Skip to main content

Table 6 Type I error and power estimates in the presence of population admixture at alpha = 1e-6 using 100 replicates.

From: A three-stage approach for genome-wide association studies with family data for quantitative traits

  Phenotype data LM LME FBAT LM-LME 1 Three-stage
Type I error (r2 = 0) original phenotypes 0.63 0.08 0.02 0.08 0.02
  residuals (1) 0.59 0.07 - 0.07 -
  residuals (2) 0.13 0.01 - 0.01 -
Power r2 = 0.5 original phenotypes 0.43 0.39 0 0.39 0.31
  residuals (1) 0.45 0.37 - 0.37 -
  residuals (2) 0.17 0.14 - 0.14  
Power r2 = 0.8 original phenotypes 0.66 0.63 0.05 0.63 0.56
  residuals (1) 0.66 0.63 - 0.63 -
  residuals (2) 0.57 0.47 - 0.47 -
Power QTL original phenotypes 0.81 0.85 0.11 0.85 0.83
  residuals (1) 0.82 0.84 - 0.84 -
  residuals (2) 0.81 0.67 - 0.67 -
  1. Marginal phenotype distribution follows a normal distribution with variance 1. The additive genetic effect in assessing power is sqrt(0.005/2/0.3/0.7) = 0.109.
  2. LM_LME1: two-stage approach with LM at first stage with p-value cut-off 0.1 and LME at second stage with p-value cut-off 1e-6.
  3. Three-stage: LM at first stage with p-value cut-off 0.1, LME at second stage with p-value cut-off 1e-6 and FBAT at third stage with p-value cut-off 0.05/n, where n is the number of SNPs entering the third stage.
  4. The original simulated phenotypes and the residuals adjusted for 10 PC obtained from all 34,625 SNPs on chromosome 1 and 100 admixture SNPs (1), and from 100 admixture SNPs (2) are analyzed. The 100 admixture simulated SNPs have expected allele frequency 0.3, F ST = 0.025, the offset value δ = 0.25, QTL variance is 0.005, and the polygenic variation is 0.3. When estimating power, 34,265 SNPs are not used.