Unobserved Genotypes Research Articles

Issues and Challenges of Imputation Techniques in Genome Wide Association Studies (GWAS): A Review

A genome-wide association study (GWAS) rapidly scans DNA markers in many individuals to find genetic links to diseases. New findings aid in disease detection, treatment and prevention. Imputation predicts untyped genotypes in genetic studies when data is missing due to quality, cost, or design issues. It’s a proven statistical technique for estimating unobserved genotypes by borrowing haplotype segments from a densely genotyped reference panel. This allows estimation and testing of associations at unassayed variants.Genotype imputation is vital in analyzing genome-wide association scans, helping geneticists evaluate evidence for association at untyped genetic markers. This summary outlines missing data issues and various imputation methods.

A diverse ancestrally-matched reference panel increases genotype imputation accuracy in a underrepresented population

Variant imputation, a common practice in genome-wide association studies, relies on reference panels to infer unobserved genotypes. Multiple public reference panels are currently available with variations in size, sequencing depth, and represented populations. Currently, limited data exist regarding the performance of public reference panels when used in an imputation of populations underrepresented in the reference panel. Here, we compare the performance of various public reference panels: 1000 Genomes Project, Haplotype Reference Consortium, GenomeAsia 100 K, and the recent Trans-Omics for Precision Medicine (TOPMed) program, when used in an imputation of samples from the Thai population. Genotype yields were assessed, and imputation accuracies were examined by comparison with high-depth whole genome sequencing data of the same sample. We found that imputation using the TOPMed panel yielded the largest number of variants (~ 271 million). Despite being the smallest in size, GenomeAsia 100 K achieved the best imputation accuracy with a median genotype concordance rate of 0.97. For rare variants, GenomeAsia 100 K also offered the best accuracy, although rare variants were less accurately imputable than common variants (30.3% reduction in concordance rates). The high accuracy observed when using GenomeAsia 100 K is likely attributable to the diverse representation of populations genetically similar to the study cohort emphasizing the benefits of sequencing populations classically underrepresented in human genomics.

Higher-order epistasis and phenotypic prediction

Contemporary high-throughput mutagenesis experiments are providing an increasingly detailed view of the complex patterns of genetic interaction that occur between multiple mutations within a single protein or regulatory element. By simultaneously measuring the effects of thousands of combinations of mutations, these experiments have revealed that the genotype-phenotype relationship typically reflects not only genetic interactions between pairs of sites but also higher-order interactions among larger numbers of sites. However, modeling and understanding these higher-order interactions remains challenging. Here we present a method for reconstructing sequence-to-function mappings from partially observed data that can accommodate all orders of genetic interaction. The main idea is to make predictions for unobserved genotypes that match the type and extent of epistasis found in the observed data. This information on the type and extent of epistasis can be extracted by considering how phenotypic correlations change as a function of mutational distance, which is equivalent to estimating the fraction of phenotypic variance due to each order of genetic interaction (additive, pairwise, three-way, etc.). Using these estimated variance components, we then define an empirical Bayes prior that in expectation matches the observed pattern of epistasis and reconstruct the genotype-phenotype mapping by conducting Gaussian process regression under this prior. To demonstrate the power of this approach, we present an application to the antibody-binding domain GB1 and also provide a detailed exploration of a dataset consisting of high-throughput measurements for the splicing efficiency of human pre-mRNA [Formula: see text] splice sites, for which we also validate our model predictions via additional low-throughput experiments.

Comparison of Genotype Imputation for SNP Array and Low-Coverage Whole-Genome Sequencing Data.

Genotype imputation is the term used to describe the process of inferring unobserved genotypes in a sample of individuals. It is a key step prior to a genome-wide association study (GWAS) or genomic prediction. The imputation accuracy will directly influence the results from subsequent analyses. In this simulation-based study, we investigate the accuracy of genotype imputation in relation to some factors characterizing SNP chip or low-coverage whole-genome sequencing (LCWGS) data. The factors included the imputation reference population size, the proportion of target markers /SNP density, the genetic relationship (distance) between the target population and the reference population, and the imputation method. Simulations of genotypes were based on coalescence theory accounting for the demographic history of pigs. A population of simulated founders diverged to produce four separate but related populations of descendants. The genomic data of 20,000 individuals were simulated for a 10-Mb chromosome fragment. Our results showed that the proportion of target markers or SNP density was the most critical factor affecting imputation accuracy under all imputation situations. Compared with Minimac4, Beagle5.1 reproduced higher-accuracy imputed data in most cases, more notably when imputing from the LCWGS data. Compared with SNP chip data, LCWGS provided more accurate genotype imputation. Our findings provided a relatively comprehensive insight into the accuracy of genotype imputation in a realistic population of domestic animals.

