- Genotype quality control is a fundamental step in bioinformatics, population genetics, genome-wide association studies (GWAS), genomic prediction, and other genetic analyses. It involves systematically evaluating genetic data to identify and address errors, missing information, unreliable genotype calls, sample contamination, incorrect sample identification, and technical artifacts. The objective is to ensure that the genetic data used in downstream analyses are sufficiently accurate, consistent, and representative of the biological samples being investigated. Poor-quality genotype data can produce false genetic associations, distort allele-frequency estimates, affect population-structure analysis, and reduce the accuracy of genomic predictions.
- A genotype describes the alleles carried by an individual at a particular genetic locus. In a typical diploid organism, an individual may carry two copies of a chromosome and therefore two alleles at an autosomal locus. For a biallelic single-nucleotide polymorphism (SNP), the possible genotypes are commonly represented as homozygous reference (0/0), heterozygous (0/1), and homozygous alternate (1/1). Genotype data may be generated using SNP genotyping arrays, whole-genome sequencing (WGS), whole-exome sequencing (WES), targeted sequencing, or other molecular technologies. Each technology has different sources of error and therefore requires appropriate quality-control procedures.
- Genotype quality control is generally performed at both the sample level and the variant level. Sample-level quality control identifies individuals whose genotype data are unreliable or inconsistent with the study design. Variant-level quality control identifies genetic markers with poor call rates, low confidence, unusual allele distributions, or other technical problems. Additional analyses may evaluate population structure, relatedness, sex consistency, contamination, batch effects, and sample identity. These procedures are interconnected, and quality-control decisions should be made in the context of the biological question, genotyping technology, and characteristics of the study population.
- The first important measure is the genotype call rate, which describes the proportion of expected genotypes successfully determined for a sample or genetic variant. A high call rate indicates that relatively few genotype observations are missing, while a low call rate may indicate poor DNA quality, technical failure, problematic probes, low sequencing coverage, or difficult genomic regions. The sample call rate is calculated as the number of successfully called genotypes divided by the total number of genotypes expected for that sample. The variant call rate is calculated in the same way across individuals for a particular marker. Researchers may remove samples or variants with excessive missingness, but thresholds should be selected according to the dataset and analysis requirements rather than applied universally.
- Sample-level call rate can be represented as:
[
\text{Sample call rate} =
\frac{\text{Number of successfully called genotypes}}
{\text{Total number of expected genotypes}}
\times 100
]
- Variant-level call rate can be calculated as:
[
\text{Variant call rate} =
\frac{\text{Number of successfully called genotypes at a variant}}
{\text{Total number of individuals}}
\times 100
]
- Missing genotype rates are also important because missingness may not occur randomly. For example, a particular SNP may be difficult to genotype in one ancestry group because of differences in nearby sequence variation or probe-binding regions. If missingness differs between cases and controls, or between population groups, it may introduce bias into association analyses. Researchers should therefore examine missingness by phenotype, ancestry, genotyping batch, and other relevant sample characteristics.
- Minor allele frequency (MAF) is another commonly evaluated measure. It represents the frequency of the less common allele at a genetic locus in the analysed population. Very rare variants may be difficult to genotype accurately using some platforms, particularly when the number of observed alternate alleles is small. Low-frequency variants can also have unstable statistical estimates in association analyses. However, rare variants may be biologically important, especially in Mendelian disease research and rare-variant association studies. Therefore, low MAF should not automatically be treated as evidence of poor quality. Variant filtering should distinguish technical unreliability from genuine biological rarity.
- For a biallelic variant, allele frequency is calculated using the number of copies of each allele:
[
AF = \frac{\text{Number of copies of the allele}}
{\text{Total number of successfully observed allele copies}}
]
- For a diploid sample of (N) individuals with complete genotypes, the total number of allele copies is normally (2N), although chromosome ploidy and missing genotypes must be considered. The minor allele frequency is the frequency of the less common allele:
[
MAF = \min(AF_{\text{reference}}, AF_{\text{alternate}})
]
- Allele frequencies should be interpreted within the relevant study population because population structure and ancestry can produce substantial differences among groups. A variant that is rare in one population may be more common in another. Filtering based on an overall cohort frequency can therefore remove variants that are important in a smaller ancestry group.
