Understanding the intricate relationships within microbial communities is crucial for ecological research. Microbial communities are shaped by the interplay between species coexistence and environmental factors, and while statistical modelling has made significant strides in dissecting these relationships, the complex nature of ecological data poses challenges. This work explores novel approaches to untangle the web of interactions shaping microbial communities using machine learning approaches and introduces the MrIML R package (multi-response interpretable machine learning). By leveraging cutting-edge machine learning techniques, MrIML approximates a graphical network model that can shed light on the complex dynamics governing species distributions and community structure.
Current approaches, such as Joint Species Distribution Models, offer insights into species associations, yet their indirect nature poses limitations in accurately discerning true interactions. The study explores how MrIML-based graphical networks overcome some of these challenges and can provide a more nuanced understanding of ecological relationships that can be used to generate new hypotheses. We validate our models using simulated and real-world data to showcase the potential of interpretable machine learning tools in assisting in unravelling ecological interactions.