Measuring weak effects in high dimensional mediation analysis

Abstract

Existing mediation analysis methods have often fallen short in accurately quantifying the contribution of omics mediators, particularly those with weak effects. To address this issue, we propose two new variance-based causal measures for the global mediation effect. Then, we develop a flexible and computationally efficient estimation procedure based on a mixed-effects working model. Through this approach, we are able to accurately quantify the total mediation effect and discover the weak effects that are largely mis-estimated by existing methods. The proposed approach is general and complements the existing mediation analysis methodologies by offering new perspectives on the global and weak effects.