Protein comparison based on rigid structure alignment and shape descriptor

• Main Authors: Wang, Hsin-Wei, 王信偉
• Other Authors: Pai, Tun-Wen
• Format: Others
• Language: en_US
• Published: 2014
• Online Access: http://ndltd.ncl.edu.tw/handle/vhtgr8

博士 === 國立臺灣海洋大學 === 資訊工程學系 === 102 === Since protein structures provide accurate and reliable biological insights for protein functions, structure alignment approaches have been applied to analyze or predict functions of proteins. A set of proteins sharing common structural compositions and physicochemical characteristics reflect holding similar functions. To discover functional relationships among rapidly increasing numbers of protein structures, an efficient and effective tool for structure alignment is extremely important. In this thesis, two approaches for analyzing both rigid and flexible structures in structural comparison were proposed. The first proposed multiple structure alignment (MStA) system for rigid structure assumption was based on secondary structure information, distances between neighboring residues and dynamic programming algorithm. In addition to performing structure alignment, the developed system could be applied to any existing systems and improve the alignment performance through an additional iterative refinement algorithm. Experimental results showed that the proposed method outperformed most well-known MStA tools in terms of the total number of aligned residues, root mean square deviation (RMSD), structural alignment score and running speed. Since some specific functions of a protein should be evoked by deforming its conformation to interact with other proteins or ligands, it is important to identify similarity and congruence between the original and the deformed conformations. To overcome conformational change problems, a novel descriptor, local average distance (LAD), based on either geodesic distances (GDs) or Euclidean distances (EDs) for pairwise flexible protein structure comparison was proposed. Each protein structure was transformed into an LAD profile by employing a sliding window scanning from N- to C-terminus, and the similarity between two proteins was calculated according to alignment results of corresponding profiles. The proposed LAD descriptors were compared with 7 structural alignment methods and 7 shape descriptors on two datasets comprising hinge bending motions from the MolMovDB. The results showed that the proposed LAD method outperformed all other methods for similar structure retrieval in terms of precision-recall curve, retrieval success rate, R-precision, and F1-measure. Both ED- and GD-based LAD descriptors are able to search deformed structures accurately and overcome the problems of self-connection caused by a large bending motion. Both proposed technologies are suitable for protein structure alignment or comparison, especially, the proposed algorithms provide an alternative approach for BLASTing structures from structure databases or recognizing deformed structures from inaccurate protein classification.