MaAsLin 3: refining and extending generalized multivariable linear models for meta-omic association discovery

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Genetics

Background:

• Microbial community analysis links microbial features to phenotypes.
• Sparsity and compositionality hinder accurate association identification.
• Existing methods struggle with complex microbiome data designs.

Purpose of the Study:

• Introduce MaAsLin 3 (microbiome multivariable associations with linear models).
• Enable simultaneous identification of abundance and prevalence relationships.
• Address compositionality and complex study designs in microbiome research.

Main Methods:

• MaAsLin 3 employs multivariable linear models.
• Accounts for compositionality via experimental or computational methods.
• Expands testable hypotheses and covariate types.

Main Results:

• MaAsLin 3 outperformed state-of-the-art differential abundance methods on synthetic and real datasets.
• Identified 77% of associations in the Inflammatory Bowel Disease Multi-omics Database based on feature prevalence.
• Demonstrated superior accuracy and specificity in complex microbiome datasets.

Conclusions:

• MaAsLin 3 enhances the accuracy and specificity of microbiome association studies.
• It is particularly effective for complex datasets with sparsity and compositionality.
• Facilitates more precise identification of microbial feature-phenotype relationships.