Abstract
Despite the importance of microbial dysbiosis in human disease, the phenomenon remains poorly understood. We provide the first comprehensive and predictive model of dysbiosis at ecosystem-scale, leveraging our new machine learning method for efficiently inferring compact and interpretable dynamical systems models. Coupling this approach with the most densely temporally sampled interventional study of the microbiome to date, using microbiota from healthy and dysbiotic human donors that we transplanted into mice subjected to antibiotic and dietary interventions, we demonstrate superior predictive performance of our method over state-of-the-art techniques. Moreover, we demonstrate that our approach uncovers intrinsic dynamical properties of dysbiosis driven by destabilizing competitive cycles, in contrast to stabilizing interaction chains in the healthy microbiome, which have implications for restoration of the microbiome to treat disease.
Competing Interest Statement
The authors have declared no competing interest.
