Cluster analysis and PCA for modeling population structure

Case-control studies for association are widely used for finding genetic variants causally associated with phenotypes. Unfortunately, population structure can induce false positives. For instance, if cases and controls have different genetic backgrounds, differences in frequencies of distinct forms of variants might be due to differences in ancestral population of origin. Traditional approaches to control for the effects of population stratification include eigen-analysis, cluster analysis and matching based on genetic markers, are employed to improve the modeling of structure. Our approach goes further in that we show how to systematically obtain optimal matching and how to determine outlying subjects that cannot be successfully matched to others in the available registry. Simulations and an application to real data show improved results applying the new method.