DecoDiPhy: Deconvolving Phylogenetic Distance Mixtures

Mixtures of multiple organisms are sequenced in many applications, including metagenomics, but characterizing their composition faces two challenges: constituents are evolutionarily related, and individual short reads provide limited signal. Here, we introduce phylogenetic distance deconvolution (PDD), a framework that jointly addresses these challenges by using distances from mixtures to reference genomes to place samples on multiple branches of a phylogeny with associated abundances. We develop both exact and heuristic algorithms, implemented in DecoDiPhy. In metagenomic analyses, DecoDiPhy consolidates overdispersed per-read placements into a small set of phylogenetic features, thereby improving interpretability and enhancing downstream phenotype differentiation and differential abundance analysis.