ABSTRACT

Results from a genome-wide association study (GWAS) can be used to generate a polygenic score (PGS), an individual-level measure summarizing identified genetic influence on a trait dispersed across the genome. For complex, behavioral traits, the association between an individual’s PGS and their phenotype may contain bias (from geographic, ancestral, and/or socioeconomic confounding) alongside the causal effect of the individual’s genes. We formalize the introduction of a different source of bias in regression models using PGSs: the effects of parental genes on offspring outcomes, known as genetic nurture. GWAS do not discriminate between the various pathways through which genes become associated with outcomes, meaning existing PGSs capture both direct genetic effects and genetic nurture effects. We construct a theoretical model for genetic effects and show that the presence of genetic nurture biases PGS coefficients from both naïve OLS (between-family) and family fixed effects (within-family) regressions. This bias is in opposite directions; while naïve OLS estimates are biased away from zero, family fixed effects estimates are biased toward zero. We quantify this bias using two novel parameters: (1) the genetic correlation between the direct and nurture effects and (2) the ratio of the SNP heritabilities for the direct and nurture effects.

摘要

来自全基因组关联研究（GWAS）的结果可用于生成多基因评分（PGS），这是一种个体水平的测量方法，总结了对分散在基因组中的性状的已鉴定遗传影响。对于复杂的行为特征，一个人的PGS及其表型之间的关联可能包含偏见（来自地理，祖先和/或社会经济混杂）以及该个体基因的因果关系。我们使用PGS在形式回归模型中正式引入了不同的偏见来源：父母基因对后代结局的影响，即遗传育种。GWAS不会区分基因与结果相关的各种途径，这意味着现有的PGS既捕获直接遗传效应又捕获遗传育种效应。我们构建了遗传效应的理论模型，并表明遗传育种的存在使单纯OLS（家庭间）和家庭固定效应（家庭内）回归的PGS系数产生偏差。这种偏向是相反的。虽然单纯的OLS估计值偏离零，但家庭固定效应估计值倾向于零。我们使用两个新颖的参数来量化这种偏倚：（1）直接效应和培育效应之间的遗传相关性；（2）直接效应和培育效应的SNP遗传比。家庭固定效应估计偏向零。我们使用两个新颖的参数来量化这种偏倚：（1）直接效应和培育效应之间的遗传相关性；（2）直接效应和培育效应的SNP遗传比。家庭固定效应估计偏向零。我们使用两个新颖的参数来量化这种偏倚：（1）直接效应和培育效应之间的遗传相关性；（2）直接效应和培育效应的SNP遗传比。