personally think that it is an oversight that SNPs/Indels in the syri output do not inherit their parent IDs from the alignment records. That would make all this stuff easier to solve. I'm not sure what's happening with syri going forward though

edit: whoops clicked submit/close by mistake :')

[lrauschning]

Hm, yeah you are right this would be better to solve upstream and provide some kind of a consensus.
I can talk with Manish and Korbinian about it next week. I don't think I have push rights to SyRI, but am probably the one most familiar with the codebase, and fingers crossed this shouldn't need too much of a change.

[lrauschning]

Made an issue over at schneebergerlab/syri#294 to track this. Egor is working on the SyRI codebase anyway, so the fix would go into his update to syri.
right now we both have exams, but can probably take a look at it after.