How does ivar deal with conflicting/overlapping variants?

[charlesfoster]

Hi,

I recently ran ivar with a sample corresponding to the Delta variant. After some tweaks I've mentioned in another issue on Github, the analysis seems to be working well. I just have one question: how does ivar deal with overlapping variants?

If I run ivar variants like so:

samtools mpileup -A -B -d 0 -Q 0 --reference $REF sample.primertrim.sorted.bam | ivar variants -p sample.refIncluded.noBAQ -q 20 -m 10 -t 0.1 -r $REF -g reference_features.gff3

I recover the following two overlapping variants (among others):

REGION	POS	REF	ALT	REF_DP	ALT_DP	TOTAL_DP	ALT_FREQ
NC_045512.2	28247	A	-GATTTC	2242	2233	2479	0.900766
NC_045512.2	28253	C	A	0	78	78	1

There's also an overlapping substitution at 28249.

As you can see, the 6 bp deletion after 28247 overlaps with a single base substitution at 28253.

I then run ivar consensus:

samtools mpileup -A -d 0 -Q 0 -B --reference NC_045512.fasta sample.primertrim.sorted.bam | ivar consensus -p refIncluded.t0.noBAQ.consensus -t 0.9 -n N

The 6 bp deletion occurs occurs as expected, but the 1 base substitution does NOT get included, despite both passing the filters in the ivar consensus command. I'm happy with this: given the much stronger read evidence for the deletion, I'd prefer the 6 bp deletion without the SNP appearing. However, I'm just confused as to how/why only one gets included: how/why is the deletion chosen over the SNP?

If I try to use the ivar variants output as an input to bcftools consensus (after converting to vcf format, and bgzipping + tabixing), I get the following warning:

The site NC_045512.2:28249 overlaps with another variant, skipping...
The site NC_045512.2:28253 overlaps with another variant, skipping...

So, it appears that the deletion is incorporated, and the next two overlapping variants are ignored. Is this what ivar does?

Thanks!