Input VCF requirements

Before submitting a job, ensure that your input VCF:

• Is compressed with bgzip
• Is called against the hg38 reference.  Please check your VCF header line to confirm before proceeding, for example: ##contig=<ID=chr1,length=248956422,assembly=38>
  • If your VCF chromosomes are encoded with numbers only (e.g., "22" instead of "chr22"), this is one indicator that your VCF was called against a different reference, and you will need to liftover your VCF to hg38 (learn more)
  • Please ensure that each contig header has a length attribute. A way to fix this can be found in the Helpful Tips section below.
• Is VCF version 4.2 or higher; VCFtools can help update your VCF version
• Contains at least one autosome (chr1-chr22)
  • The service now accepts chromosome-specific files and will not fail if an autosome is missing from the input data. However, every run will be charged quota as though the full genome was processed.
• Has a sample count that matches or is less than your remaining quota (See Quota)
• Has variants sorted by genomic position
• Is an array VCF (has fewer than 3 million sites)

The service will check that your VCF meets the above requirements and will return a relevant error message explaining which requirements are not met.

For a deeper dive into the pipeline, please see Pipeline Overview.

Helpful Tips:

Since only the GT field is required for imputation, you can reduce the size of large files by removing INFO annotations and extra format fields to speed up processing. We recommend bcftools for this:

bcftools annotate -x INFO,^FORMAT/GT file.vcf

If the input VCF contains contig headers but is missing the length attribute, this can be fixed using UpdateVCFSequenceDictionary tool by running the following command:

gatk UpdateVCFSequenceDictionary \
  -V input.vcf.gz \
  --sequence-dictionary local_path_to_reference.dict \
  --output output_with_lengths.vcf.gz

You can use the reference dictionary available at gs://gcp-public-data--broad-references/hg38/v0/Homo_sapiens_assembly38.dict (a Google-owned public bucket). Download this file to a local path and provide that path to the command above.

If you still experience errors, check our Troubleshooting page.