Variant Call Format (VCF) is a tab-delimited text format used to describe variation in DNA sequences relative to a reference genome. A VCF file records the locations and identities of sequence differences and can also contain genotypes for one or more samples, quality estimates, filtering results, and biological annotations. It is widely used to exchange results from DNA sequencing analyses, particularly single-nucleotide variants, short insertions and deletions, and larger structural variants.
VCF represents interpreted variant calls rather than the underlying sequencing reads. Its reference sequence, metadata, and field definitions are essential to interpreting its contents correctly.
Development and scope
VCF originated in the 1000 Genomes Project as a common representation for variation discovered across many human genomes. A description of the format and its associated VCFtools software was published in 2011. The format subsequently became widely used beyond human genomics, including research on other animals, plants, and microorganisms.
The specification is maintained through the HTS specifications project, alongside related high-throughput sequencing formats. Successive versions have clarified field definitions and extended support for complex variation. A file declares its specification version in its first line; support for particular features depends on both that version and the software reading the file.
Although originally associated with variant sites, VCF can also represent reference observations and intervals of reference confidence through conventions such as genomic VCF.
File organization
A VCF file has a header followed by data records. Each record normally describes a variant site or a related genomic event.
Metadata and column header
Metadata lines begin with ##. The first line declares the format version, for example:
##fileformat=VCFv4.3
Other metadata can identify the reference assembly, sequence contigs, analysis software, filtering criteria, and annotation definitions. Structured declarations define fields used in the INFO and FORMAT columns, including their identifiers, types, numbers of values, and descriptions.
A final header line begins with #CHROM. It names eight fixed columns, followed, when sample data are present, by FORMAT and one column for each sample.
Fixed columns
| Column | Meaning |
|---|---|
CHROM |
Reference sequence identifier, such as a chromosome or contig name. |
POS |
One-based position of the first reference base represented by the record. |
ID |
One or more identifiers for the variant, or . if unavailable. |
REF |
Reference allele sequence. |
ALT |
Comma-separated alternate alleles, or . when no alternate allele is specified. |
QUAL |
Phred-scaled quality associated with the assertion of variation represented by the record. |
FILTER |
Filter results: PASS when all applied filters passed, named failed filters otherwise, or . when filters have not been applied. |
INFO |
Semicolon-separated site-level annotations. |
CHROM may refer to a chromosome, an unplaced contig, or another sequence in the reference assembly. An allele is interpreted relative to that particular reference, not independently of it.
The optional FORMAT column lists colon-separated sample fields, such as GT:DP:GQ. Each subsequent sample column supplies values in that order.
Alleles, coordinates, and genotypes
Reference-relative representation
For a single-nucleotide variant, REF and ALT can each contain one base. A record with POS=100, REF=A, and ALT=G describes a G alternate allele at a position where the reference has A.
Short insertions and deletions generally include a shared anchoring base so that neither allele sequence is empty. For example:
POS=100,REF=A,ALT=ATrepresents insertion of T after the reference base at position 100.POS=100,REF=AT,ALT=Arepresents deletion of the T at position 101.
Thus, a deletion record's position may identify its anchor rather than the first deleted base. Special anchoring rules apply to events at the beginning of a reference sequence.
VCF coordinates are one-based. Confusing them with zero-based coordinate systems used by some other genomic formats can shift an interpretation by one base.
Genotype encoding
The GT field represents alleles by numerical indexes:
0denotes the reference allele.1denotes the first allele listed inALT.2denotes the second alternate allele, and so forth..denotes a missing allele.
For a diploid sample, 0/1 indicates one reference allele and one copy of the first alternate allele; 1/1 indicates two copies of that alternate allele. With ALT=C,G, 1/2 represents one C allele and one G allele.
A slash indicates that the genotype is unphased. A vertical bar, as in 0|1, indicates phase information: allele order is meaningful within the applicable phase context. Phase-set annotations can identify which records have a known phase relationship. A phased separator alone does not establish chromosome-wide haplotypes.
Genotypes can contain other numbers of allele entries to accommodate haploid organisms, sex chromosomes, and polyploidy.
Example
The following example contains one biallelic site and one sample. Fields within the header row and data row are separated by tabs.
