< Back to Article

Iterative Usage of Fixed and Random Effect Models for Powerful and Efficient Genome-Wide Association Studies

Fig 5

Power in structured populations at different levels.

Three methods were employed to examine these populations, including GLM, MLM and FarmCPU. The top panel (a to e) and bottom panel (f to j) display the low and high levels of population structure, represented by Arabidopsis and human populations, respectively. The dataset from Arabidopsis population consists of 1,178 individuals genotyped with 250,000 SNPs. The dataset from human population consists of 1,500 individuals genotyped with 500,000 SNPs. The population structures are displayed by the scatter plot on the first two principal components derived from 10% of SNPs sampled randomly from Arabidopsis thaliana (a) and human (f), respectively. Additive genetic effects were simulated with 10 and 100 QTNs. The QTNs were randomly sampled from all the SNPs in each dataset. Residuals with normal distribution were added to the genetic effect to form phenotypes with heritability of 0.5. Power was examined under different levels of FDR and Type I error. A positive SNP is considered a true positive if a QTN is within a distance of 50,000 base pairs on either side, otherwise is considered a false positive. Power under different levels of FDR is displayed in subfigures b, c, g, and h. Power under different levels of Type I error is displayed in subfigures d, e, i, and j.

Fig 5