Phenotypic covariance across the entire spectrum of relatedness for 86 billion pairs of individuals

Journal article

Estimation of the causative factors leading to human trait variation is alongstanding project of statistical and quantitative genetics. Heritability, the proportion of phenotypic variance attributable to genetic factors, is a particularly important quantity for understanding the genetic architecture of disease and social science traits. This study evaluates the statistical properties of a novel “sibling regression” estimator which uses exact measures of genetic relatedness between full siblings to estimate trait heritability. As classical twin estimates rely on expected degrees of relatedness between monozygotic and dizygotic twins across families, the estimator of focus here provides heritability estimates orthogonal to those of classical methods. The analyses presented 1) confirm unbiasedness of the estimator using data simulated from real genotypes, 2) estimate the standard errors of estimates of non-additive variance across varying n and 3) quantify effects of researchers’ prior beliefs on acceptance of sibling regression estimates. Analyses 2) and 3) suggest that sibling regression estimates of non-additive variance are precise only at large sample sizes (i.e. rs> 100?) and that prior effects only become negligible for additive estimates at these same sample sizes.

Authors: Kemper, KE; Yengo, L; Zheng, Z; Abdellaoui, A; Keller, MC

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Nature Research
• Journal: Nature Communications
• Volume: 12
• Issue: 1
• Pages: 1050-1050
• Article number: 1050
• Publication date: 2021-02-16
• DOI: 10.1038/s41467-021-21283-4
• EISSN: 2041-1723
• ISSN: 2041-1723
• Language: Undetermined
• Keywords: Gene; Mathematics; Evolutionary biology; Genetics; Phenotype; Physics; Spectrum (functional analysis); Computational biology; Statistics; Biology; Covariance