1. Unified Sequence-Based Association Tests Allowing for Multiple Functional Annotations and Meta-analysis of Noncoding Variation in Metabochip Data 
Zihuai He, Bin Xu, Seunggeun Lee, Iuliana Ionita-Laza 
The American Journal of Human Genetics 
Volume 101, Issue 3, Pages (September 2017)
DOI: /j.ajhg
Copyright © 2017 American Society of Human Genetics Terms and Conditions

2. Figure 1 Type I Error Rate of Integrating Multiple Functional Scores
Empirical type I error rate was estimated as the proportion of p values less than α = 0.001 on the basis of 104 replicates. The different curves correspond to quantitative trait and dichotomous trait. The dashed line corresponds to the α level.
The American Journal of Human Genetics  , DOI: ( /j.ajhg )
Copyright © 2017 American Society of Human Genetics Terms and Conditions

3. Figure 2 Power Comparison with Existing Methods
Each bar presents the empirical power estimated as the proportion of p values less than α = 2.5 × 10−6 on the basis of 104 replicates. Abbreviations are as follows: P1 and P2, the probability that the functional score 1 and 2, respectively, can predict the risk effect; O, original dispersion (SKAT), burden, or combined (SKAT-O) test with beta(1,25) as weights; W1, weighted dispersion (SKAT), burden, or combined (SKAT-O) test with beta(1,25) × functional score as new weights; U1 and U1+2, proposed unified dispersion, burden, or combined test incorporating one and two functional scores, respectively.
The American Journal of Human Genetics  , DOI: ( /j.ajhg )
Copyright © 2017 American Society of Human Genetics Terms and Conditions

4. Figure 3 Power of Integrating Multiple Functional Scores
βj=0.8|log10mj|, where mj is the MAF for the jth variant. The causal proportion is 10%. Each curve presents the empirical power estimated as the proportion of p values less than α = on the basis of 104 replicates. The solid curves correspond to the power of the proposed unified tests incorporating 0 (the original tests) to 20 all predictive, all non-predictive, or a mixture of predictive (first three) functional scores. The black line corresponds to the ideal setting of applying the original tests (dispersion, burden, and combined) by using the causal variants only. The dashed curves correspond to the average correlations between the true effect sizes and the weight that results in the minimum p value.
The American Journal of Human Genetics  , DOI: ( /j.ajhg )
Copyright © 2017 American Society of Human Genetics Terms and Conditions

5. Figure 4 Power of Integrating Multiple Functional Scores
βj=0.2|log10mj| for quantitative traits and βj=0.4|log10mj| for dichotomous traits, where mj is the MAF for the jth variant. The causal proportion is 50%. Each curve presents the empirical power estimated as the proportion of p values less than α = on the basis of 104 replicates. The solid curves correspond to the power of the proposed unified tests incorporating 0 (the original tests) to 20 all predictive, all non-predictive, or a mixture of predictive (first three) functional scores. The black line corresponds to the ideal setting of applying the original tests (dispersion, burden, and combined) by using the causal variants only. The dashed curves correspond to the average correlations between the true effect sizes and the weight that results in the minimum p value.
The American Journal of Human Genetics  , DOI: ( /j.ajhg )
Copyright © 2017 American Society of Human Genetics Terms and Conditions

6. Figure 5
Individual Variant p Value and Eigen-Phred Score for the Rare (MAF < 0.05) Noncoding Variants in the Metabochip Region Indexed by rs
The American Journal of Human Genetics  , DOI: ( /j.ajhg )
Copyright © 2017 American Society of Human Genetics Terms and Conditions