High throughput screening of global and local protein surfaces

Sael Lee, Purdue University

Abstract

The comparative study of protein tertiary structures provides rich information for investigating the function and evolution of proteins, which are the constructors and maintainers of all living things. Traditionally, comparative studies have focused on using genetic sequences to study the function and the evolution of proteins. However, with the accumulation of structural information, more direct approaches for gaining information through protein structure comparisons are also possible. These require methods that are able to quickly find proteins of similar structure over large datasets. However, most of the existing structure comparison methods use full structure alignment methods that are not quick enough for database searching. In this dissertation, methods for rapid comparison of protein tertiary structures are proposed. For this purpose, among the protein tertiary structure representations, the protein surface is used. The shape, physicochemical properties and other properties of the protein surface determine the recognition process of other proteins, ligands, DNA, or other molecules with which it interacts. To enable quick searching, the 3D Zernike descriptor, one of the object abstraction methods used in the graphics field, is used to efficiently represent and compare protein surfaces. The most attractive aspect of the 3D Zernike descriptor is rotational invariance. Rotation invariance means that structural alignment is not necessary, which speeds up the searching process. This dissertation makes three major contributions: 1) the development of a rapid method for protein function prediction using the global 3D shape, electrostatic potential, and hydrophobicity of the protein surface; 2) the development of local protein surface comparison methods and their application to ligand binding prediction; 3) the characterization and classification of local protein surface patches. The first method assumes that the protein structure relationships are preserved by evolution even when proteins share little sequence similarity. The latter two approaches assume that there are structural similarities among proteins of similar function even when structures are globally different. Local methods directly search for geometrical and/or physicochemical properties of significant sites; it is possible to predict molecular functions of proteins that lack global sequence/structural homology to proteins of known function.

Degree

Ph.D.

Advisors

Kihara, Purdue University.

Subject Area

Bioinformatics|Computer science

COinS