Figure 1.
MAF Threshold: Rejection Sampling (Black) vs. Dosage Approximation (Grey).
The MSE (Y-axis) between sampled genotype probability and true
using rejection sampling (black) and dosage approximation (grey) is compared across a spectrum of R2.
Figure 2.
Spearman Correlation with Gold Standard P-values.
The Spearman correlation (Y-axis) between gold standard p-values and p-values from different methods is displayed across a spectrum of MAF and R2.
Figure 3.
The statistical power (Y-axis) of the different methods is shown across a spectrum of R2 and MAF.
Table 1.
Rejection Sampling vs. Dosage Approximation for Estimation.
Table 2.
Type I Error at Significance Level = 5E-02.
Table 3.
Type I Error at Significance Level = 5E-05.
Figure 4.
Q–Q Plot for Null Variants with Low Imputation Quality in the CLHNS Study.
The observed (Y-axis) vs. expected (X-axis) –log10[p-values] are shown for 1,135 SNPs in the CLHNS data set. These SNPs are considered to be under the null hypothesis (true p-value >5×10−6), and all have low imputation quality (R2<0.3).
Table 4.
Associated Variants with R2≤0.3 in the CLHNS Study.
Table 5.
Associated Variants with MAF <5% in the WHI Study.
Figure 5.
Computing Time: Mixture Method vs EM-LRT-Prob.
The computing time of the Mixture method and our proposed EM-LRT-Prob method is displayed across a range of sample sizes. For each sample size, computing time is averaged across 2,000 simulated datasets.
Table 6.
One-sample T-test for Type I Error.