A novel normalization and differential abundance test framework for microbiome data.

Area of Science:

• Microbiome analysis
• Bioinformatics
• Computational biology

Background:

• Microbial communities are linked to diseases, making differential abundance analysis crucial for identifying pathogenic or probiotic bacteria.
• Microbiome data's zero-inflation and over-dispersion challenge traditional analysis methods.
• Existing normalization and differential abundance methods struggle with low-to-medium abundance species and lack flexibility in modeling dispersion.

Purpose of the Study:

• To develop a novel framework for accurate differential abundance analysis of sparse, high-dimensional microbiome data.
• To introduce a network-based normalization technique and a robust statistical model for microbiome data.
• To improve the identification of disease-associated microbial species.

Main Methods:

• Developed a novel network-based normalization technique to identify invariant microbial species for size factor construction.
• Implemented a two-stage zero-inflated mixture count regression model (RioNorm2) to handle under-sampling and over-dispersion.
• Separated microbial species into two groups for separate modeling to enhance flexibility.

Main Results:

• The RioNorm2 framework demonstrated consistently powerful and robust performance in comprehensive simulation studies across various settings.
• The method effectively handles different sample sizes, library sizes, and effect sizes.
• Application to a metastatic melanoma dataset yielded significant biological insights.

Conclusions:

• RioNorm2 offers a powerful and robust solution for differential abundance analysis in microbiome research.
• The novel normalization and modeling approach improves accuracy and power, particularly for low-to-medium abundance species.
• The framework facilitates the discovery of clinically meaningful microbial biomarkers for disease association.