Microbes form complex and ever-changing communities in which species compete for resources, exchange metabolic products, and respond to shifting environmental conditions like nutrient and oxygen availability. Understanding the ecological relationships that underlie these communities has implications for improving fecal microbiota transplantation and designing targeted bacteriotherapies. While existing computational methods have inferred putative ecological relationships from longitudinal abundance data alone, they typically ignore available biological knowledge, making it difficult to assess the biological plausibility of predicted

Interactions. We present a Bayesian dynamical systems framework with informative priors on the structure of the ecological network—in essence, our model incorporates a prior probability on whether a given pair of species interact, based on known biological data. For example, by assigning higher prior probabilities to interactions between species that are spatially co-localized or metabolically complementary, we produce more biologically plausible network structures while preserving the flexibility to discover novel interactions. This framework is

data-type agnostic, capable of integrating diverse biological information such as spatial proximity measurements, metabolic pathway complementarity, or phylogenetic relationships.

Our model implementation employs scalable variational inference with automatic differentiation, enabling efficient inference of large microbial communities. We demonstrate our approach using both semi-synthetic data and a mouse gut microbiome dataset that combines metagenomic time series with spatial measurements from SAMPL-seq. Our results show that incorporating biological priors improves network reconstruction accuracy in the semi-synthetic case and produces biologically plausible models compared to methods that treat all potential interactions as equally likely. This approach provides a principled framework for integrating multiple data types into dynamic models of microbial communities, advancing our ability to predict and manipulate complex microbiome dynamics.