Séance Séminaire

Séminaire de Probabilités et Statistique

Distance-based methods in phylogenetics

This talk will start with a broad introduction to the field of phylogenetics and in particular of distance-based methods: several popular methods for phylogenetic inference (but also hierarchical clustering) are based on a matrix of estimated distances between taxa (or any kind of objects); the goal is to construct a tree with edge lengths so that the distances between the leaves in that tree are as close as possible to the estimated distances.
The second part of the talk will introduce a novel framework unifying many of the approaches for distance-based inference. In a recent publication [1], we have shown that all the methods that fit into this general framework have highly desirable statistical properties (the consistency of the tree estimates) and algorithmic properties (efficiency of hill climbing heuristics).
Finally, I will speak about my work on the robustness of distance-based methods: how much can the estimated distances deviate from their 'true' values without compromising the estimation of the 'true' tree topology? It turns out that only one linear optimization principle (among all possible linear principles) guarantees optimal robustness, a very strong result...
[1] Pardi F, Gascuel O. Combinatorics of distance-based tree inference. Proc Natl Acad Sci USA 109: 16443-16448 (2012).