MNP multi-allelic site merge losing depth/AD information?

[yttria-aniseia]

from a larger WGS analysis, a multiallelic site output where i was unable to recover useful depth or VAF information

deepvariant gvcf:

chr13	87627866	.	GTATACACATA	G,<*>	8.7	PASS	.	GT:GQ:DP:AD:VAF:PL	0/1:6:22:6,16,0:0.727273,0:7,0,7,990,990,990
chr13	87627873	.	C	T,<*>	4.1	PASS	.	GT:GQ:DP:AD:VAF:PL	0/1:3:3:0,3,0:1,0:0,0,4,990,990,990

i see

• del record with DP=22; AD=6,16,0
• overlapping record with DP=3; AD=0,3,0

after glnexus

chr13	87627866	chr13_87627873_C_T;chr13_87627866_GTATACACATA_G	GTATACACATA	GTATACATATA,G	11	.	AF=0.684211,0.184211;AQ=11,11;AC=1,1;AN=2	GT:DP:AD:GQ:PL:RNC:PF	1/2:3:.,3,16:3:0,0,0,0,0,0:..:TR_OVERLAP

• multiallelic record with DP=3; AD=.,3,16

taking min DP here is... unexpected, but ignoring that, i would still want AD to sum to 22...

GLnexusConfig

{unifier_config: {drop_filtered: false, min_allele_copy_number: 1, min_AQ1: 0, min_AQ2: 0, min_GQ: 0, max_alleles_per_site: 32, monoallelic_sites_for_lost_alleles: true, preference: common}, genotyper_config: {revise_genotypes: false, min_assumed_allele_frequency: 9.99999975e-05, snv_prior_calibration: 1, indel_prior_calibration: 1, required_dp: 1, allow_partial_data: true, allele_dp_format: AD, ref_dp_format: MIN_DP, output_residuals: false, more_PL: true, squeeze: false, trim_uncalled_alleles: true, top_two_half_calls: false, output_format: BCF, liftover_fields: [{orig_names: [MIN_DP, DP], name: DP, description: "##FORMAT=", type: int, number: basic, default_type: missing, count: 1, combi_method: min, ignore_non_variants: true}, {orig_names: [AD], name: AD, description: "##FORMAT=", type: int, number: alleles, default_type: zero, count: 0, combi_method: min, ignore_non_variants: false}, {orig_names: [GQ], name: GQ, description: "##FORMAT=", type: int, number: basic, default_type: missing, count: 1, combi_method: min, ignore_non_variants: true}, {orig_names: [PL], name: PL, description: "##FORMAT=", type: int, number: genotype, default_type: missing, count: 0, combi_method: missing, ignore_non_variants: true}]}}

IGV read view

[yttria-aniseia]

i see this is probably intentional and not a bug

GLnexus/src/genotyper_utils.h

Lines 449 to 451 in fae491b

	// if a half-call (cs.second), censor only the reference allele
	// depth, which can be misleading when there are overlapping or
	// unphased records.

but find this outcome also misleading!

as i'm later splitting the multi-allelic records with bcftools norm, it's a problem to get DP=3; AD=.,16 but i guess in my case i just want to add a MAX_DP liftover instead...