The human microbiome is a complex assembly of bacteria that are sensitive to many perturbations. Subject to subject differences are the primary source of variation. Thus carefully designed perturbation experiments using longitudinal sampling provide the best approach for discovering importWe have developed specific tools for studying the vaginal, intestinal and oral microbiomes under different perturbations (pregnancy, hypo-salivation inducing medications and antibiotics are some examples).
We will show tools we have developed for analysing these data using nonparametric geometric and network-based methods. Challenges we have addressed include information leaks, the integration of phylogenetic information, testing in the presence of longitudinal dependencies and uncertainty quantification.
Our methods enable the detection of changepoints, ecological gradients and their uncertainty quantification, as well as the integration of tree-aware multivariate representations.
This contains joint work with Lan Huong Nguyen, Pratheepa Jeganathan, Claire Donnat, Sergio Bacallado, Ben Callahan, Julia Fukuyama, Kris Sankaran and David Relman's Lab.