MultiPhen: joint model of multiple phenotypes can increase discovery in GWAS.
O'Reilly PF, Hoggart CJ, Pomyen Y, Calboli FC, Elliott P, Jarvelin MR, Coin LJ. · 2012
Read MultiPhen: joint model of multiple phenotypes can increase discovery in GWAS. by O'Reilly PF, Hoggart CJ, Pomyen Y, Calboli FC, Elliott P, Jarvelin MR, Coin LJ. (2012) 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-pmc3342314.txt.
Introduction
O'Reilly and colleagues introduced a method that analyzes several traits at once to find gene variants, in a 2012 paper. Genome wide association studies have found hundreds of variants tied to diseases and measurements, yet they almost always test one trait at a time. That habit wastes information when traits overlap or move together, as blood fats do.
Their tool, MultiPhen, flips the usual regression around and asks which blend of traits best matches a given variant. In simulations, the joint approach found far more true signals, including variants affecting several traits or only one. Unlike some rival joint methods, it did not invent false positives on the uneven data of case control studies. On real Finnish birth cohort data, it found about a fifth more known variant links than the standard approach. Its trait blends echoed the formula doctors use to compute cholesterol.
Catalog record
| CREATOR | O'Reilly PF, Hoggart CJ, Pomyen Y, Calboli FC, Elliott P, Jarvelin MR, Coin LJ. |
|---|---|
| YEAR | 2012 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83 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:
MultiPhen: Joint Model of Multiple Phenotypes Can Increase Discovery in GWAS By Paul F O’Reilly, Clive J Hoggart, Yotsawat Pomyen, Federico C F Calboli, Paul Elliott, Marjo-Riitta Jarvelin, Lachlan J M Coin, Stacey Cherny Published in: PLoS ONE DOI: 10.1371/journal.pone.0034861 ABSTRACT Abstract The genome-wide association study (GWAS) approach has discovered hundreds of genetic variants associated with diseases and quantitative traits. However, despite clinical overlap and statistical correlation between many phenotypes, GWAS are generally performed one-phenotype-at-a-time. Here we compare the performance of modelling multiple phenotypes jointly with that of the standard univariate approach. We introduce a new method and software, MultiPhen, that models multiple phenotypes simultaneously in a fast and interpretable way. By performing ordinal regression, MultiPhen tests the linear combination of phenotypes most associated with the genotypes at each SNP, and thus potentially captures effects hidden to single phenotype GWAS. We demonstrate via simulation that this approach provides a dramatic increase in power in many scenarios. There is a boost in power for variants that affect multiple phenotypes and for those that affect only one phenotype. While other multivariate methods have similar power gains, we describe several benefits of MultiPhen over these. In particular, we demonstrate that other multivariate methods that assume the genotypes are normally distributed, such as canonical correlation analysis (CCA) and MANOVA, can have highly inflated type-1 error rates when testing case-control or non-normal continuous phenotypes, while MultiPhen produces no such inflation. To test the performance of MultiPhen on real data we applied it to lipid traits in the Northern Finland Birth Cohort 1966 (NFBC1966). In these data MultiPhen discovers 21% more independent SNPs with known associations than the standard univariate GWAS approach, while applying MultiPhen in addition to the standard approach provides 37% increased discovery. The most associated linear combinations of the lipids estimated by MultiPhen at the leading SNPs accurately reflect the Friedewald Formula, suggesting that MultiPhen could be used to refine the definition of existing phenotypes or uncover novel heritable phenotypes. INTRODUCTION Genome-wide association studies aim to identify associations between genotype and phenotype. While genotypes are well-defined biological entities, phenotypes are defined more subjectively and can relate to numerous biological processes. There has been much effort to characterise well-defined phenotypes that may correspond more to specific biological function for use in genome-wide association studies, but the definition of some phenotypes remains somewhat ad-hoc. Type 2 diabetes, the subject of the first major genome-wide association study [1], is diagnosed using a debated blood glucose threshold [2], the Metabolic Syndrome is commonly based on observing three of five criteria [3], while neuropsychiatric disorders rely on a complex range of overlapping clinical characteristics for diagnosis. While most quantitative traits are less ambiguously defined, many are defined as mathematical functions of several measurements in an attempt to better capture the underlying biology than individual quantities. There have been many more genetic determinants of