Experimental.run_combiner produces null genotypes

Hail Query & hailctl
1

I don’t think this is a bug, but after merging two gVCFs with run_combiner, I obtain an output with null genotypes for some samples and loci:

locus<GRCh38>	SAMP1:GT,END	SAMP2:GT,END
chr1:10001	0/0,10004	0/0,10019
chr1:10005	0/0,10005	NA,NA
chr1:10006	0/0,10024	NA,NA

I think I understand why this is happening - run_combiner is splitting the locii where the samples have different genotypes. But I wonder if there is a way to produce output such that every sample has a genotype at every locus (the following output was obtained with GATK CombineGVCFs):

||#CHROM|POS|INFO|SAMP1|SAMP2|
|---|---|---|---|---|---|
|0|chr1|10001|END=10004|./.:10:0:0,0,0|./.:29:0:0,0,0|
|1|chr1|10005|.|./.:14:3:0,3,45|./.:29:0:0,0,0|
|2|chr1|10006|END=10019|./.:23:6:0,6,90|./.:29:0:0,0,0|
|3|chr1|10020|.|./.:23:6:0,6,90|./.:39:3:0,3,45|
|4|chr1|10021|.|./.:23:6:0,6,90|./.:39:6:0,6,90|

Any help anyone could offer would be appreciated. Hail is very convenient in other ways, so I am hoping I can find a way around this problem.

2

The experimental.run_combiner functionality is deprecated in favor of new VariantDataset functionality (which has the same data model) found here: Hail | Variant Dataset

We’re working on fleshing out the docs for this.

The basic answer is that the matrix tables / VDSes returned by the combiner are sparse, but can be converted into a dense VCF-like representation. I think this can be done with hl.experimental.densify() on the old representation, and hl.vds.to_dense_mt(vds) on the new VDS functionality.