"Exome-wide phylogenetic analyses of patterns of selection using a multi-gene hierarchical Bayesian modeling approach."
Signatures of adaptation, and more generally patterns of selection, can be detected from multiple sequence alignment of protein-coding genes using phylogenetic codon models. The classical approach is to contrast the rates at which synonymous and non-synonymous substitutions have accumulated over evolutionary times. A ratio of these two rates (dN/dS) greater than 1 is interpreted as a signature of adaptation. A more recent approach relies on so-called mutation-selection codon models. These mechanistic models implement a more detailed null model of the nearly-neutral regime against which adaptation can then be detected, potentially with a greater sensitivity. With the increasing number of complete and well-annotated genomes, it now becomes possible to apply these models, both classical and mechanistic, across the entire exome (possibly restricted to the set of single-copy orthologous genes). Such large-scale applications, however, raise several methodological issues, while also offering interesting opportunities. First, they are computationally demanding. Second, multi-gene analyses raise the question of the correct control/calibration of the frequentist properties of the inference (whether in a maximum likelihood or in a Bayesian framework). Third, it would be interesting to pool information across single-gene alignments about the species-level parameters of the models, which are inherently shared across genes. To address these points, a hierarchical approach will be presented (implemented in a new software program, called BayesCode), working directly over a large set of multiple sequence alignments across the exomes of a set of species of interest. Simulation experiments will be used to illustrate the frequentist calibration properties of the approach, contrasting it to single-gene approaches such as implemented in CodeML. Finally, an application to ~10 000 orthologous genes across ~ 100 mammalian species, under the classical and the mutation-selection models, will be presented.
|
Nicolas Lartillot. CNRS Researcher, at the LIRMM, Montpellier (2004-2008) at the University of Montreal (2008-2015) and at the LBBE in Lyon (2015-2023). Main aspects of the research revolve around methodological developments in Bayesian inference and Monte Carlo algorithmics for phylogenetic models, with a particular focus on non-parametric modeling of the modulation of selection and amino-acid usage across coding sites (PhyloBayes); the correlated evolution of genetic sequences and phenotypic characters; and the mechanistic modeling of coding sequence evolution (mutation-selection codon models).
|
|
|
|
