Abstract:
The composition of microbial communities, in terms of relative abundances of constituent taxa, is constantly subjected to external perturbations, largely from changes in environmental parameters affecting dynamical trajectories. Understanding the sensitivity of community composition to such perturbations is crucial, given their role in maintenance of ecosystem function at scales from human health to global biogeochemical cycles. In this thesis, I unify statistical techniques from empirical dynamical modeling for nonlinear dynamical reconstruction with causal inference to evaluate how abundance and interactions affect the sensitivity of microbial communities to environmental parameter perturbations. Using simulated communities alongside longitudinal abundance data of taxa in three human microbiota, I develop a novel approach based on sampling small communities to detect and track the effect of external perturbations. Then, using causal inference to identify the effects of abundance and interactions on sensitivity, I identify 4 sub-community profiles based on causal effect signs. Simulated communities with different interaction strengths display a continuous transition in frequencies of different profiles, with real communities displaying frequencies similar to simulated weak interaction regimes. Detecting perturbations within each profile, I find that rarer profiles in competitive communities have consistently more sensitive responses, while in weak interaction regimes, this identity is variable. Real microbiota appear to show inconsistent sensitivity patterns as well. Despite the individualized nature of microbial compositions and the variety of possible responses, our study shows the possibility to synthesize the nonlinear nature of microbiota dynamics. This approach can be relevant for personalized approaches in monitoring and designing optimal therapeutic treatments.
