Variance Component Selection With Applications to Microbiome Taxonomic Data.

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Genetics

Background:

• High-throughput sequencing reveals the human microbiome's role in disease.
• Microbiome data analysis requires identifying disease-associated bacterial taxa.
• Existing methods often ignore crucial phylogenetic relationships.

Purpose of the Study:

• To develop a novel statistical method for identifying bacterial taxa associated with health outcomes.
• To integrate phylogenetic information into microbiome data analysis for improved accuracy.
• To address limitations of existing methods that ignore bacterial community structure.

Main Methods:

• Regression analysis treating bacterial taxa at different levels as multiple random effects.
• Utilizing phylogenetic kernel matrices as variance components in a joint model.
• Applying the least absolute shrinkage and selection operator (LASSO) penalty for taxonomic selection.

Main Results:

• The proposed method significantly improves selection accuracy compared to existing approaches.
• Simulation studies confirm the superiority of the new method over techniques like group-LASSO.
• The method was successfully applied to a longitudinal study of Human Immunodeficiency Virus (HIV) patients.

Conclusions:

• The novel method effectively integrates biological information (phylogeny) into variable selection for microbiome studies.
• This approach offers a more accurate way to identify disease-associated bacteria.
• The developed software in Julia is publicly available for broader research application.