Computational Analysis of Protein Sequence and Structure

Robert Matthew MacCallum

September 1997

Biomolecular Structure and Modelling Group
Department of Biochemistry and Molecular Biology
University College London
Gower Street
London WC1E 6BT

Abstract:

This project has combined the structural and functional analysis of proteins, protein structure prediction, and sequence analysis.

A study of antibody-antigen interactions has been undertaken. We have analysed antigen-contacting residues and combining site shape in the antibody crystal structures available in the Protein Data Bank. Antigen-contacting propensities are presented for each antibody residue, allowing a new definition for the complementarity determining regions to be proposed based on observed antigen contacts. An objective means of classifying protein surfaces by gross topography has been developed and applied to the antibody combining site surfaces. The surfaces have been clustered into four topographic classes, confirming suggestions in the literature that antigen type might influence the shape of the whole combining site.

The prediction of secondary structural class and architecture from sequence composition analysis has also been investigated. Modifications to a well established geometric prediction algorithm have led to improvements in accuracy and the estimation of reliability. The hierarchical prediction of fold architectures using these methods is then discussed.

To complement the ab initio approach of class and architecture prediction, a novel sequence alignment algorithm employing direct comparisons of predicted secondary structure and sequence-derived hydrophobicity was developed, and applied to fold recognition. The method, called SIVA, appears to perform well when sufficient multiple sequence information is available, although further testing, including blind public testing, is required.

Kohonen's self-organising map was applied to protein structure and sequence information, with the hope that fold recognition using SIVA could be improved using essential sequence features extracted by this technique. This was not successful, however the potential of the mapping approach has been illustrated, and a number of specific applications have been suggested.