Group-wise normalization in differential abundance analysis of microbiome samples.

Area of Science:

• Microbiome research
• Bioinformatics
• Statistical analysis

Background:

• Microbial sequencing data presents challenges in differential abundance analysis (DAA) due to its compositional nature, potentially biasing absolute abundance comparisons.
• Existing normalization methods for DAA struggle to control the false discovery rate (FDR) under high variance or compositional bias.
• Normalization is typically sample-level, but this study re-conceptualizes it as a group-level task to mitigate bias.

Purpose of the Study:

• To introduce a novel group-level normalization framework for microbial sequencing data.
• To develop and evaluate new normalization methods, group-wise relative log expression (G-RLE) and fold-truncated sum scaling (FTSS), within this framework.
• To improve the accuracy and robustness of differential abundance analysis.

Main Methods:

• Developed a group-wise normalization framework for compositional data.
• Introduced two new normalization methods: G-RLE and FTSS.
• Evaluated methods using model-based and synthetic data simulations, comparing performance against existing DAA normalization techniques.

Main Results:

• G-RLE and FTSS demonstrated higher statistical power in identifying differentially abundant taxa compared to existing methods.
• The novel methods successfully maintained the false discovery rate in challenging simulation scenarios where other methods failed.
• Optimal performance was achieved using FTSS normalization in conjunction with the MetagenomeSeq DAA method.

Conclusions:

• The proposed group-level normalization frameworks provide more robust statistical inference for DAA of compositional sequence count data.
• These methods possess a strong mathematical basis, validated performance, and available software, enhancing rigor in microbiome research.
• The novel approaches offer improvements for reproducible and reliable microbiome data analysis.