Missing calls overlapping RefCall/null-GT gVCF records

[DavidStreid]

In evaluating glnexus for merging a single sample, glnexus is missing calls where the gVCF call overlaps a Refcall record with a null (./) genotype. The variant calls are correct so is there an option I can add to preserve these calls?
Observed this mostly in INDELs, but there are some SNVs too. Two examples are below with the attached files here - workspace.tar.gz

INDEL

Expecting a HOM ALT, but no call is emitted from glnexus if the overlapping REF call is present

$ grep -A 2 "#CHROM" indel*.vcf
indel_call.g.vcf:#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	NA24385
indel_call.g.vcf-chr1	7655595	.	TAC	T,<*>	26.1	PASS	.	GT:GQ:DP:AD:VAF:PL	1/1:20:18:3,15,0:0.833333,0:26,21,0,990,990,990
--
indel_merge.g.vcf:#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	NA24385
indel_merge.g.vcf-chr1	7655595	.	TAC	T,<*>	26.1	PASS	.	GT:GQ:DP:AD:VAF:PL	1/1:20:18:3,15,0:0.833333,0:26,21,0,990,990,990
indel_merge.g.vcf-chr1	7655597	.	C	T,<*>	0.1	RefCall	.	GT:GQ:DP:AD:VAF:PL	./.:15:3:1,2,0:0.666667,0:0,14,31,990,990,990
$ glnexus_cli --config DeepVariant_unfiltered indel_merge.g.vcf 2> /dev/null | bcftools view - -H
$ rm -rf GLnexus.DB/
$ glnexus_cli --config DeepVariant_unfiltered indel_call.g.vcf 2> /dev/null | bcftools view - -H
chr1	7655595	chr1_7655595_TAC_T	TAC	T	26	.	AF=1;AQ=26	GT:DP:AD:GQ:PL:RNC	1/1:18:3,15:20:26,21,0:..

SNV

Same issue as before - running glnexus on the g.vcf with the overlapping null/RefCall does not output the expected call

$ grep -A 2 "#CHROM" snp_*.vcf
snp_call.g.vcf:#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	NA24385
snp_call.g.vcf-chr3	103921462	.	A	T,<*>	48.2	PASS	.	GT:GQ:DP:AD:VAF:PL	1/1:27:26:0,26,0:1,0:48,26,0,990,990,990
--
snp_merge.g.vcf:#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	NA24385
snp_merge.g.vcf-chr3	103921461	.	TA	T,<*>	0.3	RefCall	.	GT:GQ:DP:AD:VAF:PL	./.:12:28:26,2,0:0.0714286,0:0,11,40,990,990,990
snp_merge.g.vcf-chr3	103921462	.	A	T,<*>	48.2	PASS	.	GT:GQ:DP:AD:VAF:PL	1/1:27:26:0,26,0:1,0:48,26,0,990,990,990
$ glnexus_cli --config DeepVariant_unfiltered snp_call.g.vcf 2> /dev/null | bcftools view - -H
chr3	103921462	chr3_103921462_A_T	A	T	48	.	AF=1;AQ=48	GT:DP:AD:GQ:PL:RNC	1/1:26:0,26:27:48,26,0:..
$ rm -rf GLnexus.DB/
$ glnexus_cli --config DeepVariant_unfiltered snp_merge.g.vcf 2> /dev/null | bcftools view - -H
$

I've played around with other configs, but the closest to the desired behavior of outputting the variant is with a modified YAML based on DeepVariant_unfiltered with just the additional option trim_uncalled_alleles: false (in custom.yml in attached). However, the result is still unexpected because the records have a null genotype like below -

#CHROM	POS	ID	REF	ALT	QUAL	FILTER	INFO	FORMAT	NA24385
chr1	7655595	chr1_7655595_TAC_T	TAC	T	26	.	AF=1;AQ=26	GT:DP:AD:GQ:PL:RNC	./.:3:.:.:0,0,0:OO

Thank you for the any help and supporting this useful tool!

[JMdeSteAgathe]

Hi,
same issue on my side!
I struggle to keep RefCall variants where GT=./. when merging gvcf
(I missed a diagnosis which was filtered out because of GT=./.)
is there an option to keep them after merging?
thanks