Sponsored by:
A goal of genome-wide association studies (GWAS) is to estimate the causal effects of alleles carried by an individual on that individual ("direct genetic effects"). Genetic variation within a family is random, so one can obtain unbiased estimates of direct genetic effects from individuals with both parents genotyped. However, parental genotypes are often missing. By imputing the genotypes of missing parent(s) from observed parent and offspring genotypes, as in the software package snipar, one can obtain estimates of direct genetic effects from any individual with at least one genotyped sibling or parent. However, this is typically only a small fraction of the genotyped individuals in large-scale biobanks. We show that the sample of ‘unrelated' individuals provides information that constrains the set of possible direct effects, so that including unrelated individuals can increase the effective sample size for estimation of direct effects by up to 50 compared to using samples with genotyped relatives alone. We develop an efficient linear-mixed model approach that uses a sparse genetic relatedness matrix to model relatedness, and is able to estimate both population effects, as in standard GWAS, and direct effects, as in family-based GWAS, using the full sample of individuals, whether they have genotyped relatives or not. Our approach therefore unifies the family-based and standard GWAS approaches.