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

Area of Science:

• Microbiome Research
• Statistical Bioinformatics
• Computational Biology

Background:

• Identifying microbial features linked to community properties like health phenotypes is crucial but challenging.
• Microbial profiling data often exhibits sparsity and compositionality, complicating statistical analysis.
• Existing models struggle to simultaneously control false discovery rates, incorporate complex terms, and assess both prevalence and abundance associations.

Purpose of the Study:

• To introduce MaAsLin 3 (Microbiome Multivariable Associations with Linear Models), a novel framework for microbiome association studies.
• To enable simultaneous identification of abundance and prevalence relationships in complex microbiome datasets.
• To address data compositionality and expand the types of biological hypotheses testable in microbiome research.

Main Methods:

• Developed MaAsLin 3, a multivariable linear modeling framework for microbiome data.
• Incorporated methods to account for data compositionality using experimental (spike-ins) or computational techniques.
• Applied MaAsLin 3 to synthetic and real datasets, including the Inflammatory Bowel Disease Multi-omics Database.

Main Results:

• MaAsLin 3 demonstrated superior performance over state-of-the-art methods in testing and inferring associations from compositional data.
• Analysis of the Inflammatory Bowel Disease dataset revealed that 77% of microbial associations were with feature prevalence, not abundance.
• The framework successfully corroborated previously reported microbial associations with inflammatory bowel diseases.

Conclusions:

• MaAsLin 3 enhances the accuracy and specificity of microbiome association identification, particularly in complex study designs.
• The framework's ability to assess both prevalence and abundance provides a more comprehensive understanding of microbial roles.
• Findings highlight the importance of considering feature prevalence in microbiome-phenotype associations, especially in disease contexts.