- Genotype imputation is a computational method used in bioinformatics to predict missing or unobserved genetic variants in an individual’s genome using information from observed genotypes and patterns of genetic variation in reference populations. It is widely applied in population genetics, genome-wide association studies (GWAS), genetic epidemiology, medical genetics, and large-scale genomic research. Genotyping arrays and sequencing technologies do not always measure every genetic variant in a genome. Imputation helps estimate additional genotypes, increasing the number of variants available for analysis and improving the compatibility of datasets generated using different genotyping platforms.
- Genotyping arrays measure a selected set of genetic markers, usually single-nucleotide polymorphisms (SNPs), rather than every variant in the genome. These markers are chosen to capture common genetic variation and provide information about nearby variants through linkage disequilibrium (LD). Linkage disequilibrium describes the non-random association of alleles at different genomic positions within a population. Because nearby variants may be inherited together as part of shared haplotypes, observed genotypes can provide information about untyped variants. Genotype imputation uses these relationships, together with reference haplotypes, to estimate the most likely genotypes at variants that were not directly measured.
- For example, a genotyping array may measure a group of SNPs within a genomic region but omit several variants located between them. If the observed markers form a haplotype pattern that is commonly associated with a particular combination of alleles in a reference population, an imputation algorithm can estimate the likely genotypes at the unmeasured positions. These estimates are probabilistic rather than direct observations. Their reliability depends on the strength of linkage disequilibrium, the quality and size of the reference panel, the individual’s ancestry, and the accuracy of the observed genotypes.
- A reference panel is a collection of individuals with relatively dense genotype or haplotype information. It provides the genetic patterns used to infer unobserved variants in the study samples. Reference panels may be constructed from whole-genome sequencing datasets or other high-density genotype resources. Examples include the 1000 Genomes Project, the Haplotype Reference Consortium (HRC), and population-specific reference resources. The suitability of a reference panel depends on its genomic coverage, variant quality, population composition, sample size, and compatibility with the study dataset.
- The imputation process generally begins with genotype quality control. Researchers examine sample identity, missingness, heterozygosity, allele frequencies, sex consistency, relatedness, and other indicators of genotyping quality. Variant-level checks include call rates, allele frequencies, strand consistency, and possible genotyping errors. Poor-quality samples and markers can reduce imputation accuracy and should be investigated before proceeding. The study data must also be aligned to a consistent reference genome assembly and allele orientation.
- The next stage involves preparing the study genotypes for imputation. This may include converting genotype files into the required input format, checking chromosome and position information, removing incompatible markers, and harmonizing alleles with the reference panel. Allele harmonization is particularly important because the same SNP may be represented using different strand orientations. For example, an A/G SNP on one strand corresponds to a T/C SNP on the opposite strand. If strand information is inconsistent, allele frequencies and genotype assignments may be interpreted incorrectly. Ambiguous A/T and C/G SNPs require special attention because their complementary alleles cannot be distinguished using allele labels alone without additional information.
- After preparation, the study genotypes are matched with the reference panel. Statistical algorithms identify reference haplotypes that are compatible with the observed genotypes and estimate the likely haplotypes carried by each study individual. The missing genotypes are then inferred from these haplotype patterns. Many imputation methods use statistical phasing, hidden Markov models, or related probabilistic approaches to model the inheritance and sharing of haplotypes. Some methods perform phasing and imputation together, while others use phased haplotypes as an input.
- Statistical phasing is closely related to genotype imputation because imputation often requires knowledge of which alleles occur together on the same chromosome. Phasing assigns alleles to homologous chromosome copies, while imputation predicts unobserved genotypes. In many modern workflows, the two processes are integrated. Reference haplotypes help infer phase, and the inferred phase helps estimate missing variants. Tools such as SHAPEIT, Eagle, Beagle, and other statistical methods support different combinations of phasing and imputation.
- The output of genotype imputation is generally probabilistic. Instead of assigning only one genotype, an imputation program may estimate probabilities for the possible genotypes at each untyped variant. For a diploid biallelic SNP, the possible genotypes are commonly represented as 0/0, 0/1, and 1/1. An imputation result may assign probabilities such as 0.90, 0.08, and 0.02, respectively. These values indicate the estimated probabilities of the three genotype states under the model. The most likely genotype is 0/0, but the uncertainty should not be ignored.
