Identification of microbial interaction network: zero-inflated latent Ising model based approach.

Area of Science:

• Microbiome research
• Computational biology
• Statistical modeling

Background:

• Human-associated microbiota, particularly gut microbes, significantly impact health.
• Microbial interaction networks are crucial for understanding microbe relationships but face statistical challenges.
• Zero-inflation and unit-sum constraints in abundance data complicate network estimation.

Purpose of the Study:

• To propose a novel statistical model for estimating microbial interaction networks.
• To address challenges posed by zero-inflation and unit-sum constraints in microbiota data.
• To compare the performance of the new model against existing methods.

Main Methods:

• Development of the zero-inflated latent Ising (ZILI) model.
• A two-step algorithm for ZILI model selection.
• Evaluation using simulated data and application to an infant gut microbiota dataset.
• Comparison with Gaussian graphical model (GGM) and dichotomous Ising model (DIS).

Main Results:

• The ZILI model and its algorithm effectively identify graphical structures in simulated data.
• ZILI tends to produce sparser networks compared to GGM for the infant gut microbiota dataset.
• A hub taxon, Lachnospiraceae, was identified in the shared subnetwork, with recent literature linking it to human diseases.

Conclusions:

• The ZILI model offers an alternative to conventional methods like GGM for analyzing microbiota data with zero-inflation and unit-sum constraints.
• ZILI provides biologically interpretable results for microbial interactions.
• The model is applicable to studying microbial interactions in various body sites.