- Hardy–Weinberg equilibrium (HWE) is commonly examined during genotype quality control. Under the standard HWE model, genotype frequencies in a large, randomly mating population with no relevant evolutionary disturbances are expected to follow the proportions (p^2), (2pq), and (q^2), where (p) and (q) are the frequencies of two alleles and (p+q=1). Significant deviations from these expectations may indicate genotyping errors, population structure, inbreeding, non-random mating, selection, or other biological and sampling factors.
- HWE testing is often used to identify problematic variants, particularly in unrelated control samples from GWAS. A strong departure from HWE may indicate an excess of heterozygotes or homozygotes that requires investigation. However, HWE deviations are not necessarily technical errors. Population mixtures, relatedness, consanguinity, selection, and disease ascertainment can all affect genotype proportions. For this reason, HWE filtering should be applied carefully, with attention to the study design. In case-control studies, HWE is commonly evaluated in controls rather than cases because disease association itself can cause deviations among cases.
- Sample heterozygosity is another useful quality-control measure. It describes the proportion of heterozygous genotypes observed across a set of genetic markers in an individual. Unusually high heterozygosity may indicate contamination, sample mixture, or technical problems, while unusually low heterozygosity may reflect inbreeding, population background, low genetic diversity, or data-quality issues. Heterozygosity should therefore be evaluated relative to the study population and marker set rather than interpreted using a universal threshold.
- Sex consistency checks are important when genotype data include sex chromosomes. Genetic sex estimates can be inferred from patterns of heterozygosity on the X chromosome and the presence or absence of Y-chromosome markers, depending on the organism and its sex-chromosome system. These estimates can be compared with recorded sample information to identify possible sample swaps, incorrect metadata, or unusual biological conditions. Discrepancies should be investigated rather than automatically interpreted as errors because variations in sex-chromosome number, mosaicism, aneuploidy, and differences in biological sex characteristics can complicate interpretation. Sex-check methods must also be adapted for organisms with different chromosome systems.
- Relatedness analysis identifies genetically similar individuals and helps detect duplicate samples, close relatives, and unexpected family relationships. Closely related individuals may violate the assumptions of statistical methods designed for unrelated samples. They can also influence population-structure estimates, allele-frequency calculations, and association statistics. Tools such as KING, PLINK, and related software estimate genetic relationships using genome-wide markers. Depending on the study design, researchers may remove duplicate samples, retain one individual from each related pair, or use statistical models that explicitly account for relatedness. Family-based studies should not automatically remove relatives because their relationships may be central to the research question.
- Sample contamination occurs when genetic material from another individual or source is mixed with the DNA of the intended sample. It may produce unusual allele balances, excess heterozygosity, unexpected genotypes, or inconsistent sequencing patterns. Contamination can arise during sample collection, laboratory preparation, library construction, sequencing, or data processing. Tools such as VerifyBamID2 can estimate contamination in sequencing datasets under appropriate conditions. Other methods examine allele balance, genotype concordance, and unexpected genetic mixtures. Contamination estimates should be interpreted in relation to sequencing depth, reference-panel assumptions, and the technology used.
- Sample identity and concordance checks are particularly important when genotype data are combined with phenotype records, repeated measurements, or datasets generated at different times. Genetic fingerprints based on selected SNPs can be compared across datasets to identify duplicate samples, sample swaps, and mismatched identifiers. Tools such as Somalier and PLINK can support identity and relatedness checks. For longitudinal studies, sample tracking is essential because an incorrect match between genotype and phenotype data can invalidate an otherwise technically sound analysis.
- Batch effects occur when genetic measurements differ systematically because samples were processed in different laboratory batches, on different genotyping arrays, using different sequencing platforms, or under different experimental conditions. A batch effect may produce apparent genetic differences that are unrelated to biological variation. Researchers should examine call rates, missingness, allele frequencies, genotype clusters, and other quality metrics across batches. Principal component analysis (PCA) can also reveal technical clustering, although PCA separation should not automatically be interpreted as a batch effect because it may reflect genuine population structure. Batch information should be incorporated into quality-control investigations and, where appropriate, downstream statistical models.
