HARMONIES: A Hybrid Approach for Microbiome Networks Inference via Exploiting Sparsity

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Modeling

Background:

• The human microbiome comprises microorganisms crucial for host health.
• Next-generation sequencing enables high-throughput microbiome profiling.
• Microbiome data present challenges like high dimensionality, uneven sampling, over-dispersion, and zero-inflation, complicating network inference.

Purpose of the Study:

• To develop a robust statistical framework for inferring sparse microbial networks from complex microbiome sequencing data.
• To address the unique statistical challenges posed by microbiome count data.

Main Methods:

• Proposed HARMONIES (Hybrid Approach foR MicrobiOme Network Inferences via Exploiting Sparsity), a novel framework.
• Utilized a zero-inflated negative binomial (ZINB) distribution to model data characteristics, including excess zeros.
• Incorporated a stochastic process prior for sample normalization and Gaussian graphical models for regularization.

Main Results:

• HARMONIES demonstrated superior performance compared to four existing methods in simulation studies.
• The framework successfully identified a novel community of disease-enriched bacteria in colorectal cancer data.
• Inferred sparse and stable microbial networks.

Conclusions:

• HARMONIES provides a powerful and effective statistical approach for microbiome network inference.
• The framework is valuable for analyzing high-dimensional microbiome data and discovering biologically relevant microbial communities.
• HARMONIES is publicly available for use in microbiome research.