Deepvariant 1.8.0 has issues with using --proposed_variants and --variant_caller

[chaoxinzhang]

Hi！

Deepvariant 1.8.0 has issues with using --proposed_variants and --variant_caller

here is the command for deepvariant 1.8.0:

time sudo docker run  -m 480G \
   -v "/mnt/disk1/ref/bed":"/input" \
   -v "/mnt/disk2/SRA3":"/output" \
   -v "/mnt/disk1/ref":"/reference" \
   google/deepvariant:"1.8.0" \
   /opt/deepvariant/bin/run_deepvariant \
  --model_type WGS \
  --ref /reference/Homo_sapiens_assembly38.fasta \
  --reads /output/ERR10502884.markup.bam \
  --output_vcf /output/ERR10502884.output180.vcf.gz \
  --num_shards 120 \
  --intermediate_results_dir /output/tmp3 \
  --make_examples_extra_args variant_caller="vcf_candidate_importer",proposed_variants="/output/output1.vcf.gz"

The output180.vcf.gz file has over 5 million variations.

zcat /mnt/disk2/SRA3/ERR10502884.output180.vcf.gz  |grep PASS|wc -l

The ERR10502884. output 180. vcf. gz file has less than 100,000 lines with a PASS flag, and the rest have a refcall flag

But here is the command for deepvariant 1.6.1:

time sudo docker run  -m 480G \
  -v "/mnt/disk1/ref/bed":"/input"
  -v "/mnt/disk2/SRA3":"/output"
  -v "/mnt/disk1/ref":"/reference"
  google/deepvariant:"1.6.1" \
  /opt/deepvariant/bin/run_deepvariant \
  --model_type WGS \
  --ref /reference/Homo_sapiens_assembly38.fasta \
  --reads /output/ERR10502884.markup.bam \
  --output_vcf /output/ERR10502884.output161.vcf.gz \
  --num_shards 120 \
  --intermediate_results_dir /output/tmp3 \
  --make_examples_extra_args variant_caller="vcf_candidate_importer",proposed_variants="/output/output1.vcf.gz"

zcat /mnt/disk2/SRA3/ERR10502884.output161.vcf.gz  |grep PASS|wc -l

The ERR10502884. output 161. vcf. gz file has more than 3,000,000 lines with a PASS flag

[kishwarshafin]

@mulderdt can you please try your 1.8.0 command with --disable_small_model=true please?

[chaoxinzhang]

@mulderdt can you please try your 1.8.0 command with --disable_small_model=true please?
I'm glad to see your reply.

Yes, I did.

Try 1.8.0 command with --disable_small_model=true, The output file has more than 3,000,000 lines with a PASS flag.

What else do I need to try?

[danielecook]

@chaoxinzhang can you remove the --intermediate_results_dir /output/tmp3 \ flag when running? This directory should not be reused between commands.