Scale reliant mixed effects models enhance microbiome data analysis

Area of Science:

• Microbiome research
• Statistical modeling
• Bioinformatics

Background:

• Linear models and linear mixed-effects models (MEMs) are common in microbiome research.
• Existing MEMs struggle with compositional microbiome data, leading to high error rates.
• Normalization methods in current MEMs rely on unrealistic assumptions about unmeasured biological scale.

Purpose of the Study:

• Introduce scale-reliant mixed-effects models (SR-MEM) for robust microbiome data analysis.
• Address limitations of existing methods in handling compositional data and unmeasured biological scale.
• Provide a principled framework for mixed-effects modeling in complex microbiome studies.

Main Methods:

• Developed SR-MEM by explicitly modeling uncertainty in the unmeasured biological scale.
• Treated biological scale as a latent variable, avoiding normalization-based assumptions.
• Incorporated external scale measurements or information from independent studies.

Main Results:

• SR-MEM demonstrated consistent control of the false discovery rate.
• Achieved comparable or higher statistical power than standard normalization or bias-correction methods.
• Reanalyses of published datasets showed improved reproducibility and consistency with biological effects.

Conclusions:

• SR-MEM offers a robust and practical framework for mixed-effects modeling of microbiome sequence count data.
• By propagating scale uncertainty, SR-MEM enhances error control and reproducibility in longitudinal and hierarchical studies.
• An implementation is available in the ALDEx3 R package.