Figure 5. Uncertainty in real data and its influence on genetic risk stratification.

(a) Example of posterior PRS distributions for individuals with certain below-threshold (dark blue), uncertain below-threshold (light blue), uncertain above-threshold (light yellow), and certain above-threshold (dark yellow) classifications for HDL. Each density plot is a smoothed posterior PRS distribution of an individual randomly chosen from that category. The solid vertical lines are posterior means. The shaded areas are 95% credible intervals. The red dotted line is the classification threshold. (b) Distribution of classification categories across 11 traits (t=90%, ρ=95%). Each bar plot represents the frequency of testing individuals who fall into each of the four classification categories for one trait. The frequency is averaged across five random partitions of the whole dataset. (c) Correlation of PRS rankings of test individuals obtained from two MCMC samplings from the posterior of the causal effects. For each trait, we draw two samples from the posterior of the causal effects, rank all individuals in the test data twice based on their PRS from each sample, and compute the correlation between the two rankings across individuals. Each violin plot contains 5,000 points (1,000 pairs of MCMC samples and five random partitions).