This is a tool for transforming and processing VCF files in a scalable manner based on Apache Beam using Dataflow on Google Cloud Platform.
It can be used to directly load VCF files to BigQuery supporting hundreds of thousands of files, millions of samples, and billions of records. Additionally, it provides a preprocess functionality to validate the VCF files such that the inconsistencies can be easily identified.
Please see this presentation for a high level overview of BigQuery and how to effectively use Variant Transforms and BigQuery. Please also read the blog post about how a GCP customer used Variant Transforms for breakthrough clinical data science with BigQuery.
- Setup a Google Cloud account and create a project.
- Sign up and install the Google Cloud SDK
- Enable the Genomics, Compute Engine, Cloud Storage, and Dataflow APIs
- Open the billing page for the project you have selected or created, and click Enable billing.
- Create a new BigQuery dataset by visiting the BigQuery web UI, clicking on the down arrow icon next to your project name in the navigation, and clicking on Create new dataset.
The easiest way to run the VCF to BigQuery pipeline is to use the docker image and run it with the Google Genomics Pipelines API as it has the binaries and all dependencies pre-installed.
First, set up the pipeline configurations shown below and save it as
vcf_to_bigquery.yaml
. The parameters that you need to replace are:
my_project
: This is your project name that contains the BigQuery dataset.gs://my_bucket/vcffiles/*.vcf
: A location in Google Cloud Storage where the VCF file are stored. You may specify a single file or provide a pattern to load multiple files at once. Please refer to the Variant Merging documentation if you want to merge samples across files. The pipeline supports gzip, bzip, and uncompressed VCF formats. However, it runs slower for compressed files as they cannot be sharded.my_bigquery_dataset
: Your BigQuery dataset to store the output.my_bigquery_table
: This can be any ID you like (e.g. vcf_test).gs://my_bucket/staging
andgs://my_bucket/temp
: These can be any folder in Google Cloud Storage that your project has write access to. These are used to store temporary files needed for running the pipeline.
name: vcf-to-bigquery-pipeline
docker:
imageName: gcr.io/gcp-variant-transforms/gcp-variant-transforms
cmd: |
./opt/gcp_variant_transforms/bin/vcf_to_bq \
--project my_project \
--input_pattern gs://my_bucket/vcffiles/*.vcf \
--output_table my_project:my_bigquery_dataset.my_bigquery_table \
--staging_location gs://my_bucket/staging \
--temp_location gs://my_bucket/temp \
--job_name vcf-to-bigquery \
--runner DataflowRunner
Next, run the following command to launch the pipeline. Replace my_project
with your project name, gs://my_bucket/temp/runner_logs
with a Cloud Storage
folder to store the logs from the pipeline.
gcloud alpha genomics pipelines run \
--project my_project \
--pipeline-file vcf_to_bigquery.yaml \
--logging gs://my_bucket/temp/runner_logs \
--zones us-west1-b \
--service-account-scopes https://www.googleapis.com/auth/bigquery
Please note the operation ID returned by the above script. You can track the status of your operation by running:
gcloud alpha genomics operations describe <operation-id>
The returned data will have done: true
when the operation is done.
A detailed description of the Operation resource can be found in the
API documentation
The underlying pipeline uses Cloud Dataflow. You can navigate to the Dataflow Console, to see more detailed view of the pipeline (e.g. number of records being processed, number of workers, more detailed error logs).
In addition to using the docker image, you may run the pipeline directly from source. First install git, python, pip, and virtualenv:
sudo apt-get install -y git python-pip python-dev build-essential
sudo pip install --upgrade pip
sudo pip install --upgrade virtualenv
Run virtualenv, clone the repo, and install pip packages:
virtualenv venv
source venv/bin/activate
git clone https://github.com/googlegenomics/gcp-variant-transforms.git
cd gcp-variant-transforms
pip install --upgrade .
You may use the DirectRunner (aka local runner) for small (e.g. 10,000 records) files or DataflowRunner for larger files. Files should be stored on Google Cloud Storage if using Dataflow, but may be stored locally for DirectRunner.
Example command for DirectRunner:
python -m gcp_variant_transforms.vcf_to_bq \
--input_pattern gcp_variant_transforms/testing/data/vcf/valid-4.0.vcf \
--output_table projectname:bigquerydataset.tablename
Example command for DataflowRunner:
python -m gcp_variant_transforms.vcf_to_bq \
--input_pattern gs://my_bucket/vcffiles/*.vcf \
--output_table my_project:my_bigquery_dataset.my_bigquery_table \
--project my_project \
--staging_location gs://my_bucket/staging \
--temp_location gs://my_bucket/temp \
--job_name vcf-to-bigquery \
--setup_file ./setup.py \
--runner DataflowRunner
The VCF files preprocessor is used for validating the datasets such that the inconsistencies can be easily identified. It can be used as a standalone validator to check the validity of the VCF files, or as a helper tool for VCF to BigQuery pipeline. Please refer to VCF files preprocessor for more details.
You may need to constrain Cloud Dataflow job processing to a specific geographic region in support of your project’s security and compliance needs. See Running in particular zone/region doc.