Prediction and analysis of functionally important sites in protein structure

Identifying the molecular role of a target protein is becoming a pressing issue in structural bioinformatics. As the number of uncharacterised proteins accumulate and structural genomics initiatives pick up pace reliable and accurate approaches are required to make sense of the growing structural data in the most effective manner. This study aims to improve the characterisation and cataloguing of functionally important regions in proteins to improve the transfer of information from structure to functional annotation. This is of great importance as effective and accurate annotations are a pre-requisite to unlocking the biological information encoded within the genomes of organisms. A novel and automatic method is presented which identifies residues interacting with metal ions. The aim has been to focus the approach such that it is capable of the prediction of interactions even when reliable side-chain information is unavailable. The results demonstrate that a combination of sequence and structural features can be combined to achieve adequate predictions in both crystal structures as well as low resolution fold recognition models. The findings of the metal binding study are used to extend the approach and predict larger interaction regions found in DNA binding proteins. These important class of proteins are of particular interest given their fundamental control over biological processes. The results highlight effective classification is possible of both residues forming DNA contacts as well as the discrimination of DNA binding from non-DNA binding proteins. Crucially, DNA binding predictions are shown for a genome-wide study of Sacchromyces cerevisiae. The final study of the thesis focuses the attention on the alternative problem of improving the quality of protein structure predictions from fold recognition. The results indicate that site classification provides an effective basis by which protein models can be assessed: this in turn leads to the development and benchmarking of a new method which significantly improves the quality of protein models.