HARMONIES

HARMONIES infers sparse microbial interaction networks from zero-inflated, over-dispersed microbiome sequencing count data to characterize microbial associations.

Key Features:

• Statistical Modeling with ZINB Distribution: Models microbiome sequencing count data using a zero-inflated negative binomial (ZINB) distribution to account for skewness and excess zeros.
• Sample-wise Normalization: Implements a stochastic process prior for sample-wise normalization to address uneven sampling depth and over-dispersion.
• Sparse Network Inference: Infers sparse microbial networks via Gaussian graphical model regularization to reduce noise and enhance interpretability.
• Comprehensive Simulation Studies: Includes simulation benchmarks reporting improved accuracy and reliability in network inference relative to four other methods.

Scientific Applications:

• Colorectal cancer microbiome analysis: Applied to published colorectal cancer microbiome sequencing data to identify a microbial community enriched with disease-associated bacteria.
• Host-microbiome interaction discovery: Supports discovery of microbial associations relevant to host health and disease.

Methodology:

Uses a zero-inflated negative binomial (ZINB) statistical model with a stochastic process prior for sample-wise normalization and Gaussian graphical model-based regularization to exploit sparsity in network inference.