Abstract: Next-generation sequencing, high-throughput metabolomics and other measurement technologies have opened vast new horizons for collecting data on the structure and function of microbial communities. But it remains unclear how to leverage this data for effective intervention in medical and agricultural applications. We do not know which quantities can be reliably predicted, which are hopelessly contingent, and what the predictors are for the former. In this talk, I will draw on conceptual tools from Statistical Physics, which were designed to answer precisely these sorts of questions. In particular, I will argue that the key features of community structure are encoded in a susceptibility matrix, which contain the response of species population sizes to small changes in growth rates. I will show how to estimate this matrix in different scenarios from existing data sets, and then explain how it can be used to cluster species into functionally redundant groups for enhanced predictability of community composition.