Variant Normalization

Left-align indels and split multiallelic sites using bcftools norm.

Why Normalize?

The same variant can be represented multiple ways:

# Same deletion, different representations
chr1  100  ATCG  A      (right-aligned)
chr1  100  ATC   A      (left-aligned, normalized)
chr1  101  TCG   T      (different position)

Normalization ensures consistent representation for:

• Comparing variants from different callers
• Database lookups (dbSNP, ClinVar)
• Merging VCF files

bcftools norm

Left-Align Indels

bcftools norm -f reference.fa input.vcf.gz -Oz -o normalized.vcf.gz

Requires reference FASTA to determine left-most representation.

Check for Normalization Issues

bcftools norm -f reference.fa -c s input.vcf.gz > /dev/null
# Reports REF allele mismatches

Check modes (-c):

• w - Warn on mismatch (default)
• e - Error on mismatch
• x - Exclude mismatches
• s - Set correct REF from reference

Multiallelic Sites

Split Multiallelic to Biallelic

bcftools norm -m-any input.vcf.gz -Oz -o split.vcf.gz

Before:

chr1  100  .  A  G,T  30  PASS  .  GT  1/2

After:

chr1  100  .  A  G  30  PASS  .  GT  1/0
chr1  100  .  A  T  30  PASS  .  GT  0/1

Split SNPs Only

bcftools norm -m-snps input.vcf.gz -Oz -o split_snps.vcf.gz

Split Indels Only

bcftools norm -m-indels input.vcf.gz -Oz -o split_indels.vcf.gz

Join Biallelic to Multiallelic

bcftools norm -m+any input.vcf.gz -Oz -o merged.vcf.gz

Split Options

Combined Normalization

Standard Normalization Pipeline

bcftools norm -f reference.fa -m-any input.vcf.gz -Oz -o normalized.vcf.gz
bcftools index normalized.vcf.gz

This:

1. Left-aligns indels
2. Splits multiallelic sites

Remove Duplicates After Splitting

bcftools norm -f reference.fa -m-any -d exact input.vcf.gz -Oz -o normalized.vcf.gz

Duplicate removal options (-d):

• exact - Remove exact duplicates
• snps - Remove duplicate SNPs
• indels - Remove duplicate indels
• both - Remove duplicate SNPs and indels
• all - Remove all duplicates
• none - Keep duplicates (default)

Fixing Reference Alleles

Fix Mismatches from Reference

bcftools norm -f reference.fa -c s input.vcf.gz -Oz -o fixed.vcf.gz

This sets REF alleles to match the reference genome.

Exclude Mismatches

bcftools norm -f reference.fa -c x input.vcf.gz -Oz -o clean.vcf.gz

Removes variants where REF doesn't match reference.

Atomize Complex Variants

Split MNPs to SNPs

bcftools norm --atomize input.vcf.gz -Oz -o atomized.vcf.gz

Before:

chr1  100  .  ATG  GCA  30  PASS

After:

chr1  100  .  A  G  30  PASS
chr1  101  .  T  C  30  PASS
chr1  102  .  G  A  30  PASS

Atomize and Left-Align

bcftools norm -f reference.fa --atomize input.vcf.gz -Oz -o atomized.vcf.gz

Old to New Format

Update VCF Version

bcftools norm --old-rec-tag OLD input.vcf.gz -Oz -o updated.vcf.gz

Tags original record for reference.

Common Workflows

Before Comparing Callers

# Normalize both VCFs the same way
for vcf in caller1.vcf.gz caller2.vcf.gz; do
    base=$(basename "$vcf" .vcf.gz)
    bcftools norm -f reference.fa -m-any "$vcf" -Oz -o "${base}.norm.vcf.gz"
    bcftools index "${base}.norm.vcf.gz"
done

# Now compare
bcftools isec -p comparison caller1.norm.vcf.gz caller2.norm.vcf.gz

Before Database Annotation

bcftools norm -f reference.fa -m-any variants.vcf.gz -Oz -o normalized.vcf.gz
bcftools index normalized.vcf.gz
# Now annotate against dbSNP, ClinVar, etc.

Prepare for GWAS

bcftools norm -f reference.fa -m-any -d exact input.vcf.gz | \
    bcftools view -v snps -Oz -o gwas_ready.vcf.gz
bcftools index gwas_ready.vcf.gz

cyvcf2 Normalization Check

Check if Variants Need Normalization

from cyvcf2 import VCF

def needs_normalization(variant):
    # Check for multiallelic
    if len(variant.ALT) > 1:
        return True

    # Check for complex variants (potential MNPs)
    ref, alt = variant.REF, variant.ALT[0]
    if len(ref) > 1 and len(alt) > 1 and len(ref) == len(alt):
        return True

    return False

count = 0
for variant in VCF('input.vcf.gz'):
    if needs_normalization(variant):
        count += 1

print(f'Variants needing normalization: {count}')

Count Multiallelic Sites

from cyvcf2 import VCF

multiallelic = 0
total = 0

for variant in VCF('input.vcf.gz'):
    total += 1
    if len(variant.ALT) > 1:
        multiallelic += 1

print(f'Total variants: {total}')
print(f'Multiallelic sites: {multiallelic}')
print(f'Percentage: {multiallelic/total*100:.1f}%')

Quick Reference

Common Errors

Related Skills

• variant-calling - Generate VCF files
• filtering-best-practices - Filter after normalization
• vcf-manipulation - Compare normalized VCFs
• variant-annotation - Annotate normalized variants