Analysis of all rare variants and analysis of only nonsynonymous rare variants generated qualitatively similar results. We thus present here the results for the analysis of nonsynonymous variants.
Despite the simulated rate of missing genotype data, we were able to detect the association of the causal FLT1 gene with Q1 in 100% of the runs (Figure ). The causal KDR gene was detected in 23.2% of replicates with a 90% random call rate, in 23.8% of replicates for our allele-specific model, and in 26.8% of replicates when there was no missing genotype data. For the rest of the causal genes for Q1, we had low power to detect associations with ARNT (up to 3.9% of replicates, depending on the call rate) and HIF1A (0.5% of replicates with a 90% random call rate, 0.6% with a 95% random call rate, and just 0.3% with our allele-specific model). Interestingly, the power of HIF1A was lower in our models with higher random call rates, although this is likely to reflect stochastic variation in our simulations. The type I error rate for the detection of association with Q1 was higher than expected in several noncausal genes, including OR2T34, OR2T3, NOMO1, and HLA-B. The high type I error rates remained, irrespective of the call rate; for example, association with OR2T34 was detected in 78.4% of replicates for a 90% random call rate and increased to 85.9% of replicates when there were no missing data. Thus these type I errors have not occurred as a result of missing genotypes but because of extended linkage disequilibrium between rare variants across chromosomes.
Power to detect associations for Q1 phenotype using nonsynonymous markers. All gene regions affecting Q1 phenotype are presented.
For Q2, we had power to detect association with several causal genes, namely, BCHE, LPL, SIRT1, SREBF1, and VLDRL, but only in a small percentage of replicates (up to 4.2% with a 99% call rate) (Figure ). The type I error rates for Q2 were lower than those for Q1. For Q4, which is not associated with variants in any gene, the false-positive error rate was never higher than 1.1% (Figure ).
Power to detect associations for Q2 phenotype using nonsynonymous markers. All gene regions affecting Q2 phenotype are presented.
False-positive associations for Q4 phenotype using nonsynonymous markers. The ten most associated gene regions are presented.
For the disease (CC) phenotype, we were able to detect the causal FLT1 gene locus in 5% of replicates with no missing genotype data, 5.3% of replicates with a 99.9% call rate and the allele-specific model, and only 1.6% of replicates with a 90% random call rate. The second-ranked causal gene was PIK3C3, identified in just 1.7% of replicates with no missing genotype data (Figure ). In addition, the false-positive OR2T3 and OR2T34 genes, which showed associations with the Q1 phenotype, showed associations in 1.1% and 1.0%, respectively, of the runs with the full data set accordingly.
Power to detect associations for disease status using genes underlying disease liability and the genes affecting Q1 and Q2 phenotypes using nonsynonymous markers. Only gene loci with power larger than 0 are presented.