haploid contigs and PAR region options for DeepTrio

[prasundutta87]

Hi,

How do I use --haploid_contigs="chrX,chrY" and --par_regions_bed parameters in DeepTrio? This option is not present in DeepTrio for sex chromosomes.

Regards,
Prasun

[pichuan]

Hi @prasundutta87 , given that --haploid_contigs and --par_regions_bed are arguments for postprocess_variants, you can try using:

`--postprocess_variants_extra_args="--haploid_contigs=yourValue,--par_regions_bed=yourValue""

If you're using Docker, you likely need to make sure the par_regions_bed path are something that the run in Docker can access.

[pichuan]

And actually, if it's on the child sample or parent samples, you should use postprocess_variants_child_extra_args, postprocess_variants_parent1_extra_args, postprocess_variants_parent2_extra_args.

[prasundutta87]

Hi @pichuan , thanks a lot for the quick reply. I am running the "run_deeptrio" script from DeepTrio docker (v1.6.1). As I have trios, I am running it all together in one command as described in https://github.com/google/deepvariant/blob/r1.6.1/docs/deeptrio-pacbio-case-study.md. I will then be merging the single sample gvcfs using GLnexus as described there.

Since, I am using trios, will just using `--postprocess_variants_extra_args="--haploid_contigs=yourValue,--par_regions_bed=yourValue"" help or I need to use the second suggestion? I have attached the example command I am running for your reference:

apptainer run \ SNV_analysis_DeepTrio/images/deepvariant_deeptrio-1.6.1.sif \ run_deeptrio \ --model_type=PACBIO \ --ref=hg38_reference/GCA_000001405.15_GRCh38_no_alt_plus_hs38d1_analysis_set.fna \ --reads_child=BAMS/Proband_sorted.bam \ --reads_parent1=BAMS/Parent-1_sorted.bam \ --reads_parent2=BAMS/Parent-2_sorted.bam \ --output_vcf_child output/Proband.output.vcf.gz \ --output_vcf_parent1 output/Parent-1.output.vcf.gz \ --output_vcf_parent2 output/Parent-2.output.vcf.gz \ --sample_name_child 'Proband' \ --sample_name_parent1 'Parent-1' \ --sample_name_parent2 'Parent-2' \ --intermediate_results_dir output/intermediate_results_dir \ --output_gvcf_child output/Proband.g.vcf.gz \ --output_gvcf_parent1 output/Parent-1.g.vcf.gz \ --output_gvcf_parent2 output/Parent-2.g.vcf.gz \ --regions chr21 \ --num_shards=5

[pichuan]

Hi @prasundutta87 ,

Right, given that you're running on trio, you'll for sure want to specify different things for the parent1 and parent2: only the dad would be using haploid_contigs par_regions_bed flags. The mom would have two copies of chrX, so you should not specify chrX to be haploid.

And, for the child: you should only use those flags if the child is male.

[prasundutta87]

Sounds good, thanks a lot @pichuan . Just an additional query. If I want to joint call SNVs/indels in a set of multiple trios, which pipeline is recommended? Deepvariant or DeepTrio. My understanding is that if we joint call multiple samples together, the cohort allele frequency could be used to filter rare variants. If we use DeepTrio, AF calculation will be restricted to only the trio or family. What is the best practice here in terms of Deepvariant?

[AndrewCarroll]

Hi @prasundutta87

Both DeepVariant and DeepTrio can produce gVCFs, so you can joint call in a similar manner. In both cases you should use the DeepVariant_unfiltered preset for GLnexus, because there is family structure present which the other filtering presets wouldn't know about.

Because you can joint genotype multiple trio gVCFs from either DeepVariant or DeepTrio in the same way, I would use DeepTrio to produce the gVCFs, take all of the gVCFs and run them together through glnexus, and then you can still use allele frequency information.

