Umibato

Umibato estimates time-varying microbial interaction networks from quantitative time-series data to infer dynamic ecological relationships among microbes.

Key Features:

• Generalized Lotka-Volterra (gLVE): Models directed microbial interactions using a generalized Lotka-Volterra framework with time-varying interaction parameters.
• Bayesian Estimation: Infers interaction parameters and quantifies uncertainty via Bayesian estimation.
• Gaussian Process Regression (GPR): Estimates species-specific growth rates from time-series abundance data using Gaussian Process Regression to capture non-linear temporal trends.
• Continuous-Time Regression Hidden Markov Model (CTRHMM): Represents time-varying interactions as hidden interaction states in a continuous-time regression hidden Markov model.
• Variational Inference for CTRHMM: Fits the CTRHMM using variational inference with the core variational inference component implemented in C++ for computational efficiency.
• Benchmarking and Case Study: Demonstrated superior performance on synthetic datasets and was applied to mouse gut microbiota data to investigate effects of dietary changes.

Scientific Applications:

• Microbial ecology: Characterizes temporal changes in directed interactions within microbial communities.
• Host-microbe interactions: Analyzes temporal dynamics of host-associated microbiota, exemplified by mouse gut microbiota studies.
• Environmental and industrial systems: Applies to human gut, soil ecosystems, and industrial fermentation to study time-varying community interactions.
• Ecological stability and perturbation response: Investigates community stability, resilience, and responses to external perturbations such as dietary shifts.

Methodology:

Umibato combines a generalized Lotka-Volterra framework with Bayesian estimation, uses Gaussian Process Regression to estimate growth rates, models time-varying interactions with a continuous-time regression hidden Markov model (CTRHMM), and fits the CTRHMM via variational inference with the core variational inference implemented in C++.