MDiNE

MDiNE estimates and compares microbial co-occurrence networks from microbiome count data by modeling differences in precision matrices between groups defined by a binary covariate.

Key Features:

• Microbial Co-occurrence Network Estimation: Constructs co-occurrence networks of Operational Taxonomic Units (OTUs) using precision matrices derived from microbiome count data.
• Differential Network Analysis: Estimates differences between microbial networks corresponding to groups defined by a binary covariate through comparison of precision matrices.
• Multinomial Count Modeling: Models microbial taxon counts using a multinomial distribution influenced by latent Gaussian random variables.
• Sparse Precision Matrix Inference: Uses sparse precision matrices over latent variables to infer microbial interaction networks.
• Bayesian Inference via Hamiltonian Monte Carlo: Estimates model parameters and evaluates model fit using Hamiltonian Monte Carlo sampling.

Scientific Applications:

• Microbiome Network Analysis: Identifies interaction patterns among microbial taxa within microbial communities.
• Differential Microbial Network Studies: Detects changes in microbial co-occurrence networks between groups such as disease and control cohorts.
• Microbial Ecology and Disease Research: Investigates microbiome interaction structures associated with health conditions such as Crohn's disease.

Methodology:

MDiNE models microbial count data using a multinomial distribution influenced by latent Gaussian variables, estimates sparse precision matrices representing OTU co-occurrence networks, compares precision matrices between groups defined by a binary covariate, and performs parameter inference using Hamiltonian Monte Carlo.