Umibato: estimation of time-varying microbial interaction using continuous-time regression hidden Markov model

Area of Science:

• Microbiology
• Computational Biology
• Systems Biology

Background:

• Microbial interaction networks are crucial for understanding microbial communities.
• Existing methods, like generalized Lotka-Volterra equations (gLVE), can estimate directed networks but often overlook temporal dynamics.
• There is a need for methods that can capture the dynamic nature of microbial interactions over time.

Purpose of the Study:

• To develop an unsupervised learning-based method for inferring time-varying microbial interactions.
• To introduce Umibato, a novel algorithm that estimates dynamic microbial interaction networks.
• To provide a tool for analyzing the temporal complexity of microbial communities.

Main Methods:

• Umibato utilizes Bayesian estimation, combining Gaussian process regression (GPR) for growth rate estimation and a continuous-time regression hidden Markov model (CTRHMM) for interaction network inference.
• CTRHMM employs hidden variables to define interaction states, enabling the estimation of time-varying interactions.
• The method is unsupervised, reducing the need for prior assumptions about the microbial community structure.

Main Results:

• Umibato demonstrated superior performance compared to existing methods on synthetic datasets.
• The method successfully inferred time-varying microbial interactions in a mouse gut microbiota dataset.
• Analysis of the mouse gut microbiota provided novel insights into diet-microbiota relationships.

Conclusions:

• Umibato offers a robust approach for inferring dynamic microbial interaction networks.
• The method enhances our understanding of microbial community behavior and ecological dynamics.
• Umibato provides valuable tools for microbiome research, particularly in studying the impact of environmental factors like diet.