Compositional data analysis enables statistical rigor in comparative glycomics

Comparative glycomics data are compositional data, where measured glycans are parts of a whole, indicated by relative abundances. Applying traditional statistical analyses to these data often results in misleading conclusions, such as spurious "decreases" of glycans when other structures increase in abundance, or high false-positive rates for differential abundance. Our work introduces a compositional data analysis framework, tailored to comparative glycomics, to account for these data dependencies. We employ center log-ratio and additive log-ratio transformations, augmented with a scale uncertainty/information model, to introduce a statistically robust and sensitive data analysis pipeline. Applied to comparative glycomics datasets, including known glycan concentrations in defined mixtures, this approach controls false-positive rates and results in reproducible biological findings. Additionally, we present specialized analysis modalities: alpha- and beta-diversity analyze glycan distributions within and between samples, while cross-class glycan correlations shed light on previously undetected interdependencies. These approaches reveal insights into glycome variations that are critical to understanding roles of glycans in health and disease.