For each of 21 tested traits, we observed that functionally informed PRS using IMPACT captured more phenotypic variance than standard PRS (49.9% mean relative increase in R2; Fig. 5b, Extended Data Fig. 8 and Supplementary Tables 16–18). The mean phenotypic variance explained across traits by functionally informed PRS (R2=2.1%, s.e.m. = 0.4%) was greater than by standard PRS (R2 = 1.5%, s.e.m. = 0.3%, one-tailed paired Wilcoxon P < 4.8 × 10−7). For 19 of 21 traits, IMPACT-informed PRS significantly outperformed standard PRS (19 one-tailed difference of means P < 0.05); for platelet count P = 0.052 and for basophil count P = 0.40. Using 10,000 bootstraps of the PRS sample cohort, we found that the IMPACT-informed PRS R2 estimate was consistently greater than the standard PRS estimate for all traits except basophil count (all bootstrap P < .004; Methods and Supplementary Table 18). We observed the largest improvement for RA from R2 = 1.4% (s.d. = 0.33%) in the standard PRS to R2 = 4.1% (s.d. = 0.53%, one-tailed difference of means P < 9.8 × 10−6) in the
