Phylogeny-Guided Global-Local Shrinkage for Microbiome Feature Selection

Phylogeny-guided global–local shrinkage (PGL) improves ASV-level detection of exposure-associated relative-abundance differences by borrowing information along a supplied phylogenetic tree, especially when associations align.

• Why it matters: Microbiome feature selection is hard because sequencing counts are sparse and compositional, and evolutionarily related taxa can show similar exposure-linked patterns.
• What they did: PGL uses a Bayesian multinomial logit-normal working model with a zero-sum constraint for log-ratio coefficients, plus global–local shrinkage with an estimated phylogenetic weight for information sharing.
• The result: Simulations show higher detection power and better recovery of nonzero coefficients when aligned with the tree, with borrowing decreasing under tree corruption; in an HIV oral microbiome, it preserved association patterns while selecting three additional ASVs.

AI summary of the abstract · may contain errors · not medical advice · read the paper.