Efficient approaches for large-scale GWAS with genotype uncertainty.

Association studies using genetic data from SNP-chip-based imputation or low-depth sequencing data provide a cost-efficient design for large-scale association studies. We explore methods for performing association studies applicable to such genetic data and investigate how using different priors when estimating genotype probabilities affects the association results. Our proposed method, ANGSD-asso’s latent model, models the unobserved genotype as a latent variable in a generalized linear model framework. The software is implemented in C/C++ and can be run multi-threaded. ANGSD-asso is based on genotype probabilities, which can be estimated using either the sample allele frequency or the individual allele frequencies as a prior. We explore through simulations how genotype probability-based methods compare with using genetic dosages. Our simulations show that in a structured population using the individual allele frequency prior has better power than the sample allele frequency. In scenarios with sequencing depth and phenotype correlation ANGSD-asso’s latent model has higher statistical power and less bias than using dosages. Adding additional covariates to the linear model of ANGSD-asso’s latent model has higher statistical power and less bias than other methods that accommodate genotype uncertainty, while also being much faster. This is shown with imputed data from UK Biobank and simulations.

Prediction of Maize Phenotypic Traits With Genomic and Environmental Predictors Using Gradient Boosting Frameworks.

The development of crop varieties with stable performance in future environmental conditions represents a critical challenge in the context of climate change. Environmental data collected at the field level, such as soil and climatic information, can be relevant to improve predictive ability in genomic prediction models by describing more precisely genotype-by-environment interactions, which represent a key component of the phenotypic response for complex crop agronomic traits. Modern predictive modeling approaches can efficiently handle various data types and are able to capture complex nonlinear relationships in large datasets. In particular, machine learning techniques have gained substantial interest in recent years. Here we examined the predictive ability of machine learning-based models for two phenotypic traits in maize using data collected by the Maize Genomes to Fields (G2F) Initiative. The data we analyzed consisted of multi-environment trials (METs) dispersed across the United States and Canada from 2014 to 2017. An assortment of soil- and weather-related variables was derived and used in prediction models alongside genotypic data. Linear random effects models were compared to a linear regularized regression method (elastic net) and to two nonlinear gradient boosting methods based on decision tree algorithms (XGBoost, LightGBM). These models were evaluated under four prediction problems: (1) tested and new genotypes in a new year; (2) only unobserved genotypes in a new year; (3) tested and new genotypes in a new site; (4) only unobserved genotypes in a new site. Accuracy in forecasting grain yield performance of new genotypes in a new year was improved by up to 20% over the baseline model by including environmental predictors with gradient boosting methods. For plant height, an enhancement of predictive ability could neither be observed by using machine learning-based methods nor by using detailed environmental information. An investigation of key environmental factors using gradient boosting frameworks also revealed that temperature at flowering stage, frequency and amount of water received during the vegetative and grain filling stage, and soil organic matter content appeared as important predictors for grain yield in our panel of environments.

Genotype imputation using the Positional Burrows Wheeler Transform.

Genotype imputation is the process of predicting unobserved genotypes in a sample of individuals using a reference panel of haplotypes. In the last 10 years reference panels have increased in size by more than 100 fold. Increasing reference panel size improves accuracy of markers with low minor allele frequencies but poses ever increasing computational challenges for imputation methods. Here we present IMPUTE5, a genotype imputation method that can scale to reference panels with millions of samples. This method continues to refine the observation made in the IMPUTE2 method, that accuracy is optimized via use of a custom subset of haplotypes when imputing each individual. It achieves fast, accurate, and memory-efficient imputation by selecting haplotypes using the Positional Burrows Wheeler Transform (PBWT). By using the PBWT data structure at genotyped markers, IMPUTE5 identifies locally best matching haplotypes and long identical by state segments. The method then uses the selected haplotypes as conditioning states within the IMPUTE model. Using the HRC reference panel, which has ∼65,000 haplotypes, we show that IMPUTE5 is up to 30x faster than MINIMAC4 and up to 3x faster than BEAGLE5.1, and uses less memory than both these methods. Using simulated reference panels we show that IMPUTE5 scales sub-linearly with reference panel size. For example, keeping the number of imputed markers constant, increasing the reference panel size from 10,000 to 1 million haplotypes requires less than twice the computation time. As the reference panel increases in size IMPUTE5 is able to utilize a smaller number of reference haplotypes, thus reducing computational cost.

