Generalized EPICS (gEPICS): Handling species extinctions in predicting microbial community structures using effective pairwise interactions

Predicting the structures of multispecies microbial communities requires knowledge of the underlying interspecies interactions. EPICS, an efficient and scalable method, employs steady-state data from monocultures and leave-one-out subcommunities together with mathematical modelling to estimate effective pairwise interactions between species and accurately predicts community structures. A limitation, however, is its inability to handle scenarios wherein some species become extinct in leave-one-out subcommunities, restricting its scope. Here, we develop a generalization of EPICS (gEPICS) that overcomes this limitation. We reformulate the mathematical model in EPICS, built using the generalized Lotka-Volterra (GLV) equations, by accounting for missing species. This yields an under-determined system of equations in the unknown effective pairwise interaction coefficients. We estimate the coefficients using the Moore-Penrose pseudo-inverse. With these coefficients, we apply the model to predict community structures. We test gEPICS on a large number of in silico communities and find that it robustly predicts community structures. Its accuracy decreases with the number of missing species but increases with the total number of species involved, rendering it particularly suited to large communities. We apply it to two experimental systems from the literature, an 8-species synthetic soil microbial community and a 12-species synthetic gut microbial community. gEPICS accurately captures the structures of both the communities, demonstrating its applicability to diverse settings. gEPICS would serve as a useful tool to estimate interspecies interactions and engineer multispecies microbial communities.