Thawornwattana, Yuttapong; (2018) Bayesian inference in molecular phylogeography using Markov chain Monte Carlo. Masters thesis (M.Phil), UCL (University College London).

Preview

Text thawornwattana-2018-mphil-thesis.pdf Download (2MB) | Preview

Abstract

The Bayesian approach to phylogenetic inference allows quantification of all aspects of uncertainty using probability. Markov chain Monte Carlo (MCMC), a class of algorithms based on iterative simulation, is often considered a gold standard for approximate Bayesian inference. However, MCMC is computationally intensive and there are many design decisions to be made when using it in practice. We discuss few principles for designing simple and efficient MCMC algorithms. In particular, we propose several new proposal kernels for MCMC based on the idea of introducing negative correlations in the simulation draws. In many cases, these kernels can lead to efficiency >100%. Using practical examples, we illustrate that a sequence of well-designed one-dimensional proposals can be more efficient than a single d-dimensional proposal, and that variable transformations can be used as a general strategy for designing efficient MCMC. Next, we turn to the problem of species tree inference in the Anopheles gambiae species complex from whole-genome data. This is a challenging problem due to complex effects of recent and rapid radiation, introgression, chromosome inversions and natural selection. We extract over 80,000 coding and noncoding loci from the genomes of six members of this species complex and perform Bayesian inference using MCMC under the multispecies coalescent model, which takes into account genealogical heterogeneity across the genome and uncertainty in the gene trees. We obtain a robust species tree estimate, consistent with chromosome inversions. Using simulation informed by the real data, we conclude that species trees from previous studies are erroneous as a result of methodological artefacts. We also found evidence of gene flow between certain pairs of species based on direct estimation of migration rates under the isolation-with-migration model. The results highlight the importance of accommodating incomplete lineage sorting and introgression in phylogenomic analyses of species that arose through recent radiative speciation events.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item