Enviromics in breeding: applications and perspectives on envirotypic-assisted selection.

We propose the application of enviromics to breeding practice, by which the similarity among sites assessed on an "omics" scale of environmental attributes drives the prediction of unobserved genotype performances. Genotype by environment interaction (GEI) studies in plant breeding have focused mainly on estimating genetic parameters over a limited number of experimental trials. However, recent geographic information system (GIS) techniques have opened new frontiers for better understanding and dealing with GEI. These advances allow increasing selection accuracy across all sites of interest, including those where experimental trials have not yet been deployed. Here, we introduce the term enviromics, within an envirotypic-assisted breeding framework. In summary, likewise genotypes at DNA markers, any particular site is characterized by a set of "envirotypes" at multiple "enviromic" markers corresponding to environmental variables that may interact with the genetic background, thus providing informative breeding re-rankings for optimized decisions over different environments. Based on simulated data, we illustrate an index-based enviromics method (the "GIS-GEI") which, due to its higher granular resolution than standard methods, allows for: (1) accurate matching of sites to their most appropriate genotypes; (2) better definition of breeding areas that have high genetic correlation to ensure selection gains across environments; and (3) efficient determination of the best sites to carry out experiments for further analyses. Environmental scenarios can also be optimized for productivity improvement and genetic resources management, especially in the current outlook of dynamic climate change. Envirotyping provides a new class of markers for genetic studies, which are fairly inexpensive, increasingly available and transferable across species. We envision a promising future for the integration of enviromics approaches into plant breeding when coupled with next-generation genotyping/phenotyping and powerful statistical modeling of genetic diversity.

Multi-Trait Regressor Stacking Increased Genomic Prediction Accuracy of Sorghum Grain Composition

Genomic prediction has enabled plant breeders to estimate breeding values of unobserved genotypes and environments. The use of genomic prediction will be extremely valuable for compositional traits for which phenotyping is labor-intensive and destructive for most accurate results. We studied the potential of Bayesian multi-output regressor stacking (BMORS) model in improving prediction performance over single trait single environment (STSE) models using a grain sorghum diversity panel (GSDP) and a biparental recombinant inbred lines (RILs) population. A total of five highly correlated grain composition traits—amylose, fat, gross energy, protein and starch, with genomic heritability ranging from 0.24 to 0.59 in the GSDP and 0.69 to 0.83 in the RILs were studied. Average prediction accuracies from the STSE model were within a range of 0.4 to 0.6 for all traits across both populations except amylose (0.25) in the GSDP. Prediction accuracy for BMORS increased by 41% and 32% on average over STSE in the GSDP and RILs, respectively. Prediction of whole environments by training with remaining environments in BMORS resulted in moderate to high prediction accuracy. Our results show regression stacking methods such as BMORS have potential to accurately predict unobserved individuals and environments, and implementation of such models can accelerate genetic gain.

Genotype Imputation in Winter Wheat Using First-Generation Haplotype Map SNPs Improves Genome-Wide Association Mapping and Genomic Prediction of Traits.

Genome-wide single nucleotide polymorphism (SNP) variation allows for the capture of haplotype structure in populations and prediction of unobserved genotypes based on inferred regions of identity-by-descent (IBD). Here we have used a first-generation wheat haplotype map created by targeted re-sequencing of low-copy genomic regions in the reference panel of 62 lines to impute marker genotypes in a diverse panel of winter wheat cultivars from the U.S. Great Plains. The IBD segments between the reference population and winter wheat cultivars were identified based on SNP genotyped using the 90K iSelect wheat array and genotyping by sequencing (GBS). A genome-wide association study and genomic prediction of resistance to stripe rust in winter wheat cultivars showed that an increase in marker density achieved by imputation improved both the power and precision of trait mapping and prediction. The majority of the most significant marker-trait associations belonged to imputed genotypes. With the vast amount of SNP variation data accumulated for wheat in recent years, the presented imputation framework will greatly improve prediction accuracy in breeding populations and increase resolution of trait mapping hence, facilitate cross-referencing of genotype datasets available across different wheat populations.

Inferring Population Structure and Admixture Proportions in Low-Depth NGS Data.

