Designing genome-wide association studies: sample size, power, imputation, and the choice of genotyping chip.

Spencer CC, Su Z, Donnelly P, Marchini J. · 2009

Read Designing genome-wide association studies: sample size, power, imputation, and the choice of genotyping chip. by Spencer CC, Su Z, Donnelly P, Marchini J. (2009) 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-pmc2688469.txt.

Introduction

Spencer, Su, Donnelly, and Marchini set out in 2009 how to design genome-wide hunts for disease genes so the money is well spent. Two choices dominate: which commercial gene chip to buy, and how many patients and controls to type. Vendors advertised chips by coverage, the share of variants captured, but the authors argue coverage is the wrong scorecard.

What matters is statistical power, the true chance of catching a real culprit variant. Because variants travel in correlated blocks, power cannot be computed by formula; it must be simulated. Their simulations of case and control samples, built to mimic real correlation patterns, compared all major chips across effect sizes and populations. Big coverage gaps often shrank to trivial power gaps. Filling in untyped variants by imputation, educated guessing from neighbors, also bought real power cheaply. On a fixed budget, the best buy is often more samples, not the densest chip.

Catalog record

CREATORSpencer CC, Su Z, Donnelly P, Marchini J.
YEAR2009
COLLECTIONPAPERS
FORMATTXT
FILE SIZE63 KB
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC2688469

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Designing Genome-Wide Association Studies: Sample Size, Power, Imputation, and the Choice of Genotyping Chip By Chris C. A. Spencer, Zhan Su, Peter Donnelly, Jonathan Marchini, John D. Storey Published in: PLoS Genetics DOI: 10.1371/journal.pgen.1000477 ABSTRACT Genome-wide association studies are revolutionizing the search for the genes underlying human complex diseases. The main decisions to be made at the design stage of these studies are the choice of the commercial genotyping chip to be used and the numbers of case and control samples to be genotyped. The most common method of comparing different chips is using a measure of coverage, but this fails to properly account for the effects of sample size, the genetic model of the disease, and linkage disequilibrium between SNPs. In this paper, we argue that the statistical power to detect a causative variant should be the major criterion in study design. Because of the complicated pattern of linkage disequilibrium (LD) in the human genome, power cannot be calculated analytically and must instead be assessed by simulation. We describe in detail a method of simulating case-control samples at a set of linked SNPs that replicates the patterns of LD in human populations, and we used it to assess power for a comprehensive set of available genotyping chips. Our results allow us to compare the performance of the chips to detect variants with different effect sizes and allele frequencies, look at how power changes with sample size in different populations or when using multi-marker tags and genotype imputation approaches, and how performance compares to a hypothetical chip that contains every SNP in HapMap. A main conclusion of this study is that marked differences in genome coverage may not translate into appreciable differences in power and that, when taking budgetary considerations into account, the most powerful design may not always correspond to the chip with the highest coverage. We also show that genotype imputation can be used to boost the power of many chips up to the level obtained from a hypothetical “complete” chip containing all the SNPs in HapMap. Our results have been encapsulated into an R software package that allows users to design future association studies and our methods provide a framework with which new chip sets can be evaluated. INTRODUCTION The International HapMap project [1],[2] documented the strong correlations between alleles at polymorphic loci in close physical proximity along human chromosomes. As a consequence it is necessary to genotype only a subset of loci to capture much of the common variation in the genome. Combined with recent technological innovations this observation has made the concept of genome-wide association (GWA) studies a reality [3],[4]. Over the few last years these studies have been very successful in uncovering new disease genes for many different complex diseases [5]. Well over 300 such loci have already been published and many more studies are currently being planned. In the design of such studies two fundamental decisions have to be made: which loci to genotype, and in how many individuals. Both decisions have practical constraints. For example it is currently not possible to assay all known variation in the human genome at a reasonable cost and choices must be made between a set of commercially available genotyping chips. Similarly, sample sizes are often limited by the number of well characterized clinical samples. Therefore, ultimately, the researcher and funding bodies must ask how to use the financial and practical resources available in order to best further the understanding of the genetics of the disease or trait of interest. A primary consideration should be the power of the study: the probability of detecting a variant assumed to be causal. In comparing chips for GWA studies it has been common to ask what proportion of SNPs not directly genotyped are “captured” or “tagged” by the chip, i.e. are well predicted, via LD, by a SNP, or combination of SNPs, on the chip. To do so it is necessary to define the level of prediction required, or equivalently to set a threshold for the required level of correlation. Although arbitrary, this has often been set at 0.8 [6],[7],[8]. The resulting proportion of SNPs captured at this level is often referred to as the coverage of the chip. Having specified the threshold it is possible to estimate the coverage of a particular chip from HapMap data, although we note that some care is required to account for SNPs not in HapMap [8]. Here we focus instead on the power of particular chips to detect causal variants of different effect sizes, and the way in which this varies with study size and/or study cost and when using genotype imputation methods. Although coverage is straightforward to estimate, power is a complicated function of the set of SNPs on the chip, effect size, and sample size, and can only be assessed by simulation. It turns out that differences in coverage between chips are often not reflected in substantial differences in power and that the use of genotype imputation further reduces these differences. Study power is routinely used throughout science in experimental design and we argue that it should be the primary consideration in designing GWAs. This approach was used in settling several design questions in the Wellcome Trust Case Control Consortium [5]. Our results have been encapsulated in a user-friendly R package that allows the power of different chip and sample size combinations to be assessed given a total budget for the study. Knowledge of study power is also invaluable when analysing data from a study. Assessment of whether positive results at a particular significance level are “real” or due to chance requires knowledge of power [5], and the practical decision of how far down the list of potential associations one should go in replication studies should be informed by power considerations. Other comparisons of chips have been carried out but have either focussed exclusively on estimating coverage [8], have been limited in scope of which chips have been evaluated [9] or have used analytical calculations that do not properly take into account the complex LD structure of the human genome [10],[11] or failed to assess the impact of imputation correctly [11]. A recent paper [12] has used chip data to assess the performance of the chips but the small sample size (N = 359) means that these results cannot be used to assess power of new study designs of more realistic sizes. In addition, the simulations of quantitative phenotypes used the Signal to Noise Ratio (SNR) to measure effect size of the causal SNP which is non-standard and difficult to interpret. For binary traits, simulations assumed a disease prevalence of 25%, a relative risk of 3 and a sample size of only 75 cases and 75 controls. These parameter settings are not realistic for genome-wide association studies or useful when designing new studies. RESULTS Theoretical Results Study power depends on assumptions about the underlying disease model, in addition to effect sizes and sample sizes. When the true causative SNP is not on the genotyping chip there will typically be several SNPs on the chip which are correlated with it. One or more of these could give a signal of significant association and hence allow detection of the locus. The LD structure

Read the complete text