(577f) Integration of Metabolic Network and Data-Driven Modeling to Predict Context-dependent Microbial Interactions

Interspecies interactions provide microbes with an important means to promote their fitness so that they can survive in a broader range of environmental conditions as a community, while not individually. As a prime survival strategy, therefore, microbes dynamically modulate their interactions and even change their partners subject to environmental variations. A fundamental understanding of this context-dependent interaction is key for accurately predicting community dynamics and function. Dynamic Flux balance analysis (dFBA) provides mechanistic predictions of how metabolic interactions occur among species through cross-feeding and competition and how the shifts in interactions can be context-dependent across varying environments. Despite this usefulness, mechanistic models alone cannot quantify net interactions among species and their temporal variations, particularly when the system is characterized by mixed interactions. To compensate for this limitation, we propose the integration of data-driven modeling with dFBA. We used Sparse Identification of Nonlinear Dynamics (SINDy) to infer microbial interactions with each other and with the environment. Unlike conventional network inference tools that assume constant interaction coefficients, SINDy allows for the identification of interaction coefficients that are dependent on species abundances and environmental variables. In the case study of a co-culture of auxotrophic Escherichia coli strains, integration of SINDy with dFBA provided not only accurate predictions of the temporal shifts in interactions as the limiting nutrients change in time, but also mechanistic interpretations of how these shifts occur in metabolic reaction levels in individual species. Beyond simple co-cultures, the proposed method can be extended to model more complex communities.