We here present two methods for inferring population structure and admixture proportions in low-depth next-generation sequencing (NGS) data. Inference of population structure is essential in both population genetics and association studies, and is often performed using principal component analysis (PCA) or clustering-based approaches. NGS methods provide large amounts of genetic data but are associated with statistical uncertainty, especially for low-depth sequencing data. Models can account for this uncertainty by working directly on genotype likelihoods of the unobserved genotypes. We propose a method for inferring population structure through PCA in an iterative heuristic approach of estimating individual allele frequencies, where we demonstrate improved accuracy in samples with low and variable sequencing depth for both simulated and real datasets. We also use the estimated individual allele frequencies in a fast non-negative matrix factorization method to estimate admixture proportions. Both methods have been implemented in the PCAngsd framework available at http://www.popgen.dk/software/.

Genotype imputation performance of three reference panels using African ancestry individuals.

Genotype imputation estimates unobserved genotypes from genome-wide makers, to increase genome coverage and power for genome-wide association studies. Imputation has been successful for European ancestry populations in which very large reference panels are available. Smaller subsets of African descent populations are available in 1000 Genomes (1000G), the Consortium on Asthma among African ancestry Populations in the Americas (CAAPA) and the Haplotype Reference Consortium (HRC). We compared the performance of these reference panels when imputing variation in 3747 African Americans (AA) from two cohorts (HCV and COPDGene) genotyped using Illumina Omni microarrays. The haplotypes of 2504 (1000G), 883 (CAAPA) and 32,470 individuals (HRC) were used as reference. We compared the number of variants, imputation quality, imputation accuracy and coverage between panels. In both cohorts, 1000G imputed 1.5-1.6× more variants than CAAPA and 1.2× more than HRC. Similar findings were observed for variants with imputation R2 > 0.5 and for rare, low-frequency, and common variants. When merging imputed variants of the three panels, the total number was 62-63M with 20M overlapping variants imputed by all three panels, and a range of 5-15M variants imputed exclusively with one of them. For overlapping variants, imputation quality was highest for HRC, followed by 1000G, then CAAPA, and improved as the minor allele frequency increased. 1000G, HRC and CAAPA provided high performance and accuracy for imputation of African American individuals, increasing the number of variants available for subsequent analyses. These panels are complementary and would benefit from the development of an integrated African reference panel.

The social genome of friends and schoolmates in the National Longitudinal Study of Adolescent to Adult Health

Humans tend to form social relationships with others who resemble them. Whether this sorting of like with like arises from historical patterns of migration, meso-level social structures in modern society, or individual-level selection of similar peers remains unsettled. Recent research has evaluated the possibility that unobserved genotypes may play an important role in the creation of homophilous relationships. We extend this work by using data from 5,500 adolescents from the National Longitudinal Study of Adolescent to Adult Health (Add Health) to examine genetic similarities among pairs of friends. Although there is some evidence that friends have correlated genotypes, both at the whole-genome level as well as at trait-associated loci (via polygenic scores), further analysis suggests that meso-level forces, such as school assignment, are a principal source of genetic similarity between friends. We also observe apparent social-genetic effects in which polygenic scores of an individual's friends and schoolmates predict the individual's own educational attainment. In contrast, an individual's height is unassociated with the height genetics of peers.

Accuracy of genotype imputation in Nelore cattle.

BackgroundGenotype imputation from low-density (LD) to high-density single nucleotide polymorphism (SNP) chips is an important step before applying genomic selection, since denser chips tend to provide more reliable genomic predictions. Imputation methods rely partially on linkage disequilibrium between markers to infer unobserved genotypes. Bos indicus cattle (e.g. Nelore breed) are characterized, in general, by lower levels of linkage disequilibrium between genetic markers at short distances, compared to taurine breeds. Thus, it is important to evaluate the accuracy of imputation to better define which imputation method and chip are most appropriate for genomic applications in indicine breeds.MethodsAccuracy of genotype imputation in Nelore cattle was evaluated using different LD chips, imputation software and sets of animals. Twelve commercial and customized LD chips with densities ranging from 7 K to 75 K were tested. Customized LD chips were virtually designed taking into account minor allele frequency, linkage disequilibrium and distance between markers. Software programs FImpute and BEAGLE were applied to impute genotypes. From 995 bulls and 1247 cows that were genotyped with the Illumina® BovineHD chip (HD), 793 sires composed the reference set, and the remaining 202 younger sires and all the cows composed two separate validation sets for which genotypes were masked except for the SNPs of the LD chip that were to be tested.ResultsImputation accuracy increased with the SNP density of the LD chip. However, the gain in accuracy with LD chips with more than 15 K SNPs was relatively small because accuracy was already high at this density. Commercial and customized LD chips with equivalent densities presented similar results. FImpute outperformed BEAGLE for all LD chips and validation sets. Regardless of the imputation software used, accuracy tended to increase as the relatedness between imputed and reference animals increased, especially for the 7 K chip.ConclusionsIf the Illumina® BovineHD is considered as the target chip for genomic applications in the Nelore breed, cost-effectiveness can be improved by genotyping part of the animals with a chip containing around 15 K useful SNPs and imputing their high-density missing genotypes with FImpute.Electronic supplementary materialThe online version of this article (doi:10.1186/s12711-014-0069-1) contains supplementary material, which is available to authorized users.

