Navigation and control of outcomes in a generalized Lotka-Volterra model of the microbiome

Abstract: The generalized Lotka-Volterra (gLV) equations model the microbiome as a collection of interacting ecological species. Here we use a particular experimentally-derived gLV model of C. difficile infection (CDI) as a case study to generate methods that are applicable to generic gLV models. We examine how to transition gLV systems between multiple steady states through the application of direct control protocols, which alter the state of the system via the instantaneous addition or subtraction of microbial species. Then, the geometry of the basins of attraction of point attractors is compressed into an attractor network, which decomposes a multistable high-dimensional landscape into web of bistable subsystems. This attractor network is used to identify efficient (total intervention volume minimizing) protocols that drive the system from one basin to another. In some cases, the most efficient control protocol is circuitous and will take the system through intermediate steady states with sequential interventions. Clinically, the efficient control of the microbiome has pertinent applications for bacteriotherapies, which seek to remedy microbiome-affiliated diseases by directly altering the composition of the gut microbiome.