Bayesian non-parametric model reconstructs microbial community dynamics from short time series (microbiomejournal.biomedcentral.com)

• Model uses Gaussian process priors to combine short time series and infer stability landscapes.
• Estimates probabilistic exit time as a resilience metric for multistable systems.
• Validated on simulated data, lake cyanobacteria, and human gut microbiota tipping elements.

"Researchers from University of Turku and collaborators published a Bayesian non-parametric model that reconstructs microbial community dynamics from multiple short time series. The model uses Gaussian process priors to decompose dynamics into deterministic and stochastic components, predicting stable and tipping regions along a stability landscape. It estimates a probabilistic resilience metric (expected exit time) and can distinguish bistability from bimodality. Validated on simulated data and real cyanobacteria and human gut microbiota time series, the method enables analysis of community stability and tipping points with significantly fewer data points than previously required."