8.5 Genotype imputation

Genotype imputation is a powerful technique in genomics that predicts unobserved genetic variants using reference panels. It leverages linkage disequilibrium and haplotype sharing to increase marker density, enhancing genome-wide association studies and meta-analyses across different genotyping platforms.

Various statistical methods, including hidden Markov models and expectation-maximization algorithms, are used for imputation. Reference panels like the 1000 Genomes Project provide diverse haplotype data. Imputation quality is assessed using metrics such as INFO scores, with factors like reference panel size affecting accuracy.

Principles of genotype imputation

• Genotype imputation is a computational method used to infer unobserved genotypes in a sample of individuals based on a reference panel of haplotypes
• It leverages the linkage disequilibrium (LD) structure and haplotype sharing between the study sample and reference panel to predict missing genotypes
• Imputation enables increased power for genome-wide association studies (GWAS) by increasing the density of genetic markers and facilitating meta-analysis across different genotyping platforms

Statistical methods for imputation

Haplotype frequency estimation

Top images from around the web for Haplotype frequency estimation

• 

  Frontiers | Optimizing Selection of the Reference Population for Genotype Imputation From Array ... View original

  Is this image relevant?

• 

  Frontiers | Hap-E Search 2.0: Improving the Performance of a Probabilistic Donor-Recipient ... View original

  Is this image relevant?

• 

  Frontiers | Optimizing Selection of the Reference Population for Genotype Imputation From Array ... View original

  Is this image relevant?

• 

  Frontiers | Hap-E Search 2.0: Improving the Performance of a Probabilistic Donor-Recipient ... View original

  Is this image relevant?

1 of 2

Top images from around the web for Haplotype frequency estimation

• 

  Frontiers | Optimizing Selection of the Reference Population for Genotype Imputation From Array ... View original

  Is this image relevant?

• 

  Frontiers | Hap-E Search 2.0: Improving the Performance of a Probabilistic Donor-Recipient ... View original

  Is this image relevant?

• 

  Frontiers | Optimizing Selection of the Reference Population for Genotype Imputation From Array ... View original

  Is this image relevant?

• 

  Frontiers | Hap-E Search 2.0: Improving the Performance of a Probabilistic Donor-Recipient ... View original

  Is this image relevant?

1 of 2

• Estimating haplotype frequencies is a crucial step in genotype imputation
• Statistical methods such as the expectation-maximization (EM) algorithm and Markov Chain Monte Carlo (MCMC) methods are used to estimate haplotype frequencies from genotype data
• Accurate haplotype frequency estimates improve the accuracy of genotype imputation

Hidden Markov models

• Hidden Markov models (HMMs) are widely used for genotype imputation
• HMMs model the observed genotypes as emissions from hidden states representing the unobserved haplotypes
• The transition probabilities between hidden states capture the LD structure and recombination patterns in the population

Expectation-maximization algorithm

• The EM algorithm is an iterative method used for estimating haplotype frequencies and imputing missing genotypes
• It alternates between an expectation step (E-step) and a maximization step (M-step) until convergence
• The E-step computes the expected genotype counts given the current haplotype frequency estimates, and the M-step updates the haplotype frequencies based on the expected counts

Markov Chain Monte Carlo methods

• MCMC methods, such as the Gibbs sampler, are used for genotype imputation and haplotype inference
• MCMC methods generate a Markov chain of haplotype configurations that converges to the posterior distribution of haplotypes given the observed genotypes
• Samples from the Markov chain are used to estimate haplotype frequencies and impute missing genotypes

Reference panels for imputation

HapMap project

• The International HapMap Project was a landmark effort to catalog common genetic variants and LD patterns in diverse human populations
• It provided a reference panel of haplotypes for genotype imputation in early GWAS
• The HapMap data improved the power and resolution of genetic association studies

1000 Genomes Project

• The 1000 Genomes Project aimed to provide a comprehensive catalog of human genetic variation by sequencing over 2,500 individuals from diverse populations worldwide
• It generated a larger and more diverse reference panel for genotype imputation compared to the HapMap project
• The 1000 Genomes data enhanced the accuracy and coverage of genotype imputation, particularly for low-frequency and rare variants

Population-specific reference panels

• Population-specific reference panels, such as the Haplotype Reference Consortium (HRC) and the African Genome Resources Panel, have been developed to improve imputation accuracy in specific ancestral groups
• These panels capture unique LD patterns and haplotype diversity in populations underrepresented in the HapMap and 1000 Genomes data
• Using population-specific reference panels can increase the power and robustness of genetic association studies in diverse populations