- A useful measure derived from genotype probabilities is the dosage. For a biallelic variant, the expected number of alternative alleles is calculated by weighting each genotype by its probability. If (P(0/0)), (P(0/1)), and (P(1/1)) are the genotype probabilities, the alternative-allele dosage is:
[
\text{Dosage} = P(0/1) + 2P(1/1)
]
- For example, if the genotype probabilities are 0.90 for 0/0, 0.08 for 0/1, and 0.02 for 1/1, the dosage is 0.12. This represents the expected number of alternative alleles, ranging from zero to two for a diploid biallelic variant. Dosage values retain uncertainty information and are often preferred over converting every imputed genotype into a single hard genotype call.
- Imputation quality is an essential consideration. Different programs report different quality measures, and their interpretation depends on the software and statistical model. Common measures include the imputation INFO score, estimated dosage (r^2), and related accuracy statistics. These measures assess how well the imputed genotypes are estimated, but they are not identical and should not be treated as interchangeable. Researchers should consult the documentation of the imputation software to understand the definition and interpretation of each metric.
- Imputation quality is influenced by several factors. A larger and more representative reference panel generally provides more information about population haplotypes. Strong linkage disequilibrium between observed and unobserved variants can improve prediction, whereas weakly associated or rare variants may be difficult to infer. Genotyping errors, low-quality markers, poor allele harmonization, population differences, and insufficient reference coverage can reduce accuracy. Imputation is generally more reliable for common variants that are well represented in the reference panel than for very rare variants or variants found primarily in underrepresented populations.
- Population ancestry is particularly important. Reference panels that closely match the ancestry of the study population often provide more accurate imputation because they better represent the haplotypes and allele frequencies of the individuals being analysed. If a study population is poorly represented, imputation accuracy may decline, especially for rare or population-specific variants. Multi-ancestry reference panels can improve coverage across diverse populations, but they do not eliminate all differences in accuracy. Researchers should evaluate performance in the populations included in their study rather than assuming that one reference panel works equally well for everyone.
- Several reference resources are commonly used in genotype imputation. The 1000 Genomes Project provides phased genomic variation data from multiple populations and has been widely used as a reference resource. The Haplotype Reference Consortium provides a large collection of haplotypes that has supported imputation of common genetic variants. TOPMed provides extensive whole-genome sequencing data and has been used to construct reference panels for large-scale imputation. Other resources, including population-specific panels, may be more appropriate for particular ancestry groups or research populations. Availability, access requirements, genomic coverage, and assembly compatibility should be checked before selecting a panel.
- A number of software tools are widely used for genotype imputation. IMPUTE5 is designed for large-scale genotype imputation and uses reference haplotypes to estimate unobserved variants. Minimac4 is commonly used in imputation workflows and is associated with reference-panel-based imputation pipelines. Beagle supports genotype phasing and imputation and can be used in different types of genomic datasets. Eagle is widely used for statistical phasing, often as part of a workflow that subsequently performs imputation. SHAPEIT is another widely used phasing tool. The appropriate tool depends on the size of the dataset, reference panel, genotype format, computational resources, and study objective.
- A simplified example of genotype imputation using Minimac4 is shown below:
# Example structure for genotype imputation
minimac4 \
–refHaps reference_panel.m3vcf.gz \
–haps study_phased.vcf.gz \
–prefix imputed_results
- The example illustrates the main inputs: a reference haplotype panel and phased study genotypes. The actual options and file requirements depend on the Minimac4 version and the reference-panel format. Study genotypes generally need to be phased and prepared in a compatible format before imputation. The output commonly includes imputed variant information and quality measures that should be inspected before downstream analysis.
- A simplified Beagle workflow may look like this:
# Example structure for phasing and genotype imputation with Beagle
java -Xmx8g -jar beagle.jar \
gt=study_genotypes.vcf.gz \
ref=reference_panel.vcf.gz \
out=imputed_results
- Beagle can perform phasing and imputation depending on the input data and configuration. The exact command-line parameters, reference requirements, and supported file options should be checked against the installed version. A successful program run does not by itself establish that the imputed genotypes are accurate; quality assessment remains necessary.
- Before imputation, genotype data may be stored in formats such as PLINK binary files, VCF, or BCF. Many imputation services and software packages require particular formats and may impose additional requirements for chromosome naming, allele representation, genetic maps, and reference genome assembly. Data preparation may involve converting between formats using PLINK 2, BCFtools, or other tools. Variant positions and allele definitions must remain consistent throughout these conversions.
- Imputation is frequently used in genome-wide association studies. Genotyping arrays usually assay a limited number of markers, but GWAS may require a much larger set of variants to identify genetic associations. Imputation expands the available variant set by estimating genotypes at untyped positions. This can increase the number of variants tested, improve the coverage of genomic regions, and make it easier to combine studies that used different genotyping arrays. It can also support fine-mapping by providing information about variants that were not directly genotyped, although fine-mapping accuracy depends on the quality of imputation and the statistical model used.
