Learning ecosystem-scale dynamics from microbiome data with MDSINE2.

Area of Science:

• Microbiology
• Computational Biology
• Systems Biology

Background:

• Dynamical systems models are valuable for analyzing microbial ecosystems.
• Learning and interpreting these models from complex microbiome data presents significant challenges.

Purpose of the Study:

• Introduce Microbial Dynamical Systems Inference Engine 2 (MDSINE2), a Bayesian method for learning compact and interpretable ecosystem-scale dynamical systems models.
• Address challenges in analyzing microbiome time-series data and interpreting model outputs.

Main Methods:

• MDSINE2 models microbial dynamics as stochastic processes driven by interaction modules.
• Models incorporate noise characteristics of the data.
• Developed an open-source software package with tools for model interpretation (phylogeny/taxonomy of modules, stability, interaction topology, keystoneness).

Main Results:

• MDSINE2 was benchmarked using microbiome time-series data from murine cohorts subjected to perturbations.
• MDSINE2 outperforms state-of-the-art methods in learning dynamical systems models.
• Identified interaction modules providing insights into gut microbiome ecosystem-scale interactions.

Conclusions:

• MDSINE2 provides a powerful and interpretable approach to modeling microbial ecosystem dynamics.
• The method enhances understanding of gut microbiome interactions and microbial community behavior.