The goal of this project is to develop a computational framework that can learn microbe-microbe and host-microbe interactions from time series of small and large microbial communities. Molecular analyses of human-associated microbial communities have already started to reveal associations between community structure and human health and disease. Also apparent from these initial studies is the dynamic nature of the host-associated microbiomes ¿ even in healthy individuals the microbiome changes even within days, let alone over longer periods of a person's life. These changes, as well as the microbiome changes that underlie the initiation of disease, or the restoration of health after treatment, can only be fully understood by elucidating the complex networks of interactions between the members of the community. These networks cannot currently be observed experimentally as scientists have yet to fully characterize the genomic structure of the members of the community. The current proposal targets methods for inferring the interaction networks, and their parameters, by indirectly examining time-series data about the composition of host-associated communities. Novel analytical methods will be developed that can robustly learn the parameters of dynamic models from time-series data. Furthermore, several approaches will be explored for reducing the complexity of the systems derived from large microbial communities comprising hundreds to thousands of microbes. The methods developed will be evaluated on simulated and real datasets, both to validate the methods, and to evaluate the effects of experimental parameters (such as processing of microbiome data, level of noise, or sparsity of time-series information) on the ability to reconstruct dynamic models of typically encountered human-associated microbial communities.