when the genetic variants play a known and relatively specific role in the pathway of interest (e.g., ALDH2 and alcohol consumption), but these will capture a smaller proportion of the variance in the exposure than polygenic risk scores. Third, when adequate genetic variants have been identified for both the exposure and the outcome, bidirectional Mendelian randomization can be used to determine with greater confidence the likely direction of any causal relationship [33]. Fourth, a range of sensitivity analyses exist that can inform the interpretation of the findings, such as the extent to which (biological) pleiotropy has an impact on the causal estimates derived from conventional Mendelian randomization methods. This may be particularly relevant when polygenic risk scores comprising variants that act on a range of biological pathways are used. These methods include Mendelian randomization Egger regression [37] and the Kang median instrument approach [38]. Those relationships that survive this hierarchy of approaches are strong candidates for further interrogation in mechanistic or experimental studies.
