Joint Additive Gaussian Processes for Microbial Species Distribution Modeling

Microorganisms are fundamental to the functioning of every ecosystem on Earth.Yet, the majority of microbial species remain uncultured and uncharacterized. Ex-panding our understanding or generating hypotheses about how different factorsaffect species can help accelerate the discovery of new insights. Species Distribu-tion Modeling (SDM) has traditionally been the primary approach for gaining suchinsights. However, previous models have often struggled with capturing non-linearresponses and have largely focused on environmental predictors. In this paper, weinstead explore an additive Gaussian Process (GP) framework to jointly predictspecies responses to environmental features and spatial effects, while also leveragingmodel interpretability to enable domain analysis. The model is compared to existingbaseline models across several real-world datasets, showing promising results. Wedemonstrate how the interpretable nature of the model can provide insight into therelationship between environmental features and species community compositions aswell as support uncertainty estimation for species response curves.