MR-GENIUS uses gene-by-covariate interactions to generate a causal effect estimate that is robust to additive unmeasured confounding and violation of MR assumptions.
Within a one-sample MR setting with individual-level data, MR-GENIUS uses the differences in the variance of an exposure across subgroups of the genetic instrumental variable (IV), which are likely due to gene-by-covariate interactions. The method generates a single IV from the residuals of the regression between the exposure and each single nucleotide polymorphism (SNP) being used as an IV and uses this in a two-stage least squares (TSLS) model. Whilst MR-GENIUS does not require an interaction covariate to be explicitly identified, the method does rely on the presence of one or more gene-by-covariate interactions.