Prediction and analysis of protein structure

Abstract

This thesis, which contains an introduction and four manuscripts, summarises my efforts during my the past four years to understand proteins, their structure and dynamics. The first manuscript presents a protocol that refines models as part of a protein structure prediction pipeline. To achieve this, we used spatial information from determined structures and sequence information from multiple alignments. The protocol was used to improve the quality of rough models containing only one point per residue. In the second manuscript we investigated protein fold space. We compared models with known fold to determined structures and found that out models contained many folds that were not seen in the present pool of structures in the PDB. Comparison of structural features revealed no reason why the model folds could not exist. We investigated how well geometric comparison methods distinguished fold in the third manuscript. We presented a novel measure of topological similarity and showed that geometric methods have trouble distinguishing fold differences between both models and PDB structures. In the last manuscript we showed that the architecture is the most important factor for dynamics as measured by normal modes. Protein fold has some effect and cannot be discarded completely, but larger differences in fold does not necessarily correspond to larger differences in flexibility if the architecture is the same.