Koinetic: a graph-based distance as a new ecological metric for microbial community differential analysis
Briefings in Bioinformatics, vol. 27, issue 6, bbag632 ().
Abstract
Phylogeny-aware beta diversity measures are essential in microbiome research, yet widely used metrics such as UniFrac and its variants rely on independent branch length contributions and do not satisfy the Euclidean property, leading to distortions in ordination and kernel-based tests. These limitations underscore the need for a distance that is both phylogenetically coherent and geometrically valid. We introduce the Koinos Kinetic Distance (Koinetic), a phylogeny-aware dissimilarity that views microbial communities as interconnected assemblies rather than isolated taxa. “Koinos” emphasizes collective evaluation, while “kinetic” reflects how differences propagate dynamically across ecological networks and phylogenetic lineages. Koinetic leverages a graph Laplacian–based kernel to weight community differences by shared phylogeny, capturing ecological connectivity and clade relatedness. This design ensures Euclidean validity and enables seamless integration with kernel-based methods as well as geometry-preserving ordination. We demonstrate the effectiveness of Koinetic through a combination of controlled simulation studies and applications to eight microbiome datasets spanning pancreatic cancer, IBD, and metabolic disease–associated oral microbiomes. Compared to classical and phylogenetic metrics, including weighted, unweighted, and generalized UniFrac, Koinetic achieves superior preservation of geometry under dimensionality reduction and improved clustering. It bridges phylogeny-aware metrics with kernel-compatible representations, thereby providing a general foundation for phylogeny-aware analysis.