Extended Data Figure 7: Comparison of P-values between PGS-adjusted and unadjusted models within BOLT-LMM.

The violin plots (with overlaid boxplots) show the distributions of differences in -log₁₀(P) values between unadjusted and PGS adjusted models. Panels in a show results for adjustment of out-of-sample PGS, where red indicates results for BOLT-LMM-Inf models and gold shows results for BOLT-LMM models. In the left panel, the N gene-trait pairs equal 176, 180, 175 and 182 (from left to right), while in the right panel the N values equal 262, 267, 271 and 282 (from left to right). Panels in b shows results for adjustment for in-sample PGS where blue indicates BOLT-LMM-Inf and gold indicated BOLT-LMM models. In the left panel, the N gene-trait pairs equal 175, 176, 177 and 174 (from left to right), while in the right panel the N values equal 256, 257, 269 and 268 (from left to right). In both a and b, the left panel results are restricted to associations reaching Bonferroni corrected significance in either analysis (PGS adjusted or unadjusted), while the right panel is restricted to association reaching conventional exome-wide significance in either analysis. Estimated d values (difference values from paired T-test) and P-values (from paired Wilcoxon signed rank tests) are added above each violin. In all BOLT runs, 240,699 high-quality pruned common variants were used for computation of the genetic relatedness matrix. P-values are two-sided and unadjusted for multiple testing. Boxplots: center line, median; box limits, upper and lower quartiles; whiskers, 1.5x interquartile range; points, outliers. Note: $\overline{\text{d}}$, estimated paired group difference; δ, difference; α, significance cutoff.