%A D.P. Klose
%D 2008
%T Protein model construction and evaluation.
%L discovery1444214
%R PQ ETD:591516
%I University of London
%O Thesis digitised by ProQuest

%X The prediction of protein secondary and tertiary structure is becoming increasingly important as the number of sequences available to the biological community far exceeds the number of unique native structures. The following chapters describe the conception, construction, evaluation and application of a series of algorithms for the prediction and evaluation of two and three-dimensional protein structure. In chapter 1 a brief overview of protein structure and the resources required to predict protein features is given. Chapter 2 describes the investigation of sequence identity and alignments on the prediction of two-dimensional protein structure in the form of long and short range protein contacts a feature which is known to correlate with solvent accessibility. It also describes the identification of a feature which is referred to as the 'Empty Quarter' which forms the basis of an evaluation function described in Chapter 3 and developed in Chapter 4. Chapter 3 introduces the Dynamic Domain Threading method used during round six of the CASP exercise. Phobic, a protein evaluation function based on predicted solvent accessibility is described in Chapter 4. The de novo prediction of a/p proteins is described in Chapter 5, the method introduces a new approach to the old problem of combinatorial modelling and breaks the size limit previously imposed on de novo prediction. The final experimental chapter describes the prediction of solvent accessibility and secondary structure using a novel combination of the fuzzy k-nearest neighbour and support vector machine. Chapter 7 closes this piece of work with a review of the field and suggests potential improvements to the way work is conducted.