High-dimensional linear state space models for dynamic microbial interaction networks.

Area of Science:

• Microbiology
• Computational Biology
• Systems Biology

Background:

• The human microbiome significantly impacts health, necessitating an understanding of microbial interactions.
• Microbiome composition dynamically changes due to microbial interdependencies and environmental factors.
• High-dimensional microbiome data presents challenges for modeling these complex interactions.

Purpose of the Study:

• To develop a mathematical model for constructing dynamic microbial interaction networks (MINs).
• To address the challenges posed by high-dimensional microbiome data in network construction.
• To identify key dynamic interaction networks within the microbiome.

Main Methods:

• Proposed a high-dimensional linear State Space Model (SSM) with a novel Expectation-Regularization-Maximization (ERM) algorithm.
• Utilized an adaptive LASSO-based variable selection within the ERM algorithm to handle high-dimensional parameter spaces and preserve network sparsity.
• Separately specified system and measurement noise within the SSM framework.

Main Results:

• Simulation studies demonstrated the effectiveness of the proposed ERM algorithm for variable selection.
• Successfully applied the method to identify a dynamic microbial interaction network from a time-course vaginal microbiome dataset.
• The model effectively captures the sparsity inherent in microbial interaction networks.

Conclusions:

• The developed high-dimensional SSM with ERM algorithm provides a robust method for constructing dynamic microbial interaction networks.
• This approach enables the identification of key bacterial interactions within complex microbiome ecosystems.
• The methodology is adaptable for future extensions, including the integration of environmental factors into microbial interaction analyses.