To be clear, the unfiltered preset looks like this:

sudo docker run \
  -v "${PWD}/output":"/output" \
  quay.io/mlin/glnexus:v1.2.7 \
  /usr/local/bin/glnexus_cli \
  --config DeepVariant_unfiltered \
  /output/child_trio_1.g.vcf.gz\
  /output/parent1_trio_1.g.vcf.gz \
  /output/parent2_trio_1.g.vcf.gz \
  /output/child_trio_2g.vcf.gz\
  /output/parent1_trio_2.g.vcf.gz \
  /output/parent2_trio_2.g.vcf.gz \
  | sudo docker run -i google/deepvariant:deeptrio-"${BIN_VERSION}" \
    bcftools view - \
  | sudo docker run -i google/deepvariant:deeptrio-"${BIN_VERSION}" \
    bgzip -c > output/trio_cohort_merged.vcf.gz

[prasundutta87]

Hi @AndrewCarroll ,..sorry for the late reply, but aren't the inherent models different for DeepTrio and DeepVariant?

[AndrewCarroll]

Hi @prasundutta87

Yes, they are different models. You would presumably want to use either one or the other (all DeepVariant or all DeepTrio). You can merge either with glnexus though.

[prasundutta87]

Sounds good, thanks a lot, @AndrewCarroll

[zihhuafang]

Hi @pichuan,

I tried to use postprocess_variants_child_extra_args, postprocess_variants_parent1_extra_args, postprocess_variants_parent2_extra_args. with deeptrio by running the docker: docker://google/deepvariant:deeptrio-1.6.1

here's my example code where parent1 is the father in the trio

            /opt/deepvariant/bin/deeptrio/run_deeptrio \
                --model_type "WGS" \
                --ref {params.ref_genome} \
                --reads_child {input.child_cram} \
                --reads_parent1 {input.dad_cram} \
                --reads_parent2 {input.mom_cram}  \
                --output_vcf_child {output}/{params.child_name}.vcf.gz \
                --output_vcf_parent1 {output}/{params.dad_name}.vcf.gz \
                --output_vcf_parent2 {output}/{params.mom_name}.vcf.gz \
                --sample_name_child {params.child_name} \
                --sample_name_parent1 {params.dad_name} \
                --sample_name_parent2 {params.mom_name} \
                --num_shards {threads} \
                --intermediate_results_dir deeptrio_tmp/{wildcards.family} \
                --postprocess_variants_parent1_extra_args="--haploid_contigs="chrX,chrY",--par_regions_bed={input.PAR}" \
                --output_gvcf_child {output}/{params.child_name}.g.vcf.gz \
                --output_gvcf_parent1 {output}/{params.dad_name}.g.vcf.gz \
                --output_gvcf_parent2 {output}/{params.mom_name}.g.vcf.gz \
                --novcf_stats_report

but I got the following error:

FATAL Flags parsing error: Unknown command line flag 'postprocess_variants_parent1_extra_args'. Did you mean: postprocess_variants_extra_args ?
Pass --helpshort or --helpfull to see help on flags.

checking --helpfull, it seems only --postprocess_variants_extra_args is available.
Did I misunderstand the use of postprocess_variants_parent1_extra_args?

Thank you.

[pichuan]

Hi @zihhuafang ,
The flag was added after 1.6.1 was released. Sorry about that.

Note that run_deeptrio is a wrapper script that just runs the underlying binaries. So, for now you can run run_deeptrio with the --dry_run flag, which will print out all the commands it is going to run with each of the steps.

From there, you can modify to make sure the postprocess_variants has the correct flag for the corresponding samples. Sorry for the inconvenience for now. The postprocess_variants_parent1_extra_args flag will be available in the next release.

[zihhuafang]

Hi @pichuan,

Thanks a lot for the explanation.
Will proceed as you suggested!

[NourMarzouka]

If I generated a VCF file for a trio (with a father using deeptrio) or a solo male (using deepvariant) without chrX,chrY option. Can I fix the VCF after the run is finished?
As a suggestion, it would be nice if the algorithm takes care of this automatically :)