Impact of pre-imputation SNP-filtering on genotype imputation results.

BackgroundImputation of partially missing or unobserved genotypes is an indispensable tool for SNP data analyses. However, research and understanding of the impact of initial SNP-data quality control on imputation results is still limited. In this paper, we aim to evaluate the effect of different strategies of pre-imputation quality filtering on the performance of the widely used imputation algorithms MaCH and IMPUTE.ResultsWe considered three scenarios: imputation of partially missing genotypes with usage of an external reference panel, without usage of an external reference panel, as well as imputation of completely un-typed SNPs using an external reference panel. We first created various datasets applying different SNP quality filters and masking certain percentages of randomly selected high-quality SNPs. We imputed these SNPs and compared the results between the different filtering scenarios by using established and newly proposed measures of imputation quality. While the established measures assess certainty of imputation results, our newly proposed measures focus on the agreement with true genotypes. These measures showed that pre-imputation SNP-filtering might be detrimental regarding imputation quality. Moreover, the strongest drivers of imputation quality were in general the burden of missingness and the number of SNPs used for imputation. We also found that using a reference panel always improves imputation quality of partially missing genotypes. MaCH performed slightly better than IMPUTE2 in most of our scenarios. Again, these results were more pronounced when using our newly defined measures of imputation quality.ConclusionEven a moderate filtering has a detrimental effect on the imputation quality. Therefore little or no SNP filtering prior to imputation appears to be the best strategy for imputing small to moderately sized datasets. Our results also showed that for these datasets, MaCH performs slightly better than IMPUTE2 in most scenarios at the cost of increased computing time.

Estimating Individual Admixture Proportions from Next Generation Sequencing Data

Inference of population structure and individual ancestry is important both for population genetics and for association studies. With next generation sequencing technologies it is possible to obtain genetic data for all accessible genetic variations in the genome. Existing methods for admixture analysis rely on known genotypes. However, individual genotypes cannot be inferred from low-depth sequencing data without introducing errors. This article presents a new method for inferring an individual's ancestry that takes the uncertainty introduced in next generation sequencing data into account. This is achieved by working directly with genotype likelihoods that contain all relevant information of the unobserved genotypes. Using simulations as well as publicly available sequencing data, we demonstrate that the presented method has great accuracy even for very low-depth data. At the same time, we demonstrate that applying existing methods to genotypes called from the same data can introduce severe biases. The presented method is implemented in the NGSadmix software available at http://www.popgen.dk/software.

Genotype imputation accuracy in a F2 pig population using high density and low density SNP panels

BackgroundF2 resource populations have been used extensively to map QTL segregating between pig breeds. A limitation associated with the use of these resource populations for fine mapping of QTL is the reduced number of founding individuals and recombinations of founding haplotypes occurring in the population. These limitations, however, become advantageous when attempting to impute unobserved genotypes using within family segregation information. A trade-off would be to re-type F2 populations using high density SNP panels for founding individuals and low density panels (tagSNP) in F2 individuals followed by imputation. Subsequently a combined meta-analysis of several populations would provide adequate power and resolution for QTL mapping, and could be achieved at relatively low cost. Such a strategy allows the wealth of phenotypic information that has previously been obtained on experimental resource populations to be further mined for QTL identification. In this study we used experimental and simulated high density genotypes (HD-60K) from an F2 cross to estimate imputation accuracy under several genotyping scenarios.ResultsSelection of tagSNP using physical distance or linkage disequilibrium information produced similar imputation accuracies. In particular, tagSNP sets averaging 1 SNP every 2.1 Mb (1,200 SNP genome-wide) yielded imputation accuracies (IA) close to 0.97. If instead of using custom panels, the commercially available 9K chip is used in the F2, IA reaches 0.99. In order to attain such high imputation accuracy the F0 and F1 generations should be genotyped at high density. Alternatively, when only the F0 is genotyped at HD, while F1 and F2 are genotyped with a 9K panel, IA drops to 0.90.ConclusionsCombining 60K and 9K panels with imputation in F2 populations is an appealing strategy to re-genotype existing populations at a fraction of the cost.