- Genotype quality scores provide information about the confidence of individual genotype calls. In sequencing-based analyses, genotype quality (GQ) is commonly reported as a Phred-scaled measure of confidence in a genotype assignment. A higher GQ generally indicates lower estimated genotype error probability. The Phred relationship is:
[
GQ = -10\log_{10}(P_{\text{genotype error}})
]
- For example, GQ 20 corresponds to an estimated genotype error probability of 1%, while GQ 30 corresponds to 0.1%, under the probability model used by the caller. These values describe the caller’s estimated confidence and do not guarantee that a genotype is correct. Genotype quality depends on read depth, base quality, mapping quality, allele balance, local sequence complexity, and the statistical model used by the variant caller.
- Read depth (DP) is another important measure in sequencing-based genotyping. It indicates the number of sequencing reads or observations supporting a genotype at a locus, according to the caller’s definition. Low depth can result in uncertain genotype calls, while unusually high depth may indicate duplicated regions, copy-number variation, or alignment artifacts. However, high depth does not automatically mean high-quality genotypes. Researchers should consider depth together with GQ, allele depth, mapping quality, and local genomic context.
- Allele balance describes the relative support for different alleles in sequencing reads. For a typical diploid heterozygous variant, the alternate allele may be expected to account for approximately half of the supporting reads, although the observed proportion can vary. Strongly imbalanced allele support may indicate mapping bias, amplification bias, low coverage, contamination, copy-number changes, or other technical or biological factors. Allele balance is especially important when evaluating heterozygous calls and somatic variants, where tumour purity and subclonal composition can affect observed allele fractions.
- Genotype quality control also requires attention to ploidy and chromosome-specific properties. Autosomes are usually diploid in humans, while sex chromosomes may have different ploidy patterns depending on chromosomal complement. Mitochondrial DNA is generally present in multiple copies per cell and may show heteroplasmy. Other organisms may have haploid, polyploid, aneuploid, or more complex genome structures. Applying diploid assumptions to all chromosomes or organisms can produce incorrect genotype calls and misleading quality-control results. Ploidy settings should therefore match the biological system being studied.
- Population structure should be evaluated after initial genotype quality control and before many downstream association analyses. Principal component analysis can reveal major axes of genetic variation, while ancestry estimation and genetic relationship matrices can provide additional information. Population structure may affect allele frequencies and produce confounding in GWAS. However, population-structure analysis itself can be distorted by poor-quality samples, batch effects, close relatives, and highly correlated markers. Quality-control procedures such as LD pruning and careful sample selection are commonly used before PCA.
- Genotype quality control differs somewhat between SNP array data and sequencing data. SNP arrays rely on probe hybridization and genotype-cluster separation. Quality assessment may include genotype call rates, cluster plots, probe performance, intensity measurements, and allele-frequency patterns. Poorly performing probes, cross-hybridizing markers, and variants with ambiguous genomic locations may need to be excluded. Sequencing-based genotype data require assessment of read depth, mapping quality, base quality, allele balance, variant-calling confidence, and local alignment characteristics. Both technologies require sample identity checks, missingness assessment, and evaluation of population structure.
- A typical genotype quality-control workflow begins with raw data inspection and sample identification. Researchers evaluate sample and variant call rates, remove or investigate low-quality observations, assess allele frequencies, examine HWE where appropriate, and investigate heterozygosity and relatedness. They then evaluate sex consistency, contamination, batch effects, and population structure. After these steps, the retained variants and samples can be used for association analysis, imputation, genomic prediction, or population-genetic investigation. The exact order may vary because some quality-control steps depend on earlier filtering decisions, and analyses are often iterative.
- PLINK and PLINK 2 are widely used tools for genotype quality control and genome-wide analyses. They support sample and variant filtering, allele-frequency calculations, missingness summaries, HWE testing, heterozygosity-related analyses, LD pruning, PCA, and relatedness assessment. The following examples illustrate a simplified workflow using PLINK 2 with data in PGEN format. Exact options and output names may vary by software version and dataset.
# Inspect sample and variant missingness
plink2 –pfile cohort \
–missing \
–out cohort_missing
Allele frequencies can be calculated using:
# Calculate allele frequencies
plink2 –pfile cohort \
–freq \
–out cohort_frequency
- A minor allele frequency filter can be applied when justified by the study design:
# Retain variants with MAF of at least 0.01
plink2 –pfile cohort \
–maf 0.01 \
–make-pgen \
–out cohort_maf_filtered
A Hardy–Weinberg equilibrium test can be performed as follows:
# Test Hardy-Weinberg equilibrium
plink2 –pfile cohort \
–hardy \
–out cohort_hwe
- For a case-control study, HWE assessment should be performed in an appropriately defined control group when that is the intended quality-control procedure. The precise sample-selection and filtering commands depend on how case and control status are encoded in the phenotype file.
