Genotype imputation with thousands of genomes.
Howie B, Marchini J, Stephens M. · 2011
Read Genotype imputation with thousands of genomes. by Howie B, Marchini J, Stephens M. (2011) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3276165.txt.
Introduction
Bryan Howie, Jonathan Marchini, and Matthew Stephens developed a framework for genotype imputation with thousands of reference genomes, in this 2011 paper in G3. Imputation fills in unmeasured genetic variants in a study by copying from densely typed reference panels, boosting the power of association studies. As projects like 1000 Genomes made panels huge and diverse, choosing which reference haplotypes to use became a bottleneck.
Their approximation sidesteps panel selection: for each study chromosome and each genome region, local sequence similarity picks a custom set of the most useful reference haplotypes on the fly. That keeps computation manageable while using all available references, and it sharpens accuracy at low-frequency variants by catching unexpected sharing between populations. Tests on HapMap data and African malaria-network data showed accurate results across many populations. The method ships in the IMPUTE2 package.
Catalog record
| CREATOR | Howie B, Marchini J, Stephens M. |
|---|---|
| YEAR | 2011 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Genotype Imputation with Thousands of Genomes By Bryan Howie, Jonathan Marchini, Matthew Stephens, A. Chakravarti Published in: G3: Genes|Genomes|Genetics DOI: 10.1534/g3.111.001198 ABSTRACT 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. Genotype imputation is a well-established statistical technique for estimating unobserved genotypes in association studies (Browning 2008; Li et al. 2009; Marchini and Howie 2010). Imputation works by copying haplotype segments from a densely genotyped reference panel into individuals typed at a subset of the reference variants. In this way, genotypes can be estimated and tested for association at variants that were not assayed in a study. This approach can increase the power of a given study (Guan and Stephens 2008; Li et al. 2010; Marchini et al. 2007; Servin and Stephens 2007), find candidate susceptibility variants to guide fine-mapping (e.g. Liu et al. 2010), and facilitate meta-analyses that combine studies genotyped on different sets of variants (De Bakker et al. 2008; Zeggini and Ioannidis 2009). Although the statistical methods for genotype imputation are now highly developed and widely used, there are still open questions in the field. Of these, one of the most pressing is how best to use rapidly accumulating reference datasets from around the world to impute into various human populations. Our goal in this paper is to develop a framework for achieving accurate and efficient imputation with current and future reference panels. We focus specifically on low-frequency variation (defined here as polymorphisms with MAFs from 0.5 to 5%) because interrogating such variants is a major aim of contemporary association studies (McCarthy et al. 2008; Stranger et al. 2011). In the first generation of genome-wide association studies (GWAS), most imputation analyses used reference panels from Phase 2 of the International HapMap Project (The International HapMap Consortium 2005), which contain a total of 210 unrelated individuals with ancestry from West Africa, East Asia, and Europe. More recently, Phase 3 of the HapMap Project (The International HapMap Consortium 2010) increased the public reference set to over 1000 unrelated individuals sampled from 11 locations, including parts of the world that were poorly represented by HapMap 2. The 1000 Genomes Project (The 1000 Genomes Project Consortium 2010) is currently extending this resource by applying whole-genome shotgun sequencing to an even larger number of individuals (∼2500) sampled from a finer geographic grid (∼25 separate locations). Direct sequencing can identify variants not represented on the genotyping arrays that were used in the HapMap Project, so this reference set will include many more polymorphic sites than did previous imputation reference panels, with most of the new variants occurring at low population frequencies. As high-throughput genotyping and sequencing technologies continue to mature and decrease in cost, the worldwide collection of reference data will continue to grow. There are numerous approaches for building imputation reference panels from publicly available resources. Many published GWAS simply used the HapMap 2 panel that most closely matched the ancestry of the study population; e.g. CEU for European populations and CHB+JPT for East Asian populations. This “best match” strategy produced useful results in several studies, but it can yield suboptimal accuracy with more diverse reference collections (The International HapMap Consortium 2010) or in studies with no clear reference matches (Huang et al. 2009a). A simple alternative is to use a “cosmopolitan” reference set that includes all available haplotypes, each of which is assigned an equal chance of being copied a priori. This approach produces relatively accurate results in a variety of human populations and has therefore been proposed as a good fallback choice when the optimal panel composition is unclear (Guan and Stephens 2008; Huang et al. 2009a; Li et al. 2010). Another class of methods tries to maximize accuracy by weighting reference panels through cross-validation (Huang et al. 2009a) or ancestry estimation (Egyud et al. 2009; Pasaniuc et al. 2010); the Pasaniuc et al. approach differs from the others in that it uses local ancestry estimates to provide customized reference weights for each study individual. As an alternative, Jostins et al. (2011) suggested balancing accuracy and computation by using reference panels that “approximately cluster” with the study individuals on a plot of principal components (PC) that capture genetic ancestry. The standard way to impute genotypes in a GWAS is to apply one of these panel selection schemes and then pass the indicated haplotypes to an imputation method. Modern reference datasets pose a couple of challenges to this paradigm. First, many panel selection strategies become harder to execute or interpret as reference data are sampled from more populations. For example, the systematic HapMap 2 cross-validations from Huang et al. (2009a) would be more difficult to implement with additional panels, and it is harder to make clear demarcations on a plot of PCs [as recommended by Jostins et al. (2011)] as more populations are included. More importantly, larger reference panels increase the computational burden of imputation, which may compel some investigators to use smaller panels at the cost of imputation accuracy and association power. Methodological developments like “pre-phasing” (which we address in the Discussion) can alleviate this concern by speeding up imputation, but existing methods still need substantial computing power to handle reference sets with thousands of haplotypes. Our work was motivated by the idea that growing reference datasets need not make panel selection more difficult or force tradeoffs between imputation speed and accuracy; in principle, larger and more diverse reference collections could actually make it easier to identify haplotype sharing with simple models, thereby making imputation faster and more accurate. Kong et al. (2008) demonstrated this point in a large sample from a founder population, where they found that genotypes could be accurately phased and imputed using simple identity-by-state (IBS) calculations to identify shared haplotype segments. Howie et