A1475
Title: Effects of population stratification on gene-gene interaction detection
Authors: Neelotpal Das - Nagasaki University (Japan) [presenting]
Masao Ueki - Nagasaki University (Japan)
Abstract: Genome-wide association studies (GWASs) have helped identify several causal variants underlying a wide range of complex diseases. To further understand the genetic architecture of these diseases, researchers often study epistasis or gene-gene interaction, which is defined as the departure from a purely additive model between two or more loci and is typically detected by testing an interaction term in a regression model. However, a key challenge in this analysis is population structure, which refers to the presence of distinct ancestral subpopulations within a sample and is known to inflate false positive rates in GWAS. How population stratification affects epistasis detection under three modelling frameworks is investigated: (i) standard linear regression, (ii) EIGENSTRAT, a widely used method for correcting false positives due to population structure in GWAS and (iii) a novel regression-based method for this setting. For each of the three models, the bias of the interaction term for a quantitative phenotype is analytically derived, where individuals are sampled from two distinct populations. Building on analytical results, simulation studies are conducted across a wide range of population structure scenarios, comparing the false positive rates produced by each model. Finally, the analytical and simulation-based findings are validated using real Alzheimer's disease data.
