Phylogenetic shadowing through model selection

Motivation: In comparative genomics, phylogenetic shadowing is the procedure of highlighting alignment segments exhibiting unusual rates of sequence evolution, presumably due to purifying selection or adaptation. Phylogenetic shadowing has been extremely useful in delineating putative functional elements in the human genome, by juxtaposition with other primate and mammalian genomes. Results: We describe a probabilistic framework for phylogenetic shadowing in which likelihood scores are combined with a complexity penalty to restrict the false positive identification of unusual rate variation. The framework gives more flexibility in controlling sensitivity than Hidden Markov Models do. We also discuss theoretical and statistical considerations that ensure reasonable practices in choosing model parameters. Our study concentrates on the case of phylogenetic shadowing for a three-wise alignment. The techniques are illustrated by the example of a human-chimpanzee-baboon alignment in which we identify previously unannotated regions displaying unusual conservation or human-specific divergence. Availability: The Java implementations of the algorithms are publicly available from the corresponding author’s site http://www. iro.umontreal.ca/ ̃csuros/threemonkeys/. The same site also provides the segmentation of the human-chimpanzee-baboon alignment as an annotation track for the human genome in GFF