Using phenotyped but ungenotyped relatives in genetic association tests
OA Version
Citation
Abstract
In some longitudinal studies, there are individuals for whom rich phenotypic data have been collected, but who died before providing DNA for genetic studies. Genotypes of their relatives are often available. The main question we address is how and when one should incorporate phenotyped but ungenotyped relatives into genetic association tests. For genotypes missing completely at random (MCAR) and a quantitative outcome, Visscher and Duffy (2006) inferred the power increase due to the inclusion of ungenotyped individuals using information from relatives ' genotypes for the case of a single genotyped single-nucleotide polymorpherm (SNP) and a single type of relative. We derive a theoretical formula for the power gain for a dichotomous outcome. We verify and extend the theoretical result with simulations of small or moderate sized pedigrees assuming a MCAR, missing at random (MAR), or not missing at\ random (NMAR) missingness mechanism. For quantitative and binary outcomes, we observe biased effect estimates in data sets that exclude subjects with MAR genotypes and in data sets that include imputed NMAR genotypes. For most situations, power increases when ungenotyped individuals are included using imputed genotypes. The missingness mechanism, heritability, minor allele frequency, and SNP-specific heritability are important factors in the change in power for dichotomous or quantitative outcomes.
We find that the increase in the test statistic from including individuals with genotypes imputed based on relatives ' genotypes compared to omitting these individuals is about half of what could be attained using the true genotypes if they were available. Therefore, we propose a phenotypically enriched genotypic imputation (PEGI) method to impute missing genotypes using observed phenotypes in addition to genotypes. Our simulations with MCAR genotypes show that, for a SNP with moderate to strong effect on a phenotype, PEGI improves power more than imputation based solely on genotypes without excess type I errors. The effect estimate is often biased when the outcome is used for imputation while it is unbiased when a phenotype unrelated with the outcome is used. Compared to using only the observed genotypes for imputation, the PEGI method may improve power for MCAR, MAR, or NMAR genotype data.
Description
Thesis (Ph.D.)--Boston University