Low penetrance susceptibility to glioma is caused by the TP53 variant rs78378222

Background:Most of the heritable risk of glioma is presently unaccounted for by mutations in known genes. In addition to rare inactivating germline mutations in TP53 causing glioma in the context of the Li-Fraumeni syndrome, polymorphic variation in TP53 may also contribute to the risk of developing glioma.Methods:To comprehensively evaluate the impact of variation in TP53 on risk, we analysed 23 tagSNPs and imputed 2377 unobserved genotypes in four series totaling 4147 glioma cases and 7435 controls.Results:The strongest validated association signal was shown by the imputed single-nucleotide polymorphism (SNP) rs78378222 (P=6.86 × 10−24, minor allele frequency ∼0.013). Confirmatory genotyping confirmed the high quality of the imputation. The association between rs78378222 and risk was seen for both glioblastoma multiforme (GBM) and non-GBM tumours. We comprehensively examined the relationship between rs78378222 and overall survival in two of the case series totaling 1699 individuals. Despite employing statistical tests sensitive to the detection of differences in early survival, no association was shown.Conclusion:Our data provided strong validation of rs78378222 as a risk factor for glioma but do not support the tenet that the polymorphism being a clinically useful prognostic marker. Acquired TP53 inactivation is a common feature of glioma. As rs78378222 changes the polyadenylation signal of TP53 leading to impaired 3′-end processing of TP53 mRNA, the SNP has strong plausibility for being directly functional contributing to the aetiological basis of glioma.

Improving sequence-based genotype calls with linkage disequilibrium and pedigree information

Whole and targeted sequencing of human genomes is a promising, increasingly feasible tool for discovering genetic contributions to risk of complex diseases. A key step is calling an individual's genotype from the multiple aligned short read sequences of his DNA, each of which is subject to nucleotide read error. Current methods are designed to call genotypes separately at each locus from the sequence data of unrelated individuals. Here we propose likelihood-based methods that improve calling accuracy by exploiting two features of sequence data. The first is the linkage disequilibrium (LD) between nearby SNPs. The second is the Mendelian pedigree information available when related individuals are sequenced. In both cases the likelihood involves the probabilities of read variant counts given genotypes, summed over the unobserved genotypes. Parameters governing the prior genotype distribution and the read error rates can be estimated either from the sequence data itself or from external reference data. We use simulations and synthetic read data based on the 1000 Genomes Project to evaluate the performance of the proposed methods. An R-program to apply the methods to small families is freely available at http://med.stanford.edu/epidemiology/PHGC/.

Genotype imputation with thousands of genomes.

Genotype imputation is a statistical technique that is often used to increase the power and resolution of genetic association studies. Imputation methods work by using haplotype patterns in a reference panel to predict unobserved genotypes in a study dataset, and a number of approaches have been proposed for choosing subsets of reference haplotypes that will maximize accuracy in a given study population. These panel selection strategies become harder to apply and interpret as sequencing efforts like the 1000 Genomes Project produce larger and more diverse reference sets, which led us to develop an alternative framework. Our approach is built around a new approximation that uses local sequence similarity to choose a custom reference panel for each study haplotype in each region of the genome. This approximation makes it computationally efficient to use all available reference haplotypes, which allows us to bypass the panel selection step and to improve accuracy at low-frequency variants by capturing unexpected allele sharing among populations. Using data from HapMap 3, we show that our framework produces accurate results in a wide range of human populations. We also use data from the Malaria Genetic Epidemiology Network (MalariaGEN) to provide recommendations for imputation-based studies in Africa. We demonstrate that our approximation improves efficiency in large, sequence-based reference panels, and we discuss general computational strategies for modern reference datasets. Genome-wide association studies will soon be able to harness the power of thousands of reference genomes, and our work provides a practical way for investigators to use this rich information. New methodology from this study is implemented in the IMPUTE2 software package.