Empty sample/genotype-data in single samples of a pedigree cause filters to crash

[Nicolai-vKuegelgen]

Describe the bug
In a case/pedigree with multiple samples some variants may not have usable data from all samples (either due to missing coverage or also on the Y chromosome). In vcf files this can either be encoded as a single "." for that sample in the sample/genotype block or with individual missing values for each defined Format field. In the tsv file used for data-import to Varfish the sample specific data (genotype column) can - in principle - also be empty for single samples (i.e. """sample_2""": {} ). However, in cases like the the variant filtration for this sample will fail for the whole set of variants (SNVs or SVs).

To Reproduce
Steps to reproduce the behavior:

1. Generate a tsv file with empty genotype data for a single variant in a single sample (i.e. using mehari on a vcf with a "." in the sample block).
2. Import this case to Varfish
3. Attempt variant filtration
4. See error

Expected behavior
Given that some samples may not have any usable information for some variants, ideally the variant filtration should be able to deal with missing data.
Alternatively, import of variants with missing data for even a single sample should be reject, so that filtration will not fail due to this.

Additional context
This could be fixed by never writing empty sample/genotype-data into the tsv files used for varfish import, see mehari issue 672

[stolpeo]

For the Y chromosome, there is no standardized output for the genotype:

• Dragen outputs GT as ./. and other data as .
• GATK outputs GT mostly as ./. (except it is reported noise), and other data as 0
• Varfish annotator converts the . of other data to 0
• mehari converts the . of other data to -1

[stolpeo]

RCA

When starting a SV query, the variants from the database are written out to a temporary vcf file to be then submitted to the varfish-server-worker. However with some variants the genotype dictionary for an individual can be empty. If that happens to an CNV entry, the conversion fails. Per default, expected fields that are not in the genotype default to ., however, the cn entry is cast with an int() which fails the conversion.

varfish-server/backend/svs/models/jobs.py

Lines 415 to 417 in 3533238

	coerce_db_to_vcf.get(key, lambda x: x)(
	record.genotype[sample].get(key, ".")
	)

varfish-server/backend/svs/models/jobs.py

Lines 331 to 333 in 3533238

	coerce_db_to_vcf = {
	"cn": lambda x: int(x),
	}

[stolpeo]

Solution

Allow . as cn value.

        coerce_db_to_vcf = {
            "cn": lambda x: x if x == "." else int(x),
        }