Imputation software and tools

IMPUTE vs Beagle

• IMPUTE and Beagle are two widely used software packages for genotype imputation
• IMPUTE uses a hidden Markov model and a reference panel of haplotypes to estimate the probability distribution of missing genotypes
• Beagle uses a localized haplotype-cluster model and a reference panel to impute missing genotypes
• Both tools have been shown to provide accurate imputation results, with some differences in computational efficiency and handling of rare variants

MaCH vs minimac

• MaCH (Markov Chain Haplotyping) is a genotype imputation method that uses a hidden Markov model and a reference panel to estimate the most likely genotypes for each individual
• Minimac is an extension of MaCH that enables efficient imputation of large-scale datasets by leveraging pre-phased haplotypes and a compact representation of the reference panel
• Minimac has been widely adopted for imputation in large-scale GWAS and meta-analyses due to its computational efficiency and scalability

Genotype imputation servers

• Genotype imputation servers, such as the Michigan Imputation Server and the Sanger Imputation Service, provide web-based platforms for users to perform genotype imputation using a variety of reference panels and imputation algorithms
• These servers offer a user-friendly interface and handle the computational burden of imputation, making it accessible to researchers without extensive computational resources
• Imputation servers also provide quality control and post-imputation filtering options to ensure the accuracy and reliability of the imputed genotypes

Assessing imputation quality

Imputation accuracy metrics

• Several metrics are used to assess the accuracy of imputed genotypes, including the imputation quality score (INFO score), the concordance rate, and the squared correlation ($r^2$) between imputed and true genotypes
• The INFO score, ranging from 0 to 1, measures the certainty of the imputed genotypes and is commonly used to filter out poorly imputed variants
• The concordance rate and $r^2$ provide a direct measure of the agreement between imputed and observed genotypes in a validation dataset

Factors affecting imputation quality

• Imputation quality is influenced by several factors, including the size and diversity of the reference panel, the density and quality of the genotype data, and the LD structure and allele frequencies in the study population
• Larger and more diverse reference panels generally improve imputation accuracy, particularly for rare and population-specific variants
• Higher-density genotyping arrays and improved genotype calling algorithms also contribute to better imputation quality

Post-imputation filtering strategies

• Post-imputation filtering is an important step to remove poorly imputed variants and ensure the reliability of downstream analyses
• Common filtering criteria include removing variants with low imputation quality scores (e.g., INFO < 0.3), low minor allele frequencies (MAF < 1%), or high rates of missing data
• Applying stringent post-imputation filters can reduce the risk of false-positive associations and improve the reproducibility of genetic association studies

Applications of genotype imputation

Genome-wide association studies

• Genotype imputation has revolutionized GWAS by enabling the analysis of millions of genetic variants across the genome, even if they were not directly genotyped
• Imputation increases the power to detect associations by leveraging information from correlated markers and facilitating meta-analysis across studies with different genotyping platforms
• Imputed genotypes have been successfully used to identify numerous genetic risk factors for complex traits and diseases

Fine-mapping of causal variants

• Genotype imputation can aid in the fine-mapping of causal variants by increasing the resolution of genetic association signals
• By imputing genotypes at a higher density, researchers can better localize the causal variants driving the association and prioritize them for functional follow-up studies
• Fine-mapping with imputed genotypes has led to the identification of functional variants and insights into the biological mechanisms underlying complex traits

Meta-analysis of genetic studies

• Genotype imputation enables the harmonization of genetic data across studies with different genotyping platforms, facilitating large-scale meta-analyses
• Meta-analysis combines the results from multiple studies to increase statistical power and identify robust genetic associations
• Imputation to a common reference panel allows for the direct comparison and aggregation of genetic effects across studies, leading to the discovery of novel risk loci and the refinement of effect size estimates

Limitations and challenges

Rare and structural variants

• Genotype imputation is less accurate for rare variants (MAF < 1%) and structural variants (e.g., copy number variations) due to their limited representation in reference panels and the reduced LD with surrounding markers
• Imputing rare and structural variants may require larger and more diverse reference panels, as well as specialized algorithms that can handle the complexity of these variants
• The functional impact and clinical relevance of rare and structural variants imputed with lower accuracy should be interpreted with caution

Computational resources required

• Genotype imputation is computationally intensive, requiring substantial memory and processing power, particularly for large-scale datasets and dense reference panels
• The computational burden of imputation can be a limiting factor for researchers with limited computational resources
• Cloud-based imputation services and high-performance computing clusters have emerged as solutions to address the computational challenges of imputation

Imputation in admixed populations

• Genotype imputation in admixed populations, such as African Americans and Latinos, can be challenging due to the complex LD patterns and the limited representation of ancestral haplotypes in reference panels
• Imputation accuracy in admixed populations can be improved by using population-specific reference panels or by leveraging local ancestry information to guide the imputation process
• Careful quality control and validation of imputed genotypes in admixed populations are crucial to ensure the reliability of downstream analyses and the generalizability of genetic findings