Bayesian Estimation and Inference of Large-Scale Duplications

Background Recent analysis of eukaryotic genomes shows that genome duplication has occurred multiple times in fish, plants and yeast. This thesis presents a Bayesian procedure for estimating and inferring the timing of largescale duplications. The estimation procedure calculates the expected number of duplications over each edge in a discretized species tree, to find sites where many duplications have occurred. A large-scale duplication hypothesis can then be constructed and tested in an extended GSR model, which is a probabilistic model for gene and sequence evolution. The methods are then verified on two yeast data sets that each are thought to contain a genome duplication. Results The estimation procedure performs well on the first data set and clearly indicate the sites of large-scale duplications. On the second data set the estimate is located in the ancestor of the species previously predicted to contain a genome duplication. It is discussed whether this is caused by an inherent assumption of the estimation procedure, but it cannot be resolved without further studies. Several other improvements on the estimation procedure are discussed as well. Conclusions The conclusion is that the method performs well despite relatively small data sets, and the assumption of the estimation procedure can be relaxed by normalizing the expected number of duplications appropriately. In other cases the estimation procedure can be used to generate qualitative hypotheses of large-scale duplication events. Although this thesis suggest a method for testing large-scale duplication hypotheses in the GSR model, it remains to evaluate it experimentally.