Extended Data Fig. 10: Receiver operating characteristic curves using our shrinkage method as a classifier to identify true causal variants in simulated data.

Using hail, we simulated GWAS summary statistics for 80 binary phenotypes with heritability sampled uniformly between 0.1 and 0.6. We repeated this for four scenarios where the probability of a variant being causal (pi) is 0.001, 0.002, 0.005, and 0.01. We then applied our shrinkage procedure and classified those variants with a posterior probability of being from the null model of <10% as being causal (colored dots on lines represent this threshold). See Supplementary Table 16 for details on classification performance.