Specify the correct sample name for empty VCFs instead of default

[arostamianfar]

Describe the issue:
DeepVariant currently outputs default as the sample name for empty VCFs (related to the fix in #186 ). Ideally, the sample_name should be transferred from the original BAM file, but if that is too difficult to implement (since there are no examples), I think it would also be ok if we let the user customize this ID. One suggestion is that it can use the value provided by the --sample_name flag (currently, used for make_examples, but I think it can also be reused for postprocess_variants).

Using default causes issues in pipelines where the VCF is used downstream of DeepVariant (e.g. merging the VCF with other callers; or even within DeepVariant in a pipeline that scatters calling across multiple chromosomes and tries to gather them and some of those VCFs are empty).

Setup

• DeepVariant version: v0.10.0
• Installation method (Docker, built from source, etc.): docker

[pichuan]

Thanks for the suggestion @arostamianfar . Sounds like a good suggestion and should be easy to do : adding a flag to postprocess_variants. If sample_name is specified for run_deepvariant.py, we'll use it for both.

I'll file an internal issue to track this, and we should be able to have this in our next release.

[gunjanbaid]

The functionality has been implemented in the internal codebase and will be available in the next release.

[pichuan]

Hi @arostamianfar

Like @gunjanbaid said, in our internal codebase, I added the --sample_name flag to postprocess_variants as well, which will allow users to pass in a different sample_name at that stage.

But, I do want to point out that even with the current version (0.10.0), if you pass in --sample_name in the one-step command like Quick Start, you should be already able to get the VCF output with the specified sample_name.

Because the --sample_name specified in the make_examples stage does get pass on to call_variants, and then to postprocess_variants.

I tried with exactly the steps in https://github.com/google/deepvariant/blob/r0.10/docs/deepvariant-quick-start.md , and added --sample_name:

sudo docker run \
  -v "${INPUT_DIR}":"/input" \
  -v "${OUTPUT_DIR}:/output" \
  google/deepvariant:"${BIN_VERSION}" \
  /opt/deepvariant/bin/run_deepvariant \
  --model_type=WGS \
  --ref=/input/ucsc.hg19.chr20.unittest.fasta \
  --reads=/input/NA12878_S1.chr20.10_10p1mb.bam \
  --regions "chr20:10,000,000-10,010,000" \
  --output_vcf=/output/output.vcf.gz \
  --output_gvcf=/output/output.g.vcf.gz \
  --num_shards=1 \
  --sample_name=FOOBAR

After running this command (with the current version, 0.10.0), I do get FOOBAR set as my sample name.

$ zcat quickstart-output/output.vcf.gz | grep FOOBAR
#CHROM  POS     ID      REF     ALT     QUAL    FILTER  INFO    FORMAT  FOOBAR

Just an FYI, and hopefully I didn't misunderstand the issue you brought up. But, in the new code (which we plan to release soon), the --sample_name flag is added to postprocess_variants as well.

[arostamianfar]

Thank you for the clarification, @pichuan ! However, my issue isn't with changing the sample_name, but with empty VCF files. In the above example, if you change regions to --regions "chr20:1,000-10,000", you'd get default in the resulting empty VCF file.

I believe passing sample_name to postprocess_variants would fix this problem. Thanks for fixing this!

[pichuan]

Ah yes, thanks @arostamianfar for providing the --regions "chr20:1,000-10,000" example and I can now reproduce the issue. It seems like my previous internal "fix" hasn't really fixed it, so I'll send in another fix, which will come out in the upcoming release :)