Genome Wide Association studies (GWAs) are widely used to investigate the connection between genotypic and phenotypic variation with respect to a given trait (e.g. a given disease). Assessing the statistical power of such studies is crucial. Power is empirically estimated by simulating realistic samples under a disease model H1. For this purpose, the gold standard consists in simulating the genotypes given the observed phenotypes (case or control) ; thus ensuring that the total number of cases stays unchanged. This method is implemented in the software of reference Hapgen. I will present an alternative approach for simulating samples under H1 that does not require generating new genotypes for each simulation but only phenotypes. This method is based on a backward sampling algorithm and make it possible to simulate new phenotypic datasets under the constraints that a) the phenotypes are in accordance with the corresponding observed genotypes under the chosen model H1 ; b) the total number of cases is the same as in the observed dataset. I will show that our backward sampling algorithm outperforms other standard approaches such as simple rejection algorithm and MCMC. Moreover our algorithm is faster than Hapgen. At last, I will discuss the results of a power analysis on a fictive GWAs we conducted on real data from the 1000 Genomes Project.

This is joint work with Gregory Nuel (MAP5, Université Paris Descartes), Christine Sinoquet and Raphael Mourad (Université de Nantes).