Work

Topics In The Quantitative Analysis of Complex Trait Genetic Architectures

Year: 2019
Type: dissertation
Abstract: The unprecedented availability of whole-genome data in recent years has allowed researchers to ask new questions regarding the genetic underpinnings of human traits while simultaneously introducing numerous theoretical and computational challenges. The vast majority of commonly studied phenotypes have revealed themselves to be complex traits—outcomes influenced by numerous genetic and non-genetic factors, the former of which remain largely unknown. Large samples of high-dimensional data are required to effectively study the genetic architectures of such traits and the development of new theory and methods for effectively analyzing such data have become essential. The present dissertation investigates several topics in this emerging area in the context of three studies. In the first study, Falsification of candidate gene hypotheses for major depression, we apply genome-wide methods and probabilistic arguments to the critical interrogation of previous “candidate gene” hypotheses regarding the genetics of major depression. We demonstrate that modern, well-powered samples provide little evidence supporting these pre-whole genome era hypotheses and that these discrepancies cannot be explained by measurement error or non-additive effects, concluding that the majority of published findings in this area represent false positives. In the second study, Stochastic Lanczos residual maximum likelihood algorithms, we introduce two novel algorithms for residual maximum likelihood (REML) estimation of genomic variance components in the context of linear mixed-effects models. The principle of Krylov subspace shift-invariance is applied to exploit problem structure and speed computation beyond existing methods, which we demonstrate via theoretical argument and numerical experiments. In the final study, Assortative mating and whole-genome heritability estimation, we present on-going work characterizing the effects of assortative mating on commonly used heritability estimation procedures. We provide analytic and simulation-based arguments demonstrating that three widely used estimators demonstrate substantial bias when assumptions regarding independence of causal variants are violated.
Author: Border, Richard
Language: English
Subfield: Genetics
Open Access status: green
OpenAlex ID: W7062470608
Cites: 86
Cited by:
Related to: 10