Physics-constrained neural ordinary differential equation models to discover and predict microbial community dynamics

Microbial communities play essential roles in shaping ecosystem functions and predictive modeling frameworks are crucial for understanding, controlling, and harnessing their properties. Competition and cross-feeding of metabolites drives microbiome dynamics and functions. Existing mechanistic models that capture metabolite-mediated interactions in microbial communities have limited flexibility due to rigid assumptions. While machine learning models provide flexibility, they require large datasets, are challenging to interpret, and can over-fit to experimental noise. To overcome these limitations, we develop a physics-constrained machine learning model, which we call the Neural Species Mediator (NSM), that combines a mechanistic model of metabolite dynamics with a machine learning component. The NSM is more accurate than mechanistic or machine learning components on experimental datasets and provides insights into direct biological interactions. In summary, embedding a neural network into a mechanistic model of microbial community dynamics improves prediction performance and interpretability compared to its constituent mechanistic or machine learning components. Significance statementMicrobial communities drive essential biological processes in every ecosystem. Predicting microbial community dynamics and functions and uncovering their interaction networks are challenging due to their complexity and emergent interactions. We develop a framework that combines biological knowledge with machine learning to accurately predict community and metabolite dynamics. This approach outperforms existing ecological and machine learning models by capturing environmentally mediated interactions while providing interpretable insights into microbe-metabolite interactions. Our framework enables the rational design and control of microbial communities for applications in medicine, agriculture and biotechnology