Fig 5. A directed acyclic graph of the causal inference.

X and Y are two correlated phenotypes (the DNA methylation level at a CpG site and a biomarker level). The instrumental variable (IV) is the cis-regulatory genetic variant that directly influences X. If the IV is influencing Y through X in this way, there is a causal relation between X and Y. If the IV is not influencing Y through X, there might be an unknown confounding factor influencing both X and Y (U2), or there might be a reverse causation (Y influences X). To be a valid instrument three criteria must be fulfilled: IV criterion 1) the genetic variant must not be associated with an unknown confounding factor (U1) that influences Y, and IV criterion 2) the genetic variant must not be directly associated with Y, and IV criterion 3) the genetic variant should be associated with X.