An efficient and scalable top-down method for predicting structures of microbial communities

Modern applications involving multispecies microbial communities rely on the ability to predict structures of such communities in defined environments. The structures depend on pairwise and high-order interactions between species. To unravel these interactions, classical bottom-up approaches examine all possible species subcommunities. Such approaches are not scalable as the number of subcommunities grows exponentially with the number of species, n. Here we present a top-down method wherein the number of subcommunities to be examined grows linearly with n, drastically reducing experimental effort. The method uses steady-state data from leave-one-out subcommunities and mathematical modeling to infer effective pairwise interactions and predict community structures. The accuracy of the method increases with n, making it suitable for large communities. We established the method in silico and validated it against a five-species community from literature and an eight-species community cultured in vitro. Our method offers an efficient and scalable tool for predicting microbial community structures.