Seminars

We explain why the least squares formalism for distance- based phylogenetics provides a foundation for understanding many of the distance methods that have been proposed during the past 40 years. Our main result is a characterization of least squares methods that satisfy a phylogenetically desirable property. We use this to provide a statistical interpretation for the popular neighbor-joining algorithm, and argue that certain least squares methods (for example the Fitch-Margoliash criterion) are problematic. A key concept is the phylogenetic orange, introduced by Junhyong Kim. We explain its relevance, and also its broader significance in phylogenetics. Finally, we discuss obvious next steps that follow from our results, leading to practical improvements in distance-based phylogenetic algorithms and connections to interesting mathematical objects.

This is joint work with Radu Mihaescu.