Improved Bayesian methods for detecting recombination and rate heterogeneity in DNA sequence alignments

Abstract

DNA sequence alignments are usually not homogeneous. Mosaic structures may result as a consequence of recombination or rate heterogeneity. Interspecific recombination, in which DNA subsequences are transferred between different (typically viral or bacterial) strains may result in a change of the topology of the underlying phylogenetic tree. Rate heterogeneity corresponds to a change of the nucleotide substitution rate. Various methods for simultaneously detecting recombination and rate heterogeneity in DNA sequence alignments have recently been proposed, based on complex probabilistic models that combine phylogenetic trees with factorial hidden Markov models or multiple changepoint processes. The objective of my thesis is to identify potential shortcomings of these models and explore ways of how to improve them. One shortcoming that I have identified is related to an approximation made in various recently proposed Bayesian models. The Bayesian paradigm requires the solution of an integral over the space of parameters. To render this integration analytically tractable, these models assume that the vectors of branch lengths of the phylogenetic tree are independent among sites. While this approximation reduces the computational complexity considerably, I show that it leads to the systematic prediction of spurious topology changes in the Felsenstein zone, that is, the area in the branch lengths configuration space where maximum parsimony consistently infers the wrong topology due to long-branch attraction. I demonstrate these failures by using two Bayesian hypothesis tests, based on an inter- and an intra-model approach to estimating the marginal likelihood. I then propose a revised model that addresses these shortcomings, and demonstrate its improved performance on a set of synthetic DNA sequence alignments systematically generated around the Felsenstein zone. The core model explored in my thesis is a phylogenetic factorial hidden Markov model (FHMM) for detecting two types of mosaic structures in DNA sequence alignments, related to recombination and rate heterogeneity. The focus of my work is on improving the modelling of the latter aspect. Earlier research efforts by other authors have modelled different degrees of rate heterogeneity with separate hidden states of the FHMM. Their work fails to appreciate the intrinsic difference between two types of rate heterogeneity: long-range regional effects, which are potentially related to differences in the selective pressure, and the short-term periodic patterns within the codons, which merely capture the signature of the genetic code. I have improved these earlier phylogenetic FHMMs in two respects. Firstly, by sampling the rate vector from the posterior distribution with RJMCMC I have made the modelling of regional rate heterogeneity more flexible, and I infer the number of different degrees of divergence directly from the DNA sequence alignment, thereby dispensing with the need to arbitrarily select this quantity in advance. Secondly, I explicitly model within-codon rate heterogeneity via a separate rate modification vector. In this way, the within-codon effect of rate heterogeneity is imposed on the model a priori, which facilitates the learning of the biologically more interesting effect of regional rate heterogeneity a posteriori. I have carried out simulations on synthetic DNA sequence alignments, which have borne out my conjecture. The existing model, which does not explicitly include the within-codon rate variation, has to model both effects with the same modelling mechanism. As expected, it was found to fail to disentangle these two effects. On the contrary, I have found that my new model clearly separates within-codon rate variation from regional rate heterogeneity, resulting in more accurate predictions.