UC Berkeley

This thesis explores methods for computational comparative modeling of genetic sequences. The framework within which this modeling is undertaken is that of sequence alignments and associated phylogenetic trees. The first part explores methods for building ancestral sequence alignments making explicit use of phylogenetic likelihood functions. New capabilities of an existing MCMC alignment sampler are discussed in detail, and the sampler is used to analyze a set of HIV/SIV gp120 proteins. An approximate maximum-likelihood alignment method is presented, first in a tutorial-style format and later in precise mathematical terms. An implementation of this method is evaluated alongside leading alignment programs. The second part describes methods utilizing multiple sequence alignments. First, mutation rate is used to predict positional mutational sensitivities for a protein. Second, the flexible, automated model-specication capabilities of the XRate software are presented. The final chapter presents recHMM, a method to detect recombination among sequence by use of a phylogenetic hidden Markov model with a tree in each hidden state.