Not available / text

Homology modeling is integral to structure-based drug discovery. Robust homology modeling to atomic-level accuracy requires in the general case successful prediction of protein loops containing small segments of secondary structure. For loops identified to possess α-helical segments, an alternative dihedral library is employed composed of (phi,psi) angles commonly found in helices. Even with imperfect knowledge coming from sequence-based secondary structure, helix or hairpin embedded loops, up to 17 residues in length, are successfully predicted to median sub-angstrom RMSD. Having demonstrated success with these cases, performance costs for these and other similar long loop predictions will be discussed. Dramatic improvements in both speed and accuracy are possible through the development of a Cβ-based scoring function, applicable to hydrophobic residues, that can be applied as early as half-loop buildup. With this scoring function, up to a 30-fold reduction in the cost to produce competitive sub-2 A loops are observed. Through the use of this scoring function, an efficient method will be presented to achieve ultra-high resolution buildup that restrains combinatorial explosion and offers an alternative to the current approach to full-loop buildup. This novel method is designed to be inherently suitable for homology model refinement.

Orientadores: José Mario Martínez Pérez, Margarida Pinheiro Mello / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Matemática, Estatística e Computação Científica / Made available in DSpace on 2018-08-19T12:33:50Z (GMT). No. of bitstreams: 1 Lima_RodrigoSilva_D.pdf: 3475914 bytes, checksum: e7fad42e859de59d5e4db3f0a5a41417 (MD5) Previous issue date: 2012 / Resumo: Neste trabalho estudamos inicialmente o problema da reconstrução 3D de uma proteína dadas as distâncias entre pares de átomos de sua estrutura. Formulamos a situação como um problema de otimização não linear com função objetivo contínua no domínio de variáveis e mostramos através de experimentos computacionais que a estrutura original da proteína é recuperada mesmo quando admitimos conhecidas apenas um subconjunto de distâncias intra- átomos. Em seguida, estudamos problema da representação de um conjunto de proteínas comparadas em relação as suas estruturas tridimensionais. Propomos algumas formulações para este problema onde as proteínas são representadas por objetos em espaços euclidianos e elaboramos também um procedimento para classificar proteínas novas sem a necessidade de realizar exaustivas comparações estruturais envolvendo as proteínas analisadas / Abstract: In this work we initially study the problem of reconstruct the 3D structure of a protein given the distances between pairs of its atoms. We formulate this situation as a nonlinear optimization problem with a continuous objective function over the domain of variables. We show by computational experiments that the original protein structure is recovered even when we do not use all the distances between its atoms. Next, we study the problem of representing a set of proteins. The proteins are compared with respect to their 3D structures. We propose some formulations to this problem, where the proteins are represented by objects in euclidean spaces and we elaborate also a form of use these representations to classify new proteins without perform many comparisons between the analyzed structures / Doutorado / Doutor em Matemática Aplicada