Algorithms for Analyzing Protein Conformational Flexibility with Noisy Data

Many proteins undergo changes in their structure during biochemical processes such as binding and transport. Such conformational changes are crucial to the functioning of proteins, and by extension, to most biological processes. This thesis describes algorithms for detecting conformational changes from protein structure data. Conformational changes may be determined by comparing protein structures at both ends of a conformational motion, which allows us to determine the parts of the structure that are flexible and the parts that are rigid. A number of protein structures have been experimentally determined in multiple conformational states, and this thesis describes an efficient and robust algorithm to determine conformational changes from such data. Many structures, however, have not been determined in multiple states. In such cases, it has been found that comparing protein structures of interest against related protein structures can still allow inference of conformational changes. Algorithms to detect evidence of possible conformational changes between two related protein structures are presented in this work as well. The method is extended to allow efficient scanning of a given target against a large database of related structures. A key contribution of this work is a principled approach to noisy structural data, a persistent problem with protein structures, due to uncertainty in determining precise atomic positions of atoms using curent technology. Noise is analyzed statistically by assuming that it has a Gaussian distribution. The validity of this model is affirmed by testing on many protein structures from the PDB undergoing a wide variety of conformational changes. This treatment of noise allows the algorithms presented in this work to be more robust and accurate than other competing methods.