Efficient search and comparison algorithms for 3D protein binding site retrieval and structure alignment from large-scale databases
Metadata[+] Show full item record
Finding similar 3D structures is crucial for discovering potential structural, evolutionary, and functional relationships among proteins. As the number of known protein structures has dramatically increased, traditional methods can no longer provide the life science community with the adequate informatics capability needed to conduct large-scale and complex analyses. A suite of high-throughput and accurate protein structure search and comparison methods is essential. To meet the needs of the community, we develop several bioinformatics methods for protein binding site comparison and global structure alignment. First, we developed an efficient protein binding site search that is based on extracting geometric features both locally and globally. The main idea of this work was to capture spatial relationships among landmarks of binding site surfaces and bfuild a vocabulary of visual words to represent the characteristics of the surfaces. A vector model was then used to speed up the search of similar surfaces that share similar visual words with the query interface. Second, we developed an approach for accurate protein binding site comparison. Our algorithm provides an accurate binding site alignment by applying a two-level heuristic process which progressively refines alignment results from coarse surface point level to accurate residue atom level. This setting allowed us to explore different combinations of pairs of corresponding residues, thus improving the alignment quality of the binding site surfaces. Finally, we introduced a parallel algorithm for global protein structure alignment. Specifically, to speed up the time-consuming structure alignment process of protein 3D structures, we designed a parallel protein structure alignment framework to exploit the parallelism of Graphics Processing Units (GPUs). As a general-purpose GPU platform, the framework is capable of parallelizing traditional structure alignment algorithms. Our findings can be applied in various research areas, such as prediction of protein inter
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License.