Abstract
Recent advances in DNA sequencing have led to a boom in microbiome analyses to determinecommunity composition. Fitting such community composition data to mathematical
models allows for the estimation of interspecies interactions within a microbial community.
In this thesis, we explore the extent to which noise inherent to time-series microbiome data
interferes with the inference of interspecies interactions. We first create a small synthetic
test community with structure mimicking real microbial communities based on the generalized
Lotka-Volterra (gLV) model, incorporating di↵ering levels of two types of noise, process
and measurement noise. We then establish a method of parameter estimation for both the
gLV and multivariate autoregressive (MAR) models, and apply the method to our synthetic
dataset with varying levels of noise. We find that interspecies interactions can be well estimated
even with moderate levels of process noise, but even modest amounts of measurement
noise lead to poor estimates of interactions.