Dr. Bernard Moret
EPFL (Swiss Federal Institute of Technology), Lausanne, Switzerland

Phylogenetic Analyses: Past, Present, and Future Bernard M. Moret School of Computer and Communication Sciences EPFL (Swiss Federal Institute of Technology) Lausanne, Switzerland http://lcbb.epfl.ch/ Phylogenies are simplified histories of the evolution of a group of taxa (organisms, genes, biological networks, computer malware, artistic styles, etc.) These phylogenies are inferred from modern-day specimens, in a process that starts by collecting comparable data about the taxa (such as the sequences of a few genes), then devising an appropriate model of evolution for the data, and finally running an inference procedure (machine-learning) to obtain a tree and some parameter values about that tree. Each year, thousands of citations are made to existing phylogenetic inference packages, mostly in the life sciences, but also in computer science, linguistics, forensics, and art history. As enounced by Th. Dobzhansky in the title of one of his papers, "biology makes no sense except in the light of evolution" and phylogenetic analyses are our spotlights. Yet in this talk I will argue that phylogenetic analyses are underused and in need of generalization. For the last 80 years, phylogenies have used sequence data as the basis for inference; at first these sequences coded for morphological characteristics or simple genomic characteristics such as chromosomal banding; for the last 40 years, they have been RNA or DNA sequences. Phylogenetic analyses of languages, artistic styles, criminal activities, biological networks, or entire genomes have had to use tools developed to analyze relatively short sequences with very simple evolutionary models: the complexity of evolutionary models for other data, along with the relative paucity of studies based on such data, prevented the development of analysis techniques better adapted to the data. We thus need to enlarge and generalize existing techniques to improve the quality of phylogenetic analyses of data other than genome sequence data and to enable phylogenetic analyses for entirely new types of data. In particular we need new models, sophisticated preprocessing, and reasonable optimization criteria. Most importantly, we need an enlightened view of phylogenetic analyses in science. We are all familiar with comparative methods, but a comparison between two taxa or a collection of pairwise comparisons among a collection of taxa is just a \emph{degenerate} phylogenetic analysis, one that makes no (or minimal) use of evolution and models. In any area where the objects of study are subject to some form of evolution, phylogenetic analyses will yield much better results than simple comparative studies.