Resampling-based inferences for compositional regression with application to beef cattle microbiomes.

Area of Science:

• Microbiome research
• Statistical modeling
• Bioinformatics

Background:

• Gut microbiomes influence human and animal health.
• Analyzing compositional microbiome data with many taxa and few samples is challenging.
• Existing sparse log-contrast regression lacks formal inference procedures.

Purpose of the Study:

• To develop a novel estimation and inference procedure for linear regression with low-sample-sized compositional predictors.
• To establish a formal statistical inference for individual regression coefficients in microbiome studies.
• To identify key bacterial taxa associated with phenotype responses.

Main Methods:

• A two-step approach using compositional log-contrast regression.
• Step 1: Sparse penalty to screen relevant predictors.
• Step 2: Non-sparse model with nonparametric testing (permutation, bootstrap) for regression coefficients.

Main Results:

• The proposed method outperforms traditional approaches in simulation studies.
• It is effective even with extremely low sample sizes.
• Successfully identified key bacterial taxa related to cattle quality measures in an application.

Conclusions:

• The new procedure provides a robust framework for microbiome data analysis.
• It enables reliable identification of microbial taxa associated with specific health outcomes.
• This method advances the statistical toolkit for microbiome-driven health research.