Variant Allele Frequency and Genotype Quality

  • Variant allele frequency and genotype quality are important measures in bioinformatics used to evaluate genetic variation and the reliability of genotype calls. Variant allele frequency describes how frequently an alternate allele is observed among the chromosomes or sequencing reads examined, while genotype quality indicates the confidence assigned to a particular sample’s genotype at a variant site. These measures are widely used in germline and somatic variant analysis, population genetics, clinical genomics, cancer research, evolutionary studies, and agricultural genomics. Although they are related to variant interpretation, they describe different properties and should be considered together with sequencing depth, mapping quality, allele balance, and other quality metrics.
  • A genetic variant is a difference between an individual’s DNA sequence and a reference sequence or another defined comparison sequence. At a biallelic variant site, two alleles are considered: the reference allele and one alternate allele. For example, if the reference nucleotide is A and the alternate nucleotide is G, an individual may carry the genotypes AA, AG, or GG in a typical diploid organism. The genotype indicates which alleles are present in the individual, whereas allele frequency describes how common an allele is in a group of chromosomes or observations. The distinction is important because a genotype belongs to an individual, while an allele frequency is calculated across a defined collection of observations or individuals.
  • The term variant allele frequency (VAF) is commonly used in sequencing analysis to describe the proportion of allele observations supporting an alternate allele at a genomic position. It is calculated by dividing the number of reads supporting the alternate allele by the total number of reads supporting the reference and alternate alleles. For a simple biallelic site, the calculation is:

[
VAF = \frac{\text{Alternate allele read count}}{\text{Reference allele read count}+\text{Alternate allele read count}}
]

  • For example, if 40 reads support the reference allele and 20 reads support the alternate allele, the VAF is 20 divided by 60, or approximately 0.33. This indicates that one-third of the informative reads support the alternate allele. The calculation should use appropriate allele-depth information and exclude observations that do not provide reliable evidence for either allele. Depending on the software and data format, the reported read counts may be affected by filtering and the caller’s definitions.
  • VAF is particularly important in somatic variant analysis because acquired mutations may occur in only a fraction of cells within a tissue or tumour. A mutation present in a small proportion of tumour cells may have a low VAF, while a mutation present in most tumour cells may have a higher VAF. However, VAF is not a direct measurement of the percentage of cells carrying a mutation. It is influenced by tumour purity, local copy number, loss of heterozygosity, subclonal composition, normal-cell contamination, and sequencing bias. Consequently, a VAF of 0.20 does not necessarily mean that exactly 20% of cells carry the mutation.
  • In a typical diploid germline sample without copy-number changes or substantial technical bias, a heterozygous variant is often expected to have a VAF near 0.5, because one chromosome carries the reference allele and the other carries the alternate allele. A homozygous alternate genotype is generally expected to have a VAF near 1.0. These values are expectations rather than strict rules. Sampling variation, mapping bias, allele-specific amplification, contamination, mosaicism, and local genomic changes can produce different observed values. In addition, sex chromosomes, mitochondrial DNA, polyploid organisms, and regions with variable copy number require interpretation according to their biological ploidy.
  • Population allele frequency is related to VAF but represents a different concept. Population allele frequency is calculated from the number of copies of an allele among the chromosomes in a defined population. For a diploid autosomal population, each individual generally contributes two alleles at a locus, subject to missing genotypes and other ploidy considerations. If 30 alternate alleles are observed among 200 successfully genotyped chromosomes, the alternate allele frequency is 0.15. Population allele frequency is used in population genetics, evolutionary studies, genetic association research, and clinical variant interpretation. Resources such as gnomAD, dbSNP, and the 1000 Genomes Project provide population variation information, although their sample composition and coverage differ.
  • Minor allele frequency (MAF) is the frequency of the less common allele at a biallelic site in a specified population or dataset. If the reference allele frequency is 0.85 and the alternate allele frequency is 0.15, the MAF is 0.15. At multiallelic sites, frequency interpretation is more complex because several alternate alleles may be present. MAF is widely used in genome-wide association studies, population-genetic analyses, and rare-variant research. However, a variant’s frequency does not establish whether it is harmful or beneficial. A rare variant may be benign, while a common variant may influence a trait or disease risk. Population ancestry and the representation of different groups in reference databases are also important when interpreting allele frequencies.
  • Genotype quality (GQ) measures the confidence that a variant caller assigns to a particular genotype in an individual sample. It is commonly represented as a Phred-scaled probability that the called genotype is incorrect. A high GQ indicates stronger evidence for the assigned genotype, while a low GQ indicates greater uncertainty. GQ is usually calculated from the relative likelihoods of possible genotypes based on the observed sequencing evidence. In a typical diploid biallelic setting, the possible genotypes are homozygous reference, heterozygous, and homozygous alternate. The caller evaluates the read evidence for these alternatives and assigns the genotype with the strongest support, together with a quality estimate.
  • The Phred scale is used to express error probabilities logarithmically. Its general relationship is:

