Duke BERD Seminar: Genetic association models for related samples and population structure

The goal of genetic association studies is to identify genetic variants that are associated with a trait or phenotype, which is often a disease. This problem is challenging to solve because there are often millions of tests to be performed, numerous causal variants, non-genetic factors that are often unknown, and structure in the genetic data.

Genetic data can be structured, in the sense that samples are not independent, under a variety of real-world scenarios, including the presence of close or distant relatives, and multiple genetic ancestries such as in multiethnic studies and recently admixed populations such as African-Americans and Hispanics. Genetic data is also structured due to linkage disequilibrium, which results in correlations between variants that are in physical proximity. In addition to reviewing all of these concepts, the presentation will focus on the two most common models for this task, namely linear regression with principal components as covariates, and linear mixed-effects models, and includes some recent evaluations comparing the two.

Intended audience: biostatisticians and clinical/translational researchers
Technical level: intermediate

Presenter: Alex Ochoa, PhD