Variant Call Format (VCF)

Every VCF file begins with a structured header section prefixed by ##. This metadata defines the file's provenance, the reference genome used, and the semantics of all downstream fields.

• INFO fields: Define per-variant annotations like allele frequency (AF) and quality depth (QD)
• FORMAT fields: Specify per-sample data fields such as genotype (GT) and read depth (DP)
• FILTER entries: Declare the meaning of filter flags like PASS or LowQual
• Contig lines: Explicitly list reference sequence names and lengths for validation

The GT FORMAT field encodes diploid or polyploid genotypes using a compact numeric representation. Alleles are indexed from 0 (reference) upward, separated by / (unphased) or | (phased).

• 0/0: Homozygous reference — both alleles match the reference
• 0/1: Heterozygous — one reference and one alternate allele
• 1/1: Homozygous alternate — both alleles are the variant
• 0|1: Phased heterozygous — the first haplotype carries the reference, the second carries the variant
• ./.: Missing genotype — no call could be made at this position

A single VCF record can represent multiple alternate alleles at the same genomic position, avoiding redundant entries. The ALT column uses comma-separated values, and genotype indices reference these alleles numerically.

• A site with two alternate alleles lists them as A,T in the ALT column
• Genotype 1/2 indicates the sample carries one copy of the first alternate allele and one copy of the second
• Allele-specific annotations use the AD (allele depth) FORMAT field to report read counts per allele
• Tools like bcftools norm can decompose multi-allelic records into biallelic representations for compatibility

The QUAL column stores a Phred-scaled probability that a variant exists at this site, calculated as -10 * log10(prob(call is wrong)). A QUAL of 30 indicates a 1 in 1000 chance of error.

• FILTER column: Records whether a variant passed all quality filters (PASS) or failed specific criteria
• VQSR integration: Variant Quality Score Recalibration uses Gaussian mixture models to assign well-calibrated quality scores based on known truth sets
• INFO/DP: Total read depth at the site, essential for distinguishing true variants from sequencing noise
• FORMAT/GQ: Genotype Quality, a Phred-scaled confidence score for the assigned genotype call

VCF v4.3 and later support encoding large-scale genomic rearrangements using symbolic alleles in the ALT column, marked with angle brackets.

• <DEL>: Deletion — a segment of DNA is missing relative to the reference
• <DUP>: Duplication — a segment is tandemly duplicated
• <INV>: Inversion — a segment is reversed in orientation
• <INS>: Insertion — novel sequence inserted at the breakpoint
• INFO/SVLEN: The length of the structural variant in base pairs
• INFO/END: The end coordinate, stored in the INFO column for symbolic alleles
• BND format: Breakend notation uses ] and [ brackets to specify translocation partner coordinates

Large VCF files are routinely compressed with bgzip (block gzip) and indexed with tabix to enable fast random access without full decompression.

• .vcf.gz: The standard compressed VCF extension using block gzip compression
• .tbi: The tabix index file enabling region-based queries via genomic coordinates
• BCF format: A binary variant call format that stores the same information as VCF but in a more compact, machine-readable binary representation
• Streaming queries: Tools like bcftools view -r chr1:1000000-2000000 use the index to extract specific genomic regions in milliseconds

A Gaussian mixture model that assigns a well-calibrated error probability to each variant in a VCF. It uses known truth resources like HapMap and 1000 Genomes as positive training examples and the most confident novel calls as negative examples.

• Generates the VQSLOD score stored in the VCF INFO field
• Replaces hard filters with a continuous, probabilistic quality metric
• Critical for false discovery rate control in production pipelines

While standard VCF rows capture single nucleotide variants and small indels, structural variants (SVs) require specialized VCF records with BND (breakend) ALT alleles and INFO fields like SVLEN, SVTYPE, and CIPOS.

• Long-read sequencing dramatically improves SV breakpoint resolution
• Local reassembly resolves complex rearrangements before VCF encoding
• VCF v4.4+ includes enhanced symbolic allele representations for large events

VCF genotypes can be unphased (0/1) or phased (0|1) to indicate which alleles reside on the same parental chromosome. Read-backed phasing uses paired-end reads spanning multiple heterozygous sites.

• Joint genotyping simultaneously analyzes multiple samples to improve rare variant accuracy
• Phased VCFs are essential for imputation and population genetics
• The PS (phase set) INFO tag groups variants into haplotype blocks