Added a force output sites argument to GenotypeGVCFs

[davidbenjamin]

Closes #6239.

@ldgauthier @bbimber Here's what I settled on:

• GenotypeGVCFs gets a new argument for force-genotyping intervals. This is parsed exactly like any other interval argument, so the following are all valid:

--force-output-intervals 20:45000
--force-output-intervals intervals.interval_list
--force-output-intervals dbsnp.vcf
--force-output-intervals 20:1000-2000 --force-output-intervals 20:3000-4000

• Only the sites matter, so if a VCF is given the alleles are ignored.
• If the site is absent in the input gVCF, there is no output. No data in yields no data out.
• If the site overlaps a record in the input gVCF, there is output, regardless of whether the record is a variant, a REF block, or a spanning deletion.

[davidbenjamin]

As far as I can tell the failures are just Travis acting up. GenotypeGVCFsIntegrationTest passes locally.

[ldgauthier]

@davidbenjamin What's the difference between --force-output-intervals myIntervals.list and --include-non-variants -L myIntervals.list?

[bbimber]

@ldgauthier I have not had a chance to check out this code in detail (I'll try to get someone in my lab to do this today); however, I would hope that the difference is that --force-output-intervals would output at those intervals (whether variant or not) AND include any additional variant sites. The latter, --include-non-variants and -L would only output the intervals from -L.

[ldgauthier]

Yes, thank you for jogging my memory @bbimber. @davidbenjamin adding something to that extend in the BadArgumentException message would be helpful.

[davidbenjamin]

@ldgauthier I wrote a detailed error message. Anything else?

[ldgauthier]

Looks good!

[bbimber]

@davidbenjamin Hello, we started using this feature more seriously and have a question. the output from GenotypeGVCFs with --force-output-sites seems to include sites with <NON_REF> as the ALT allele. Is this expected? I am guessing that these sites used to be filtered (no passing evidence of variation), and the point of this feature is the include sites based on whitelist?

The problem is that some downstream tools dont know what to do with this. There is probably some subtlety here, but I would think either a given sample has callable variation, it is REF, or it is no-call? Is <NON_REF> in the output by design?

[davidbenjamin]

@bbimber Do I understand correctly that your PR #6406 resolves the issue you found?

[bbimber]

yes, it seems to. it seems this was a simple bug where GenotypeGVCFsEngine.removeNonRefAlleles() wasnt working as intended for multi-allelic sites; however, I'm not sure I understand the entire genotyping process well enough to be certain on this. we can iterate on #6406