Figure 4: Simulations for Binary Outcome.

The bar-plots show the power to detect the mediating set when using the TS (yellow), MIN (green), MARG (orange), LIN (red), and QUAD (brown) procedures. The baseline scenario includes $m_s=$ 20 biomarkers per set, $q=$ 20 disjoint sets, $q_m=$ 1 mediating set, $q_{OD}=$ 0 one-dimensional sets, $q_{TD}=$ 0 two-dimensional sets, $m_M=$ 3 mediating biomarkers, $m_E=$ 0 noise-biomarkers, and a correlation of ${\rho }_M=$ 0. We evaluate the effect of varying a single parameter, keeping all other parameters set to their baseline value.