Body-mass-index (BMI), defined as the ratio of weight to height squared, discovered than those of weight [4], and the significance of rs8050136 in FTO was reported to be 11 orders of magnitude more significant for BMI than weight [5]. Illig et al. [6] investigated genetic association with metabolites and all metabolite ratios, using metabolite ratios as proxies for enzymatic reaction rates. A primary aim of this study is to fully assess the performance of a multivariate approach to genome-wide association studies in comparison to that of the usual single-phenotype strategy. Rather than exploiting an available multivariate method, we introduce a new fast and interpretable approach, which addresses some of the limitations of previously suggested methods. In particular, we propose that linear combinations (equivalent to ratios on pre log-transformed phenotypes) of directly measured phenotypes may often capture unmeasured aspects of complex biological networks, such as reaction rates, protein mediators or other uncharacterised or clinically-inferred phenotypes. Modelling the association between linear combinations of phenotypes and the genotypes at each SNP may uncover genetic association hidden to both single phenotype GWAS and those based on a priori defining a phenotype as a fixed function of traits. Such an approach circumvents the problems of phenotype definition by enabling association of genotype with flexible linear combinations of robust measurements, and could potentially be exploited by biologists and clinicians to define phenotypes that better reflect underlying biological processes. We introduce a new method, MultiPhen, that rapidly performs GWAS on multiple phenotypes by identifying the linear combination of the phenotypes most associated with genotype at each SNP. This is achieved by reversing the regression, such that genotype is regressed on phenotype, rather than phenotype-on-genotype as in the standard GWAS approach. By applying ordinal regression (proportional odds logistic regression), which models the genotype data as ordinal, multiple phenotypes can then be jointly modelled as predictors of the SNP genotypes to test for multi-phenotype associations (see Methods). This model makes no assumption on the phenotype distribution and so can accommodate both binary and continuous measurements. Our test for association is a likelihood ratio test for model fit, which provides a P value for evidence of association between the SNP and the phenotypes. The usual genome-wide significance level is applied. While several approaches to multivariate analysis of GWAS have been introduced in recent years, the univariate approach is still routinely favoured [7]–[9]. We believe that this is partly due to a lack of thorough comparison between a multivariate and the univariate approach, which we seek to address here, and also potentially due to limitations in the multivariate methods, such as: not allowing joint analysis of continuous and binary phenotypes [10]–[11], having inflated type 1 error [11], reducing the effective sample size by requiring cross-validation [12], and not explicitly modelling the correlation structure between traits [13]–[14]. MultiPhen does not suffer from these problems, but the benefit of using MultiPhen over ‘canonical correlation analysis’ (CCA) [15], when testing one-SNP-at-a-time, is less clear due to their similarity. MultiPhen and CCA (as implemented by Ferreira and Purcell [15]) both test the linear combination of the phenotypes most associated with the genotypes of a SNP against a null hypothesis of no association, but while CCA treats the genotypes as normally distributed MultiPhen appropriately models the genotypes as ordinal. Based on this, we specifically investigate whether MultiPhen offers better performance than CCA (which is equivalent to ‘reversed’ linear regression, with SNP as outcome, and MANOVA). We show via simulation that MultiPhen has markedly higher power than the standard single-phenotype approach to detect SNP-phenotype associations in many scenarios, even for identifying variants that affect only one of the phenotypes, as a result of jointly modelling the phenotypes. In addition, there are several convenient aspects of adopting our approach. As the phenotypes are treated as predictors rather than outcomes, there is no need to ensure their normality via transformation. Our MultiPhen software (available as a CRAN package for R) appropriately handles imputed SNP data and CNVs, can optionally perform a score test, and can perform univariate linear (standard) and ordinal (reversed) regressions for comparison with multivariate results (see Methods). A further advantage of our approach is its computational speed; applying MultiPhen to 10 traits is around 1.3 times faster than the corresponding set of standard single-phenotype analyses. With a natural test for combining results across studies as the (weighted) sum of the likelihood