A simulation study comparing three methods for individual admixture estimation.

MARTINEZ MARIGNAC VL; M.R. VAUGHN; I. HALDER; M.D. SHRIVER; E.J. PARRA

Toronto

Congreso; American Society of Human Genetics Annual Meeting 2004; 2004

American Society of Human Genetics

Population-based association studies are a promising approach for mapping genes involved in complex traits and diseases. However, these studies are prone to false positive results due to population structure. One potential source of population structure is admixture stratification (e.g. variation of individual ancestry between individuals in an admixed sample). Admixture stratification can have a profound effect in association studies in admixed populations, potentially leading to significant results for a substantial number of genetic markers that are not functionally related to the complex trait being studied. The problems outlined above can be eliminated by controlling for individual ancestry in the statistical analysis. A number of software programs are available that implement different strategies to estimate individual ancestry in admixed populations. However, there has not been a systematic evaluation of the relative merits of these programs. In this study, we simulated different admixture scenarios and evaluated three widely used programs to estimate individual ancestry: STRUCTURE, ADMIXMAP and a maximum likelihood method (ML, Chakraborty 1986). We created, by simulation, admixed samples with different ancestral contributions and different levels of admixture stratification. The programs STRUCTURE, ADMIXMAP and ML were run on the simulated population data to determine the ability of each program to correctly identify individual admixture and population stratification under the alternative admixture scenarios. We explored the effect of factors such as the number of parental populations, the number of informative markers, the availability of parental frequency data (only tested for STRUCTURE and ADMIXMAP) and the effect of misspecification of parental frequencies (only tested for ADMIXMAP).