Development of protein modeling methods for structure refinement in the context of unreliable environments

Abstract

The number of experimentally determined protein structures is increasing exponentially. Based on this abundant structural information, homology modeling is now the most popular method for protein structure prediction. Still however, knowledge of high resolution structures is critical for applications using the protein structure such as drug discovery and protein design. By realizing this, protein structure refinement methods have been developed to improve the structure quality of low resolution experimental structures or model structures. Another realm of protein structure refinement is to predict the protein structure in the environment of interest, such as binding to a specific partner, when only structures resolved in different conformational states or model structures are provided. In this thesis, four modeling methods (GalaxyLoop-PS2, GalaxyRefine2, GalaxyVoyage, and Galaxy7TM) developed in the scope of refining predicted protein structures are introduced. The methods were evolved by either extending the range of structure targeted for refinement or considering the interaction with a particular binding partner. The shared problem of these methods was that the environment of modeling was unreliable due to errors embedded in model structures. Commonly, two approaches were taken to tackle this problem. These were initially searching the conformational space in low resolution and developing a hybrid energy function less sensitive to environmental error. The development and application results of the approaches taken for each modeling method will be addressed in detail.