Tropical Geometry of Phylogenetic Tree Space: A statistical Perspective
TL;DR
By representing phylogenetic trees in a tropical geometric space called "palm tree space," researchers achieved faster geodesic computations and better performance in statistical tasks like PCA compared to the standard BHV tree space.
Problem / question
Sets of phylogenetic trees form a complex, non-Euclidean space (BHV space) where standard statistical methods fail, suffering from computational bottlenecks and "sticky" means that prevent exact parametric statistical inference.
Methods
The authors modeled trees in "palm tree space" using a tropical metric, proving it is a Polish space suitable for probability measures. They developed tropical Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). They tested tropical PCA on 1,089 influenza A H3N2 hemagglutinin sequences (from 1993-2017, split into 21 datasets of approximately 20,000 trees each) and tested tropical LDA on simulated coalescent model gene trees with 10 leaves.
Key findings
Palm tree space reduced the time complexity of computing geodesics to O(N^2 log N) from the O(N^4) required in BHV space. In the influenza dataset, tropical PCA explained a higher and more consistent proportion of standard deviation (46% to 96%) than BHV PCA (3.7% to 97%) and captured richer tree topologies, projecting 6 distinct topologies versus 2 in the 2008 dataset. Tropical LDA successfully classified simulated gene trees, reaching 100% accuracy when the species depth to population size ratio was 5 or 10.
Why it matters
This tropical geometric framework provides a mathematically rigorous and computationally efficient foundation for applying standard machine learning and parametric statistical tools directly to complex evolutionary tree datasets.
Limitations
Unlike BHV space, geodesics (shortest paths between points) are generally not unique in palm tree space, which can complicate certain computational analyses, and the BHV distance cannot be generally bounded by the tropical distance.
Takeaway
Tropical geometry offers a highly efficient and statistically viable alternative to traditional non-Euclidean spaces for analyzing large datasets of phylogenetic trees.