[
GQ = -10\log_{10}(P_{\text{genotype error}})
]

  • where (P_{\text{genotype error}}) is the estimated probability that the assigned genotype is incorrect under the caller’s model. A GQ of 10 corresponds to an estimated error probability of 10%, a GQ of 20 corresponds to 1%, and a GQ of 30 corresponds to 0.1%. A GQ of 40 corresponds to 0.01%. These interpretations describe the probability represented by the quality model and do not guarantee that the genotype is correct in every biological or technical context.
  • Genotype quality is different from variant quality. Variant quality, commonly represented by the QUAL field in a VCF file, describes confidence in the variant site or variant call according to the caller’s model. Genotype quality describes confidence in the genotype assigned to an individual sample at that site. A variant can have a high site-level quality score but a low genotype quality in one sample if the sample has insufficient or conflicting read evidence. Conversely, a sample may have a confident genotype at a site even when the overall evidence for the site is not strong across the cohort. Both measures are useful for quality assessment, but they answer different questions.
  • Variant Call Format (VCF) files commonly store genotype quality and allele-depth information in the FORMAT and sample columns. The GT field represents the genotype, GQ represents genotype quality, DP represents read depth, and AD commonly records the number of reads supporting the reference and alternate alleles. For a simple biallelic variant, a VCF record may look like this:

#CHROM  POS     ID  REF ALT QUAL FILTER INFO FORMAT          SAMPLE

chr1    10500   .   A   G   60   PASS   .    GT:AD:DP:GQ    0/1:18,17:35:45

  • In this example, the genotype is 0/1, indicating a heterozygous genotype in a typical diploid setting. The allele-depth field reports 18 reference-supporting reads and 17 alternate-supporting reads, giving an approximate VAF of 0.49. The total depth is 35, and the genotype quality is 45. The site-level quality score is 60. These values provide complementary information about the variant and the sample genotype. Actual VCF records may contain additional fields, and the precise meaning of depth and allele counts can vary by caller.
  • A useful distinction is that sequencing read depth and allele frequency are not the same. Read depth describes the number of observations at a position, while VAF describes the proportion of those observations supporting an alternate allele. Two samples may have the same VAF but very different read depths. For example, a VAF of 0.5 based on 10 reads is less strongly supported than a VAF of 0.5 based on 200 high-quality reads, assuming comparable read characteristics. Greater depth generally improves genotype estimation, but only when the additional reads provide reliable and sufficiently independent evidence.
  • Allele balance is closely related to VAF and is often used to assess heterozygous genotype calls. In a typical diploid germline sample, a heterozygous variant is expected to have roughly equal reference and alternate allele support. A substantial deviation may indicate sequencing or mapping bias, contamination, copy-number changes, allelic imbalance, or mosaicism. However, the expected balance depends on the sample and genomic context. For example, tumour samples may show strong allelic imbalance because of tumour purity, copy-number changes, or subclonal mutations. A universal allele-balance threshold should therefore not be applied without considering the biological setting.
  • In somatic cancer analysis, VAF is particularly useful for examining tumour heterogeneity and clonal composition. A mutation present in most tumour cells may be described as clonal, while a mutation found in only a subset of tumour cells may be subclonal. The observed VAF can help investigate these patterns, but it must be interpreted together with tumour purity and copy-number information. For example, a heterozygous mutation present in every tumour cell in a sample containing a mixture of tumour and normal cells may have a VAF substantially below 0.5. A mutation present in a small subclone may have an even lower VAF. Copy-number gains, deletions, and loss of heterozygosity can further alter the expected fraction.
  • Mosaicism also affects VAF interpretation. A mosaic individual carries genetically distinct cell populations that originated from the same fertilized egg. If a variant arises during early embryonic development, it may be present in several tissues but only in a proportion of cells. A later mutation may be restricted to a particular tissue. Mosaic variants can therefore show VAF values below the levels expected for ordinary heterozygous germline variants. Their detection may require high-depth sequencing and, in some cases, testing of multiple tissues. Low VAF alone does not prove mosaicism because sequencing artifacts and acquired mutations in blood cells can produce similar patterns.
  • Genotype quality is influenced by several factors, including read depth, base quality, mapping quality, allele balance, local sequence complexity, and the statistical model used by the variant caller. Low read depth can make it difficult to distinguish between possible genotypes. Low mapping quality can reduce confidence when reads may originate from multiple genomic locations. Low base quality can make alternate alleles less reliable, while strong allele imbalance may indicate that the observed genotype does not fit the expected model. Genotype quality should therefore be considered alongside the underlying evidence rather than used as an isolated measure.
  • Different variant callers calculate genotype quality using their own methods. GATK HaplotypeCaller estimates genotype likelihoods using local haplotype reconstruction and read evidence. BCFtools provides genotype likelihood and calling approaches based on alignment and pileup information. FreeBayes uses a Bayesian framework to evaluate candidate genotypes. DeepVariant uses a deep-learning model to classify candidate variants and produces genotype and quality information. Because the underlying models differ, GQ values from different callers should not automatically be compared as though they were calibrated identically.
  • In cohort analysis, genotype quality is important because uncertain genotypes can distort allele-frequency estimates, population structure analysis, linkage disequilibrium calculations, and genome-wide association studies. Researchers may apply genotype-level filters to mark low-confidence genotypes as missing while retaining the variant site. This is different from removing an entire variant record. For example, if a variant is confidently genotyped in most samples but has low-quality evidence in a few individuals, genotype-level filtering can preserve the reliable observations. Site-level filtering may then be applied separately using measures such as variant quality, missingness, or other site-level annotations.
  • An illustrative BCFtools command for filtering genotypes by genotype quality is:

