No document available.
Abstract :
[en] It is generally believed that epistasis makes an important contribution to the genetic architecture of complex disease, and numerous statistical and bioinformatics methods have been developed to detect it.
We compare several state-of-the-art epistasis detection methods in terms of empirical power, type-I error control, and CPU time. The methods compared include Model-Based Multifactor Dimensionality Reduction (MB-MDR) [1, 2], BOolean Operation-based Screening and Testing (BOOST) [3], EPIBLASTER [4], Random Jungle (RJ) [5], Logistic Regression and PLINK.
Our comparative study is based on an extensive simulation study using different two-locus models, exhibiting both main effects and epistasis [3]. In these simulations, 100 SNPs are generated, no LD between them. All genotypes are assumed to be in Hardy-Weinberg equilibrium. Furthermore, 2 disease-associated SNPs are selected, with MAFs set to 0.1, 0.2 and 0.4. The MAFs of the non-disease associated SNPs are uniformly distributed on [0.05, 0.5]. In order to achieve high accuracy in empirical power estimation, all simulation settings involve 1000 replicates. All methods are applied to WTCCC Crohn's Disease data.
[1] Calle, M.L. et al. (2008), Tech. Rep. No. 24, Dep. of Systems Biology, Univ. de Vic
[2] Cattaert, T. et al. (2011), Ann. Hum. Gen. 75, 78-89
[3] Wan, X. et al. (2010), Am. J. Hum. Gen. 87, 325-340
[4] Kam-Thong, T. et al. (2011), Eur. J. Hum. Gen. 19, 465-471
[5] Schwartz, D.F. et al. (2010), Bioinf. 26, 1752-1758
