Much of my recent work has focused on how rapidly the microbiome changes over time. There's increasing evidence that a healthy microbiome tends to be fairly stable, whereas in a variety of diseases, the microbiome may be less stable. It turns out that "stability" is not a straightforward thing to quantify and compare between high-dimensional bacterial communities, so I'm working on ways to quantify it and to formally test whether volatility differs based on disease state. 

Data: Moving Pictures study, Caporaso et al (2011), Genome Biology. 

A second major area of research is kernel machine regression models, which test whether an entire microbial community is associated with the outcome. For example, we could answer a question like: are the throat microbiomes of smokers more similar to other smokers than to non-smokers? On the graph above, points that are closer together represent more similar microbiomes -- so it seems that yes, there is some association between smoking status and throat microbiome. Kernel machine regression models would help us formally test this association. I've worked specifically on time-to-event outcomes (MiRKAT-S) and incorporating more omnibus tests into the MiRKAT R package. 

Data: Charlson et al (2010) throat microbiome dataset.