normalise multiallelic deletions in merged VCFs

[mparker2]

Hi @lrauschning,

I had an issue with the merged vcfs of msyd that stems from the way that multiallelic deletions are encoded. This is a simplified section of VCF:

Chr4	19310313	DEL996661	CGAGA	C	.	PASS	.	GT	0	0	0	0	0	0	0	0	0	0	0	1
Chr4	19310313	DEL964119	CGAGAGA	C	.	PASS	.	GT	0	1	0	0	0	0	0	0	0	0	0	0
Chr4	19310313	DEL1041418	CGAGAGAGA	C	.	PASS	.	GT	0	0	0	0	0	1	0	0	0	0	0	0
Chr4	19310313	DEL917455	CGAGAGAGAGA	C	.	PASS	.	GT	1	0	0	0	0	0	0	0	0	0	0	0
Chr4	19310313	DEL963898	CGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	1	0	0	0	1	0	0	0	0	0
Chr4	19310313	DEL1044120	CGAGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	0	0	0	0	0	1	0	0	0	0
Chr4	19310313	DEL1208140	CGAGAGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	0	0	0	0	0	0	1	0	0	0
Chr4	19310313	DEL673752	CGAGAGAGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	0	0	0	0	0	0	0	1	0	0
Chr4	19310313	DEL1069688	CGAGAGAGAGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	0	1	0	0	0	0	0	0	0	0
Chr4	19310313	DEL1323846	CGAGAGAGAGAGAGAGAGAGAGAGA	C	.	PASS	.	GT	0	0	0	0	1	0	0	0	0	0	1	0

Instead of outputting a single record with a single reference and many alts, msyd produces many records with different refs. Ideally the longest deletion should be taken as the ref and many alts should be made:

Chr4	19310313	DEL996661	CGAGAGAGAGAGAGAGAGAGAGAGA	CGAGAGAGAGAGAGAGAGAGA,CGAGAGAGAGAGAGAGAGA,CGAGAGAGAGAGAGAGA,CGAGAGAGAGAGAGA,CGAGAGAGAGAGA,CGAGAGAGAGA,CGAGAGAGA,CGAGAGA,CGAGA,C	.	PASS	.	GT	4	2	5	9	10	3	5	6	7	8	10	1

I can achieve this with bcftools norm, so its not a big deal, but it might be worth implementing since multiallelic SNPs and insertions are already handled correctly.

Cheers
Matt

[lrauschning]

Hi @mparker2,

phew, good catch! Really annoying cornercase.
I think the underlying issue is that the identification of records that should be merged is done by ref allele (if on the same position). This is necessary to prevent msyd from merging small variant with synteny/SV annotations.
Cf. how mapping is used here.

I could change the matching to allow matching of prefixes without too much work, though I'm not sure how this would work if there are other annotations at the same position, like indels or SNPs, or how to handle recrods at a later position that sholud become part of the deletion after its length is extended.
Do you know how bcftools handles these cases?
Might indeed be easier to either call bcftools norm directly from msyd or recommend users to always renormalize a msyd VCF, this feels a bit like tapping a barrel of worms and hoping it doesn't burst.

Cheers,
Leon

[lrauschning]

Ah, for some reason markdown doesn't like the line numbered link.
Its defined in l. 414 in vcf.pyx, and used in the two while loops from ll. 416-435.