Zero-Inflated Beta Regression for Differential Abundance Analysis with Metagenomics Data.

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Genetics

Background:

• Next-generation sequencing (NGS) technologies generate vast amounts of metagenomics data.
• Analyzing human microbial data aims to detect abundance differences across clinical conditions.
• Metagenomics data are compositional, sparse, high-dimensional, and often exhibit skewed distributions.

Purpose of the Study:

• To propose a novel statistical method, ZIBSeq, for identifying differentially abundant features in metagenomics data.
• To address the challenges of compositional, sparse, and high-dimensional nature of metagenomics data.
• To provide an efficient tool for comparing microbial communities across multiple clinical conditions.

Main Methods:

• Developed a zero-inflated beta regression approach (ZIBSeq).
• The method accounts for the sparse nature of metagenomics data.
• Efficiently handles compositional data structures.

Main Results:

• ZIBSeq demonstrated superior performance with high AUC values in simulation studies compared to existing methods.
• The approach effectively identified biologically relevant taxa in a human metagenomics dataset.
• The method successfully handles sparse and compositional data.

Conclusions:

• ZIBSeq is an effective tool for identifying differentially abundant features in human metagenomics studies.
• The proposed method offers improved performance for analyzing complex microbial community data.
• The R software package is available for researchers.