MixMPLN

MixMPLN infers multiple microbial association networks from a sample-taxa count matrix by modeling counts with a mixture of K Multivariate Poisson Log-Normal distributions to capture factor- or condition-dependent community interactions.

Key Features:

• Mixture Model Framework: Models a sample-taxa count matrix as a mixture of K Multivariate Poisson Log-Normal distributions to represent multiple microbial networks.
• Parameter Estimation: Estimates parameters by maximizing the likelihood using the minorization-maximization principle with gradient ascent and block updates.
• Synthetic Data Generation and Performance Assessment: Generates synthetic sample-taxa count matrices and evaluates performance on absolute count data, compositional data, and normalized data.
• Sparse Network Recovery: Recovers sparse taxa association networks using an l1-penalty to promote sparsity.

Scientific Applications:

• Microbiome network inference: Infers condition- or factor-specific microbial associations from sample-taxa count matrices.
• Environmental microbiology: Characterizes how environmental factors influence microbial network structure.
• Clinical research: Identifies host-associated differences in microbial association networks relevant to clinical studies.
• Ecological studies: Analyzes community-level interaction patterns and shifts across ecological contexts.

Methodology:

Modeling is performed with a mixture of K Multivariate Poisson Log-Normal distributions; parameters are estimated via a minorization-maximization algorithm that combines gradient ascent with block updates to maximize likelihood; an l1-penalty is applied for sparse network recovery; synthetic data generation is used for performance assessment on absolute, compositional, and normalized data.