Create a issue in ivar - indels

[Alema91]

Create a issue to examine the ivar performance in indels

[Alema91]

Describe the bug

We have observed that ivar variants can generate false positive variant calls for SARS-CoV-2 genomes that contain insertions or deletions. Here is an example from a private genome that contains the 6bp ORF8 deletion:

CHROM	POS	REF	ALT	GENE	EFFECT	HGVS_C	HGVS_P	DP	REF_DP	ALT_DP	AF	sample	software	lineage
NC_045512.2	28247	AGATTTC	A	ORF8	conservative_inframe_deletion	c.355_360delGATTTC	p.Asp119_Phe120del	76166	48839	64275	0.84	218025	ivar	AY.33

Position 28247 has a well supported deletion (nearly 70000x coverage) and ivar variants is calling a variant inside that deletion (IGV image and variant table):

CHROM	POS	REF	ALT	GENE	EFFECT	HGVS_C	HGVS_P	DP	REF_DP	ALT_DP	AF	sample	software	lineage
NC_045512.2	28253	C	A	ORF8	missense_variant	c.360C>A	p.Phe120Leu	3954	61	3851	0.97	218025	ivar	AY.33

This variant should be included in the consensus according to our quality criteria (variants with an AF > 0.75). Therefore, the AF is overestimated due to the misscalculation in the variant depth (because of the deletion).

CHROM	POS	REF	ALT	GENE	EFFECT	HGVS_C	HGVS_P	DP	REF_DP	ALT_DP	AF	sample	software	lineage
NC_045512.2	28247	AGATTTC	A	ORF8	conservative_inframe_deletion	c.355_360delGATTTC	p.Asp119_Phe120del	75039	10741	63575	0.847	218025	VarScan	AY.33
NC_045512.2	28248	GATTTCA	G	ORF8	disruptive_inframe_deletion	c.356_361delATTTCA	p.Asp119_Ile121delinsVal	64673	126	189	0.003	218025	VarScan	AY.33
NC_045512.2	28249	ATTTC	A	ORF8	frameshift_variant	c.357_360delTTTC	p.Asp119fs	64614	98	79	0.001	218025	VarScan	AY.33
NC_045512.2	28251	TTCATC	T	ORF8	frameshift_variant&stop_lost&splice_region_variant	c.360_364delCATCT	p.Phe120fs	69573	5295	48	0.001	218025	VarScan	AY.33
NC_045512.2	28252	TC	T	ORF8	frameshift_variant	c.360delC	p.Phe120fs	69316	4895	151	0.002	218025	VarScan	AY.33
NC_045512.2	28253	C	A	ORF8	missense_variant	c.360C>A	p.Phe120Leu	69561	15	5009	0.072	218025	VarScan	AY.33

Other variant callers such us Varscan detect this variant with an AF << 0.25 because the depth of that position is calculated taking into account the deletion reads. Thus, the AF differs from ivar variants.

Expected behavior

We might suggest that ivar variants overestimate this variant based on the depth calculation and therefore can cause issues with variant prediction in indels (insertions and deletions).

This issue may be related to #79, #83, #85, #103

To Reproduce

Run ivar variants with these params:

samtools mpileup \\
        -a \\
        --count-orphans \\
        --no-BAQ \\
        --ignore-overlaps \\
        --max-depth 20 \\
        --fasta-ref fasta \\
        --min-BQ  | ivar variants -q 30 -t 0.25 -m 10 -r fasta gff -p sample

[Alema91]

Better comprehension and need data to reproduce the issue.

[Alema91]

Done #126