Skip to content

Variant Filtration Specifications

Process Details

  • Name: VariantFiltration
  • Process UUID: bbZQLsqWRyN51ajDSKx2EAKIMIrCTs
  • Process Group: gatk4

Overview

The VariantFiltration process applies hard filters to raw variant calls (SNPs and indels) to remove low-quality variants and reduce false positives. This process implements the Broad Institute's recommended filtering criteria to distinguish high-confidence variants from artifacts and sequencing errors. The filtering approach uses various quality metrics to flag variants that fail specific thresholds while preserving them in the output file for downstream analysis decisions.

This process is implemented in Bash, which invokes GATK4 VariantFiltration commands for both SNPs and indels.

Key Functionality

  • SNP Quality Filtering: Applies six different quality filters to SNPs including variant confidence normalization (QD), strand bias detection (FS, SOR), mapping quality assessment (MQ), and allele position evaluation (MQRankSum, ReadPosRankSum)
  • Indel Quality Filtering: Applies three specialized filters for indels with adjusted thresholds appropriate for insertion/deletion variants
  • Filter Annotation: Marks variants that fail filters with specific filter names while preserving all variants in the output VCF file for transparency

Input/Output Specification

Inputs

Required Inputs

  • VCF Set

    • Description: Raw variant calls containing both SNPs and indels to be filtered
    • Format: VCF (Variant Call Format)

  • Genome Dictionary

    • Description: Reference genome dictionary file containing sequence information
    • Format: DICT (sequence dictionary)

Outputs

  • Filtered VCF Set
    • Description: Variant calls with quality filter annotations applied. Variants passing all filters are marked as "PASS" while those failing specific criteria are marked with corresponding filter names
    • Format: VCF (Variant Call Format)

Parameters & Settings

This process uses fixed filtering thresholds based on GATK best practices and does not expose user-configurable parameters in the Foundry UI.

SNP Filtering Criteria: - QD < 2.0: Variant confidence normalized by unfiltered depth - FS > 60.0: Strand bias probability (Phred-scaled) - MQ < 40.0: Root mean square mapping quality - SOR > 4.0: Symmetric odds ratio for strand bias - MQRankSum < -12.5: Mapping quality difference between reference and alternate alleles - ReadPosRankSum < -8.0: Read position bias between reference and alternate alleles

Indel Filtering Criteria: - QD < 2.0: Variant confidence normalized by unfiltered depth - FS > 200.0: Strand bias probability (adjusted threshold for indels) - SOR > 10.0: Symmetric odds ratio for strand bias (adjusted threshold for indels)

References & Resources

  • Tool Documentation: GATK4 VariantFiltration
  • Related Papers: McKenna, A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Research 20, 1297-1303 (2010). DOI: 10.1101/gr.107524.110