Identification of microbiome markers for ordered groups.

Background: Identifying microbiome markers associated with ordered phenotypes, such as disease stages or severity levels, is crucial for understanding disease progression and advancing precision medicine. Despite this importance, most existing methods for differential abundance analysis are designed for binary group comparisons and do not incorporate ordinal information, limiting their ability to capture trends across ordered categories.

Objective: To develop and evaluate statistical methods that explicitly account for ordinal phenotype structure in microbiome data, addressing challenges such as sparsity and zero inflation, and improving the detection of meaningful microbial associations.

Methods: In this study, we propose and evaluate three novel approaches specifically tailored for microbiome association analysis with ordered groups: the binary optimal test, the linear trend test, and the proportional odds model-based permutation test (POMp). These methods explicitly account for the ordinal structure of phenotypes and address the sparsity and zero-inflation commonly observed in microbiome data through permutation-based inference. We applied the proposed methods to three publicly available gut microbiome datasets, including two related to obesity and one concerning colorectal cancer.

Results: All three proposed methods successfully identified differentially abundant features (DAFs) that exhibited stronger ordinal associations compared to those identified by existing methods. In particular, POMp consistently outperformed other approaches in terms of correlation with phenotype order, demonstrating its potential to identify biologically relevant markers.

Conclusion: The findings of this study highlight the importance of incorporating ordinal information in microbiome studies and provide robust statistical tools for advancing microbial biomarker discovery in complex disease contexts.