Extended Data Fig. 9. Mendelian randomisation analysis.

a) Causal effects of protein levels on disease risk estimated through two-sample Mendelian randomisation analysis of pQTL summary statistics and disease GWAS summary statistics. OR: consensus estimate of the odds ratio conferred per standard deviation increase in protein levels across five Mendelian randomisation methods. * Estimated causal effect is directionally consistent with PGS-protein associations in Fig. 1b. 95% CI: 95% confidence interval. P-value: Two-sided P-value obtained by averaging Z-scores across five Mendelian randomisation methods. Entries are greyed out where P > 0.05, and red where P < 0.0038 (Bonferroni correction for 13 tests). Pleiotropy P-value: two-sided P-value for the intercept term in Egger regression, indicating where P < 0.05 confounding of the causal estimate by horizontal pleiotropy. Full summary statistics including exact P-values are detailed in Supplementary Table 6. b) Dose response curves showing the estimated causal effect of changes in protein levels on disease risk for each protein and disease. Points on each plot show the cis-pQTLs used as genetic instruments for each test. On the x-axes, points show the standard deviation change in protein levels per copy of the minor allele in the pQTL summary statistics, and horizontal bars show + /- the standard error. On the y-axes, points show odds ratio conferred per copy of the minor allele in the GWAS summary statistics, and vertical bars indicate show + /- the standard error. Effect sizes, standard errors, and exact two-sided P-values from pQTL and GWAS summary statistics are detailed in Supplementary Table 7. The slope of the orange dashed line corresponds to the estimated causal effect (consensus Odds Ratio from a). The yellow ribbon shows the 95% confidence interval for the estimated causal effect (slope), accounting also for the 95% confidence interval for the intercept term in Egger regression.