Simplifying the description, not the system: Advanced ecological theory for real-world microbiomes

Washington University, Saint Louis MO

Investigators

Abstract

Project summary Microbial ecosystems play a major role in human health. An ability to quantitatively predict their function and dynamics, and design communities with desired properties, would be transformative for microbiome-based diagnostics and treatments of diseases. However, natural systems are highly complex, harboring hundreds of species. Recently, an important theoretical development was the realization that high-diversity ecosystems are subject to different emergent laws, arising from diversity. These emergent behaviors limit our ability to obtain generalizable insight from simplified examples in the lab. Understanding the behavior and properties of complex communities requires ecological theory and computational tools specific to the high-diversity regime, considering microbiomes in their natural complex context. The solution is to simplify the description, not the system. Recent advancements in my group have demonstrated that increasing community diversity can enhance the predictive power of simple ecological models, a phenomenon we refer to as “emergent simplicity.” This breaks from traditional perspectives, opening the door to new methodologies that work because of diversity, not despite it. Over the next five years, my group will: (1) Build computational tools to identify functionally significant taxa groupings, applicable to diverse ecosystems like marine, soil, and gut microbiomes; (2) Create statistical methods to design microbial communities with desired functions, with initial validations focused on synthetic communities suppressing pathogens such as K. pneumoniae; (3) Test the power and limitations of coarse-grained modeling in a complex natural community (soil); and (4) Systematize the insights from these empirical examples into a theory of ecosystem coarse-graining, classifying ecosystem properties by their “coarse-grainability,” and guiding data collection by identifying the most informative perturbations for predictive coarse models to be learned. The research will yield (A) statistical and computational tools of broad applicability to fundamental biology, such as de novo discovery of relevant dynamical variables; (B) new statistical methodologies to discover bacterial interactions from community-level observations; and (C) a systematic methodology for developing interpretable coarse-grained models trainable on minimal data, enabling prediction in clinical contexts where obtaining millions of training examples is infeasible.