Compositional zero-inflated network estimation for microbiome data

Area of Science:

• Microbiome research
• Computational biology
• Statistical ecology

Background:

• Microbiome network estimation is crucial for understanding ecological relationships.
• High-throughput microbiome data present statistical challenges: compositional nature and zero-inflation.
• Existing methods struggle with these data characteristics.

Purpose of the Study:

• To introduce a novel method, COmpositional Zero-Inflated Network Estimation (COZINE), for microbial network inference.
• To address the compositional and zero-inflated properties of microbiome data.
• To provide a computationally scalable solution for network estimation.

Main Methods:

• Utilized the multivariate Hurdle model within the COZINE framework.
• Inferred sparse conditional dependencies considering continuous, binary, and mixed data representations.
• Developed a method that maintains computational scalability.

Main Results:

• COZINE demonstrated superior performance in capturing diverse microbial relationships compared to existing methods.
• Simulations confirmed the method's effectiveness in various scenarios.
• Applied COZINE to analyze the oral microbiome network in leukemic patients.

Conclusions:

• COZINE successfully addresses key challenges in microbiome network estimation.
• The method effectively discovers diverse dependence relationships within microbial communities.
• COZINE is publicly available for broader scientific application.