Translational microbiome research exists under the promise of uncovering microbial level interventions, which will promote health and eliminate disease. However, most of the available observational and experimental study designs do not allow for unambiguous determination of the direct causal role of the microbiome, as alternative interpretations are always plausible. For example, an antimicrobial intervention to manipulate the microbes may also have a direct effect on the measured outcome. For this reason, causal mediation analysis is desirable to estimate the extent to which microbes and intervention are responsible for observed study phenotypes. The univariate single mediator model framework achieves such inference by estimation of linear regressions. Even in the simplest case of univariate intervention and response, the microbiome measurements are highly multivariate, under-sampled, compositional and over-dispersed, which imposes modeling challenges in translating the single mediator model to microbiome data. Using distance-based energy statistics, I will derive a multivariate causal mediation framework suitable for application to microbiome data.