Unifying community detection across scales from genomes to landscapes

Biodiversity science encompasses multiple disciplines and biological scales from molecules to landscapes. Nevertheless, biodiversity data are often analyzed separately with discipline-specific methodologies, constraining resulting inferences to a single scale. To overcome this, we present a topic modeling framework to analyze community composition in cross-disciplinary datasets, including those generated from metagenomics, metabolomics, field ecology, and remote sensing. Using topic models, we demonstrate how community detection in different datasets can inform the conservation of interacting plants and herbivores. We show how topic models can identify members of molecular, organismal, and landscape-level communities that relate to wildlife health, from gut microbes to forage quality. We conclude with a future vision for how topic modeling can be used to design cross-scale studies that promote a holistic approach to detect, monitor, and manage biodiversity.