Mendelian randomization accounting for horizontal and correlated pleiotropic effects using genome-wide summary statistics.

Mendelian randomization (MR) is a valuable tool for detecting evidence of causal relationships between pairs of traits. Opportunities to apply MR are growing rapidly as the number of genome-wide association studies (GWAS) with publicly available summary statistics grows. Unfortunately, existing MR methods are prone to false positives caused by pleiotropic variants. Correlated pleiotropy, which arises when genetic variants affect both traits through a heritable shared factor, is a particularly challenging problem and is not addressed by most existing methods. Additionally, most MR methods only use genome-wide significant loci, which can limit power and introduce bias. We propose a new method (Causal Analysis Using Summary Effect Estimates; CAUSE) that uses genome-wide summary statistics to identify patterns that are consistent with causal effects, while accounting for pleiotropic effects, including correlated pleiotropy. We demonstrate in simulations that CAUSE is much better at controlling false positive rate in the presence of pleiotropic effects than other methods. We apply CAUSE to study relationships between pairs of complex traits and between blood cell composition and autoimmune disorders. We find that CAUSE detects causal relationships with strong literature support, including an effect of blood pressure on heart disease risk that is not found using other methods. Our results suggest that many pairs of traits identified as causal using alternative methods may be false positives driven by pleiotropic effects.