Dissertation

A model of amino acid replacements in protein sequences

Olsen, Rolf Miles

2004

Overview

A study is made of the patterns and rates of amino acid replacements observed between (likely) orthologous proteins sequences. These sequences are from 15 eukaryotic organisms with representatives from several of large taxonomic groups (animals, plants, yeasts, Conosa, protists). These organisms are not very closely related, the lineage divergences between the major groups occurred approximately 1 to 1½ billion years ago. The primary means of studying the observed amino acid replacements is by the design and optimization of a probabilistic model of amino acid replacements. The model of amino acid replacements is part of a larger model, intended for sequence homology classification, which was used to iteratively define the model estimation training datasets. The homology classification techniques include: the introduction of terms for relationships between clusters of phylogenetic characters; (i) comparative hypothesis tests; (ii) a proof about the limit distributions of and characterization of finite size distributions for sequence alignment scores which have different features to standard scoring systems. Simple counting studies were also done to. (i) investigate whether patterns of replacements differ with timescale; (ii) to test whether amino acids replacements are reversible; (iii) establish the correlation, on long time scales, between replacements and similarity of amino acid physical properties. Some of the empirical results of the thesis are that. (i) patterns of replacements differ with timescale and the differences on shorter timescales can be correlated to the genetic code; (ii) the hypothesis that replacements are reversible was not rejected; (iii) on timescales of a billion years replacements in protein sequences are correlated to similarities of amino acid physical properties; (iv) averaged across the timescales associated with the divergences separating the 15 eukaryotes represented in the dataset, bias in amino acid replacements is largely correlated to similarity of physical properties. The major achievement of the thesis was the development of a design procedure for reversible Markov transition matrices which can incorporate prior knowledge about amino acid replacements into the structure of the matrices and which could be applied to similar model building problems. It came through studies into the formal properties of these matrices. Design facilitated. (i) model estimation on datasets that others thought unfeasibly large; (ii) a model with superior performance to the current standard model for phylogenies on moderately large datasets.

• Publisher: ProQuest Dissertations & Theses
• Subject: Biochemistry / Biophysics
• ISBN: 9780496678075, 0496678078