Supplementary Figure 4: Power of BUHMBOX in polygenic modeling. | Nature Genetics

Supplementary Figure 4: Power of BUHMBOX in polygenic modeling.

From: A method to decipher pleiotropy by detecting underlying heterogeneity driven by hidden subgroups applied to autoimmune and neuropsychiatric diseases

Supplementary Figure 4

We simulated GWAS and examined the power of BUHMBOX by including moderately significant loci up to P < 0.01. We modeled the genetic architecture of disease using rheumatoid arthritis, on the basis of results from Stahl et al. (Nat. Genet. 44, 483–489, 2012). To simulate 2,231 causal variants, we combined 71 independent known rheumatoid arthritis risk loci with an additional 2,160 loci sampled from the joint posterior distribution of RAF and OR values presented in Stahl et al. For null loci, we also used the null RAF distribution presented in Stahl et al. Given this disease model, we simulated a GWAS with 3,964 cases and 12,052 controls (sample sizes from Stahl et al.), assuming disease prevalence of 0.01. Given these GWAS results, we used only the top k GWAS loci defined by P-value threshold t and their observed OR estimates for BUHMBOX power simulations. We assumed N = 5,000 and π = 0.5 for power evaluation and tried different P-value thresholds t from 5 × 10−8 to 0.01.

Back to article page