##fileformat=VCFv4.3
##contig=<ID=chr1,length=1000000>
##FORMAT=<ID=GT,Number=1,Type=String,Description="Genotype">
##FORMAT=<ID=DP,Number=1,Type=Integer,Description="Read depth">
#CHROM POS ID REF ALT QUAL FILTER INFO FORMAT SAMPLE_A
chr1 100 . A G 50 PASS . GT:DP 0/1:18
The record describes an A-to-G variant at position 100 on chr1. SAMPLE_A has an unphased heterozygous genotype, and its reported read depth is 18. The site has quality 50 and passed the filters that were applied. No variant identifier or site-level annotations are supplied.
This record does not, by itself, establish a biological effect, inheritance pattern, or association with a phenotype.
Annotations and quality measures
VCF distinguishes site-level annotations in INFO from sample-level annotations identified by FORMAT.
Common fields include:
AC: alternate-allele counts among the genotypes considered.AN: total number of alleles counted in called genotypes.AF: alternate-allele frequency.DP: read depth, with scope and counting details dependent on the field definition and producing software.AD: sample-level depths for the reference and alternate alleles.GQ: sample-level genotype quality.PL: Phred-scaled genotype likelihoods.
Quality values commonly use the Phred transformation:
where is the relevant error probability. A value of 30 corresponds to an error probability of under that interpretation. Site quality, genotype quality, and genotype likelihoods describe different quantities and are not interchangeable. Their empirical calibration also depends on the calling model and data.
Header declarations specify annotation cardinality. For example, Number=A means one value per alternate allele, Number=R one per reference and alternate allele, and Number=G one per possible genotype under the applicable ploidy. These distinctions matter when splitting or merging records.
Structural variants and symbolic alleles
Larger changes, collectively called structural variation, may be represented by explicit sequences or symbolic alleles such as <DEL>, <DUP>, and <INV>. Additional fields can describe variant type, length, endpoints, and uncertainty in breakpoint positions.
VCF also defines breakend notation for rearrangements. Brackets and reference coordinates describe how an end of one sequence connects to another. Depending on the event, multiple linked records may be needed.
Two special alternate-allele forms have distinct meanings:
*denotes an allele absent at the current position because of a deletion represented by another record.<NON_REF>, used in genomic VCF workflows, represents possible non-reference alleles not explicitly listed.
Neither should be interpreted as an ordinary DNA sequence.
Compression, indexing, and related formats
VCF is human-readable but can become large in studies containing many samples. It is commonly compressed with BGZF, a block-based compression format that supports efficient access to portions of a compressed file. Coordinate-sorted compressed VCF files can be indexed using tools such as Tabix, allowing retrieval of records overlapping a genomic region without reading the entire file.
Binary Call Format (BCF) is a binary representation of the VCF data model. It reduces text-parsing overhead and is widely used for computational processing.
Genomic VCF (gVCF) is a VCF-based representation that includes information about non-variant positions or blocks of positions, as well as candidate variants. It is commonly used as an intermediate in joint genotyping. The absence of a site from an ordinary variant-only VCF, by contrast, does not establish that the sample confidently matches the reference there.
Software including VCFtools and BCFtools supports operations such as filtering, querying, merging, annotation, and conversion.
Applications and limitations
VCF provides a common exchange format for population genetics, genome-wide association studies, variant annotation, and comparison of sequencing analyses. Its support for multiple samples permits site-level summaries alongside individual genotypes.
Interoperability nevertheless depends on more than syntactic validity:
- Reference compatibility: Coordinates and alleles depend on a specific assembly. Converting between assemblies requires coordinate mapping and validation of allele sequences.
- Equivalent representations: The same insertion, deletion, or complex event can have different valid representations, especially in repetitive sequence. Trimming redundant bases and left-aligning indels can improve comparisons, but do not resolve every complex equivalence.
- Multiallelic transformations: Splitting a record into separate biallelic records requires appropriate updates to genotypes, allele-indexed annotations, and genotype-indexed values.
- Missingness: A missing genotype is not a homozygous-reference genotype. An omitted site may reflect filtering, insufficient evidence, or the output conventions of the pipeline.
- Annotation consistency: Custom fields, differing depth calculations, and caller-specific filters can make superficially similar files difficult to compare.
- Representation limits: Complex rearrangements and variation relative to a pangenome may be cumbersome to express through records anchored to one linear reference.
VCF standardizes how results are represented, not how they are inferred. A valid file can still contain incorrect calls, poorly calibrated quality scores, or incomplete annotations.