RT Journal Article SR Electronic T1 Testing the principles of Mendelian randomization: Opportunities and complications on a genomewide scale JF bioRxiv FD Cold Spring Harbor Laboratory SP 124362 DO 10.1101/124362 A1 M Taylor A1 KE Tansey A1 DA Lawlor A1 J Bowden A1 DM Evans A1 Smith G Davey A1 NJ Timpson YR 2017 UL http://biorxiv.org/content/early/2017/04/07/124362.abstract AB Background Mendelian randomization (MR) uses genetic variants as instrumental variables to assess whether observational associations between exposures and disease reflect causal relationships. MR requires genetic variants to be independent of factors that confound observational associations.Methods Using data from the Avon Longitudinal Study of Parents and Children, associations within and between 121 phenotypes and 13,720 genetic variants (from the NHGRI-EBI GWAS catalog) were examined to assess the validity of MR assumptions.Results Amongst 7,260 pairwise comparisons between the 121 phenotypes, 2,188 (30%) provided evidence of association, where 363 were expected at the 5% level (observed:expected ratio=6.03; 95% CI: 5.42, 6.70; χ2=9682.29; d.f. =1, P≤1x10-50). Amongst 1,660,120 pairwise associations between phenotypes and genotypes, 86,748 (5.2%) gave evidence of association at the same threshold, where 83,006 were expected (observed:expected ratio=1.05; 95% CI: 1.04, 1.05; χ2=117.57; d.f. =1, P=2.15x10-27). Amongst 1,171,764 pairwise associations between the phenotypes and LD pruned independent genetic variants, 60,136 (5.1%) gave evidence of association, where 58,588 were expected (observed:expected ratio=1.03; 95% CI: 1.03, 1.08; χ2= 43.05; d.f. = 1, P=5.33x10-11).Conclusion These results confirm previously observed patterns of phenotypic correlation. They also provide evidence of a substantially lower level of association between genetic variants and phenotypes, with residual inflation the likely product of indistinguishable real genetic association, multiple variables measuring the same biological phenomena, or pleiotropy. These results reflect the favorable properties of genetic instruments for estimating causal relationships, but confirm the need for functional information or analytical methods to account for pleiotropic events.