Variational upper bounds for probabilistic phylogenetic models

摘要: Probabilistic phylogenetic models which relax the site independence evolution assumption often face problem of infeasible likelihood computations, for example task selecting suitable parameters model. We present a new approximation method, applicable wide range probabilistic models, guarantees to upper bound true data, and apply it models. The method is complementary known variational methods that lower likelihood, uses similar optimize bounds from above below. applied our aligned DNA sequences various lengths human in region CFTR gene homologous eight mammals, found be appreciably close whenever could computed. When computing exact was not feasible, we demonstrated proximity bounds, implying tight likelihood.