- In population genetics, imputation can improve the completeness of genotype datasets used to study genetic diversity, population structure, linkage disequilibrium, demographic history, and relatedness. It allows researchers to analyse more variants across individuals and can facilitate comparisons between datasets with different marker coverage. However, population-specific variation and differences in reference-panel representation must be considered when interpreting imputed genotypes.
- Genotype imputation is also useful in genetic epidemiology and large cohort studies. Large studies may combine samples genotyped using different platforms or at different times. Imputation can help harmonize these datasets by estimating a more consistent set of variants across individuals. This is particularly useful when studying complex traits and diseases influenced by many genetic loci. Nevertheless, differences in sample preparation, genotyping technology, population ancestry, and quality-control procedures can introduce systematic errors that imputation cannot automatically correct.
- In medical genetics, imputation may help identify variants in genomic regions that were not directly measured by a genotyping array. It can support research into disease susceptibility, genetic risk prediction, and the identification of candidate variants. However, imputed genotypes should not automatically be treated as equivalent to directly observed sequencing results. Rare, clinically important, or treatment-relevant variants generally require appropriate confirmation and interpretation using validated clinical methods. Imputation alone is not sufficient to establish a diagnosis or determine a variant’s clinical significance.
- Imputation quality control should be performed after the analysis. Researchers commonly examine the number of imputed variants, allele frequencies, genotype distributions, imputation quality scores, and missingness. Variants with poor imputation quality may be excluded from downstream analyses. Thresholds should be selected according to the software, study design, allele frequency, and intended analysis rather than applying a single universal cutoff. Some studies use stricter criteria for rare variants because their imputation is often less reliable.
- An important distinction exists between genotype imputation and missing genotype filling within a dataset. Imputation predicts genotypes using statistical relationships and reference haplotypes, whereas simple missing-value filling may replace missing entries using a basic rule such as the most frequent genotype. These approaches are not equivalent. Genotype imputation uses population-level genetic information and probabilistic models to estimate unobserved variation, while simple filling does not adequately represent haplotype structure or genotype uncertainty.
- Genotype imputation also differs from variant calling. Variant calling identifies genetic differences from sequencing reads aligned to a reference genome. Imputation estimates genotypes at variants that were not directly observed or measured, using existing genotype information and reference data. Similarly, imputation is distinct from genotype phasing, although the two processes are often integrated. Phasing determines which alleles occur together on the same chromosome, while imputation predicts unobserved alleles or genotypes.
- Despite its advantages, genotype imputation has important limitations. It cannot reliably recover every variant, particularly rare variants that are absent or poorly represented in the reference panel. Accuracy may be reduced in highly repetitive or structurally complex genomic regions. Differences in ancestry, population history, allele frequency, and linkage disequilibrium can influence prediction quality. Genotyping errors and incorrect strand alignment may also propagate into imputed results. In addition, the imputed genotype probabilities depend on the assumptions and data used by the statistical model.
- Structural variants and complex genomic regions present further challenges. Standard imputation methods are often designed primarily for SNPs and small insertions and deletions. Larger deletions, duplications, inversions, and complex rearrangements may require specialized reference panels and methods. Regions with extensive copy-number variation or highly similar paralogous sequences may also be difficult to impute accurately. Researchers should therefore avoid assuming that the performance of SNP imputation methods extends automatically to all types of genetic variation.
- For reproducible research, the complete imputation workflow should be documented. Important details include the genome assembly, reference panel and version, genotyping platform, quality-control criteria, phasing and imputation software versions, genetic map, command-line parameters, and post-imputation filtering thresholds. The original genotype data should be preserved, and imputed outputs should be stored separately. Researchers should also record the number of variants retained or excluded and the quality measures used to evaluate the results.
- Genotype imputation is a fundamental method in modern bioinformatics because it expands genetic datasets beyond the variants directly measured by sequencing or genotyping arrays. By using linkage disequilibrium, haplotype information, and reference panels, it provides probabilistic estimates of unobserved genotypes. Tools such as IMPUTE5, Minimac4, Beagle, and related programs support large-scale imputation in population genetics, GWAS, and genetic epidemiology. Reliable results require appropriate reference panels, careful data harmonization, accurate phasing, quality assessment, and transparent reporting of uncertainty. Understanding these principles is essential for using imputed genotypes responsibly in genomic research.