- Variant and sample missingness can be filtered using thresholds selected for the dataset:
# Example: filter variants and samples by missingness
plink2 –pfile cohort \
–geno 0.02 \
–mind 0.02 \
–make-pgen \
–out cohort_missing_filtered
- Here, –geno 0.02 excludes variants with more than 2% missing genotypes, while –mind 0.02 excludes samples with more than 2% missing genotypes. These are illustrative thresholds, not universal recommendations. More stringent or relaxed criteria may be appropriate depending on genotyping technology, sample size, variant frequency, and research objectives.
- Linkage disequilibrium pruning is often used before PCA and other population-structure analyses:
# Select approximately independent variants
plink2 –pfile cohort_missing_filtered \
–indep-pairwise 200 50 0.2 \
–out cohort_pruning
The resulting pruned variant set can be used to calculate principal components:
# Calculate the first 10 principal components
plink2 –pfile cohort_missing_filtered \
–extract cohort_pruning.prune.in \
–pca 10 \
–out cohort_pca
- These examples demonstrate common operations but do not constitute a complete quality-control pipeline. Researchers should inspect the generated reports, confirm that the software options match their installed version, and determine whether filtering has removed biologically important variants or specific population groups.
- For sequencing-based genotype data, BCFtools and related tools can be used to inspect and filter VCF or BCF files. For example, a basic variant-level filter may retain sites with sufficient quality and sequencing depth:
# Retain variants meeting example site-quality and depth criteria
bcftools filter \
-i ‘QUAL >= 30 && INFO/DP >= 10’ \
input.vcf.gz \
-Oz -o filtered.vcf.gz
# Index the compressed VCF
bcftools index -t filtered.vcf.gz
- This example applies site-level filters and should not be mistaken for complete sample-level or genotype-level quality control. Site quality, sample genotype quality, read depth, allele depth, mapping quality, and other evidence should be evaluated separately. Filtering criteria must also be adapted to the variant-calling pipeline and the biological purpose of the study.
- After quality control, researchers should evaluate how filtering has changed the dataset. They should record the number of samples and variants before and after each filtering step, examine missingness and allele-frequency distributions, and compare retained and excluded samples. Excessive filtering may disproportionately remove individuals from underrepresented populations or eliminate rare variants of biological importance. Sensitivity analyses using alternative thresholds can help determine whether major conclusions depend strongly on particular filtering choices.
- Genotype quality control is particularly important in GWAS because technical errors can produce false-positive or false-negative associations. Poorly genotyped variants may appear associated with a phenotype if genotype error rates differ between cases and controls. Population structure, relatedness, batch effects, and differential missingness can also distort association statistics. Reliable genotype data should therefore be combined with phenotype quality control, population-structure correction, appropriate statistical models, and multiple-testing procedures.
- In genomic prediction and animal or plant breeding, genotype quality control influences the accuracy of genomic estimated breeding values and other prediction results. Incorrect genotypes can distort genomic relationship matrices, allele-frequency estimates, and marker-effect estimates. Quality control is also important when combining animals or plants from different breeds, lines, populations, or breeding generations. Researchers should consider breed composition, ploidy, genotyping platform, marker density, and population-specific allele frequencies when deciding which samples and markers to retain.
- For reproducible research, every genotype quality-control decision should be documented. Important information includes the genome assembly, genotyping or sequencing platform, variant-calling pipeline, sample and variant inclusion criteria, missingness thresholds, MAF thresholds, HWE procedures, relatedness rules, contamination checks, and software versions. Researchers should preserve the original genotype data, retain quality-control reports, and use scripts or workflow systems to make processing repeatable. When data are shared, appropriate metadata should describe the processing steps and known limitations.
- Genotype quality control is essential for producing accurate and reliable genetic analyses. It identifies technical errors, unreliable samples, poor-quality variants, contamination, sample mismatches, and other problems that can compromise downstream research. By combining sample-level and variant-level assessment with appropriate population-genetic and statistical checks, researchers can improve the quality of GWAS, genomic prediction, evolutionary studies, and other bioinformatics applications. Careful documentation and biologically informed filtering are necessary to ensure that quality control improves data reliability without unnecessarily removing genuine genetic diversity.