Skip to content

Latest commit

 

History

History
139 lines (122 loc) · 6.25 KB

bigquery_to_vcf.md

File metadata and controls

139 lines (122 loc) · 6.25 KB
Raw

BigQuery to VCF

The BigQuery to VCF pipeline is used for loading a BigQuery table to one VCF file. You may also customize the pipeline to load a subset of samples and/or genomic regions.

Running BQ to VCF

Similar to running the VCF to BigQuery pipeline, the BigQuery to VCF pipeline can also be run using docker or directly from the source.

Using docker

Run the script below and replace the following parameters:

  • Dataflow's required inputs:
    • GOOGLE_CLOUD_PROJECT: This is your Google Cloud project ID where the job should run.
    • GOOGLE_CLOUD_REGION: You must choose a geographic region for Cloud Dataflow to process your data, for example: us-west1. For more information please refer to Setting Regions.
    • TEMP_LOCATION: This can be any folder in Google Cloud Storage that your project has write access to. It's used to store temporary files and logs from the pipeline.
  • INPUT_TABLE: BigQuery table that will be loaded to VCF. It must be in the format of GOOGLE_CLOUD_PROJECT:DATASET.TABLE.
  • OUTPUT_FILE: The full path of the output VCF file. This can be a local path if you use DirectRunner (for very small VCF files) but must be a path in Google Cloud Storage if using DataflowRunner.
#!/bin/bash
# Parameters to replace:
GOOGLE_CLOUD_PROJECT=GOOGLE_CLOUD_PROJECT
GOOGLE_CLOUD_REGION=GOOGLE_CLOUD_REGION
TEMP_LOCATION=gs://BUCKET/temp
INPUT_TABLE=GOOGLE_CLOUD_PROJECT:DATASET.TABLE
OUTPUT_FILE=gs://BUCKET/loaded_file.vcf

COMMAND="bq_to_vcf \
  --input_table ${INPUT_TABLE} \
  --output_file ${OUTPUT_FILE} \
  --job_name bq-to-vcf \
  --runner DataflowRunner"

docker run -v ~/.config:/root/.config \
  gcr.io/cloud-lifesciences/gcp-variant-transforms \
  --project "${GOOGLE_CLOUD_PROJECT}" \
  --region "${GOOGLE_CLOUD_REGION}" \
  --temp_location "${TEMP_LOCATION}" \
  "${COMMAND}"

In addition, the following optional flags can be specified in the COMMAND argument in the above script:

  • --representative_header_file: If provided, meta-information from the provided file (e.g., INFO, FORMAT, FILTER, etc) will be added into the output_file. Otherwise, the meta-information is inferred from the BigQuery schema on a "best effort" basis (e.g., any repeated INFO field will have Number=.). It is recommended to provide this file to specify the most accurate and complete meta-information in the VCF file.
  • --genomic_regions: A list of genomic regions (separated by a space) to load from BigQuery. The format of each genomic region should be REFERENCE_NAME:START_POSITION-END_POSITION or REFERENCE_NAME if the full chromosome is requested. Only variants matching at least one of these regions will be loaded. For example, --genomic_regions chr1 chr2:1000-2000 will load all variants in chr1 and all variants in chr2 with start_position in [1000,2000) from BigQuery. If this flag is not specified, all variants will be loaded.
  • --sample_names: A list of sample names (separated by a space). Only variants for these calls will be loaded from BigQuery. If this parameter is not specified, all calls will be loaded.
  • --allow_incompatible_schema: If representative_header_file is not provided, the meta-information is inferred from the BigQuery schema. There are some reserved fields based on VCF 4.3 spec. If the inferred definition is not the same as the reserved definition, an error will be raised and the pipeline will fail. By setting this flag to true, the incompatibilities between BigQuery schema and the reserved fields will not raise errors. Instead, the VCF meta-information are inferred from the schema without validation.
  • --preserve_sample_order: By default, samples names in the output VCF file are generated in ascending order. If set to true, the order of sample names will be the same as the BigQuery table, but it requires all extracted variants to have the same sample ordering (usually true for tables from single VCF file import).
  • --number_of_bases_per_shard: The maximum number of base pairs per chromosome to include in a shard. A shard is a collection of data within a contiguous region of the genome that can be efficiently sorted in memory. This flag is set to 1,000,000 by default, which should work for most datasets. You may change this flag if you have a dataset that is very dense and variants in each shard cannot be sorted in memory.

The pipeline can be optimized depending on the size of the output VCF file:

  • For small VCF files (e.g. a few megabytes), you may use --runner DirectRunner to speed up the pipeline as it avoids creating a Dataflow pipeline.
  • For large VCF files (e.g. >100GB), using Cloud Dataflow Shuffle can speed up the pipeline, by adding the parameter --experiments shuffle_mode=service to the COMMAND argument.
  • For medium sized VCF files, you can use the above script as is.

Running from github

In addition to using the docker image, you may run the pipeline directly from source.

Example command for DirectRunner:

python -m gcp_variant_transforms.bq_to_vcf \
  --input_table bigquery-public-data:human_genome_variants.1000_genomes_phase_3_variants_20150220 \
  --output_file gs://BUCKET/loaded_file.vcf \
  --job_name bq-to-vcf-direct-runner \
  --genomic_regions 1:124852-124853 \
  --sample_names HG00099 HG00105 \
  --project "${GOOGLE_CLOUD_PROJECT}" \
  --temp_location "${TEMP_LOCATION}"

Example command for DataflowRunner:

python -m gcp_variant_transforms.bq_to_vcf \
  --input_table bigquery-public-data:human_genome_variants.1000_genomes_phase_3_variants_20150220 \
  --output_file gs://BUCKET/loaded_file.vcf \
  --job_name bq-to-vcf \
  --genomic_regions 1:124852-124853 \
  --representative_header_file gs://BUCKET/representative_header_file.vcf \
  --sample_names HG00099 HG00105 \
  --setup_file ./setup.py \
  --runner DataflowRunner \
  --project "${GOOGLE_CLOUD_PROJECT}" \
  --region "${GOOGLE_CLOUD_REGION}" \
  --temp_location "${TEMP_LOCATION}"