Fig. 2

Simulation results. A–C compare the type-I error and statistical power of SUPERGNOVA, GNOVA, and LDSC in global genetic covariance estimation. We use the proportion of p values that are less than 0.05 to estimate type I error or statistical power when true parameters are zero or nonzero, respectively. A Two GWASs were simulated on two non-overlapping datasets (set 1 and set 2). B GWASs were simulated on two datasets with a 50% sample overlap (set 1 and set 3). C Two GWASs were both simulated on the same dataset (set 1) with a 100% sample overlap. D–F compare SUPERGNOVA and ρ-HESS on local genetic covariance estimation using GWASs D without sample overlap (set 1 and set 2), E with a partial sample overlap (set 1 and set 3), and F with a complete sample overlap (set 1 only)