# Set genotypes with GQ below 20 to missing

bcftools filter \

    -S . \

    -e ‘FMT/GQ<20’ \

    cohort.vcf.gz \

    -Oz \

    -o cohort.gq_filtered.vcf.gz

# Index the output VCF

bcftools index cohort.gq_filtered.vcf.gz

  • This example illustrates genotype-level filtering, but the exact filtering expression and behaviour should be checked against the installed BCFtools version and the input VCF structure. A genotype with missing or undefined GQ may require separate handling. Researchers should also evaluate the number of genotypes affected and the consequences for sample and variant missingness.
  • Allele frequency can be calculated from genotype data using tools such as PLINK, PLINK 2, and BCFtools. For example, BCFtools can calculate allele frequencies from a multi-sample VCF:

# Calculate allele frequencies

bcftools +fill-tags cohort.vcf.gz \

    -Oz \

    -o cohort.with_frequency.vcf.gz \

    — -t AF,AC,AN

# Index the output

bcftools index cohort.with_frequency.vcf.gz

  • The AF tag represents allele frequency, AC represents the alternate allele count, and AN represents the total number of called alleles used in the calculation. The values are based on the genotypes in the dataset and therefore depend on which samples and alleles are included. For a diploid autosomal dataset, two called alleles generally contribute per successfully genotyped individual, while missing genotypes do not contribute. Sex chromosomes, mitochondrial DNA, polyploid organisms, and copy-number-variable regions require appropriate ploidy interpretation.
  • PLINK 2 can also calculate allele frequencies from genotype data:

# Import VCF data into PLINK 2 format

plink2 \

    –vcf cohort.vcf.gz \

    –make-pgen \

    –out cohort

# Calculate allele frequencies

plink2 \

    –pfile cohort \

    –freq \

    –out cohort_frequency

  • The resulting frequency reports can be used in population-genetic analyses and quality-control procedures. Researchers should distinguish between frequencies calculated across the entire cohort and frequencies calculated within particular ancestry groups or study subgroups. A variant may be common in one population and rare in another, and pooled estimates can conceal this variation.
  • VAF and GQ are also useful in clinical genetics, but neither measure alone establishes pathogenicity. A high-quality genotype can represent a benign variant, while a disease-associated variant may be detected with lower confidence in a challenging genomic region. Clinical interpretation should consider population frequency, inheritance pattern, gene–disease validity, functional evidence, phenotype, segregation, and clinical classification frameworks. Resources such as ClinVar and ClinGen can provide supporting information, but database records should be evaluated in their evidence context and may differ in review status or interpretation.
  • Several limitations must be considered when using VAF and genotype quality. VAF can be affected by sequencing depth, allele-specific mapping, amplification bias, contamination, copy-number changes, and ploidy. GQ depends on the caller’s statistical model and the quality of the evidence supplied to it. A high GQ does not guarantee biological correctness if the reads are systematically misaligned or affected by an unrecognised artifact. A low GQ does not necessarily mean that the alternate allele is false; it may reflect insufficient coverage or a difficult genomic region. Both measures are therefore best interpreted together with read-level evidence, mapping quality, base quality, and relevant biological information.
  • Reproducibility is essential in analyses involving VAF and genotype quality. Researchers should record the reference genome assembly, sequencing platform, read-processing procedures, aligner, variant caller, software versions, and filtering criteria. The definition of read depth and allele depth should be documented, particularly when comparing outputs from different callers. Cohort allele frequencies should specify the samples and population groups included, while somatic VAF interpretation should consider tumour purity and copy-number context. Quality-control reports and original variant files should be retained so that filtering and interpretation decisions can be reviewed.
  • Variant allele frequency and genotype quality provide complementary information about genetic variation. VAF describes the proportion of allele observations supporting an alternate allele, while GQ measures confidence in the genotype assigned to an individual sample. Together with read depth, allele balance, mapping quality, and variant-level quality, they help researchers evaluate the reliability of genetic calls. Their correct interpretation is essential for accurate germline and somatic analysis, population studies, clinical genomics, and reproducible bioinformatics research.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *