genome-variant-filtering

name	genome-variant-filtering
description	Filter raw variant calls using GATK VQSR, GATK hard filters, or bcftools expression-based filtering.

Skill: genome-variant-filtering

Use When

• User wants to filter a raw VCF to remove low-confidence variants.
• User needs to apply GATK best-practice hard filters or VQSR.
• User wants to use custom bcftools filter expressions.
• User needs to separate SNPs from indels for different filter strategies.

Inputs

• Required:
  • Raw VCF file (.vcf.gz).
  • Reference FASTA (for GATK operations).
• Optional:
  • Filter strategy: vqsr, hard, or bcftools (default: hard).
  • Resource VCFs for VQSR: HapMap, Omni, 1000 Genomes, dbSNP.
  • Custom filter expressions (for bcftools mode).
  • Truth sensitivity level for VQSR (default: 99.5 for SNPs, 99.0 for indels).

Workflow

1. If hard-filter: split VCF into SNPs and indels with gatk SelectVariants.
2. Apply GATK recommended hard filters:
   • SNPs: QD < 2.0, FS > 60.0, MQ < 40.0, MQRankSum < -12.5, ReadPosRankSum < -8.0.
   • Indels: QD < 2.0, FS > 200.0, ReadPosRankSum < -20.0.
3. If VQSR: run gatk VariantRecalibrator + gatk ApplyVQSR for SNPs then indels with resource annotations.
4. If bcftools: apply user-provided filter expressions via bcftools filter.
5. Merge filtered SNPs and indels back into a single VCF.
6. Index the output VCF.
7. Report: variants before and after filtering, pass rate, filter category breakdown.

Output Contract

• Filtered VCF file (.vcf.gz).
• Filter summary statistics: total variants, PASS count, filtered count by category.

Limits

• VQSR requires a sufficiently large call set to train the model (at least ~30 exomes or 1 WGS); use hard filters for small sample sets.
• Resource VCFs must match the reference genome build (e.g., GRCh38).
• GATK is required for VQSR and hard-filter modes; bcftools mode only requires bcftools.
• Hard filters are recommended when VQSR training data is insufficient.
• Common failure cases:
  • VQSR failing due to too few variants in the training set for model convergence.
  • Resource VCFs on a different genome build than the input VCF (e.g., GRCh37 vs GRCh38).
  • bcftools filter expression syntax errors causing silent empty output.