Penalised complexity priors for the reconstruction of past population size from phylogenies
Carregando...
Data
2021-12
Orientador(res)
Carvalho, Luiz Max Fagundes de
Métricas
Título da Revista
ISSN da Revista
Título de Volume
Resumo
This work seeks to understand the influence of prior distributions on inferences about
population dynamics obtained from phylogenies. More specifically, considering the posterior
results obtained from the Skyride model (Minin et al., 2008). The Skyride model assumes
that the effective population size has a smooth trajectory and assigns a Gaussian Markov
Random Fields (GMRF) prior to the logarithm of the effective population size. This model
has an unknown precision parameter τ , associated with the smoothness of the curve. We
propose an analysis investigating the choice of the prior for the τ parameter. In particular,
we study the penalised complexity priors in comparison with the default choice of the
Gamma distribution. A review of the literature was conducted, considering previous studies on the estimation
of effective population size using phylogenies. Then, we conducted simulation studies to
estimate the effective population size from phylogenetic trees. We run the experiments for
phylogenies generated with the phylodyn package. We looked at populations with constant,
exponential, cyclic, and bottleneck effect trajectories, and then performed experiments
analyzing a real dataset of influenza A/H3N2. Finally, through the performance evaluation
metrics, it was possible to show that a PC prior leads to more accurate results, regardless
of the number of leaves in the phylogenetic tree.
Descrição
Trabalho de conclusão de curso