Made available in DSpace on 2015-04-11T14:02:35Z (GMT). No. of bitstreams: 1 Kellen Fabiane da Silva Pinage.pdf: 968503 bytes, checksum: 287e8546d1bf57097a18d1540c5cc374 (MD5) Previous issue date: 2007-09-21 / Protein structure prediction is a process in Molecular Biology by way of which the 3D structure of a protein is determined based on known structures of other proteins. This is an important process because the structure of a protein is a determinant factor for its function in the cell. Knowing a protein s structure allows scientists to describe the kind of activity that the protein performs in the cell and to develop drugs to treat diseases. The prediction process is based on similarity between the amino acid sequences that form proteins: the structure of a target protein is predicted by re-using knowledge on proteins whose structures are already determined and whose amino acid sequences are similar to the former s. Therefore, knowledge reuse occurs in the process of predicting protein structure, but without employing a domain ontology. In this work, we apply a technique of ontology-driven knowledge re-use in protein structure prediction aiming at improving the prediction process in its efficiency and in the quality of the obtained structures. An experiment has been carried out in which the technique was applied to predict the structure of 286 target sequences. There has been improvement as well as loss of quality of predicted structures, whereas a run time performance gain in 38% of the target structures was observed. / Predição de estrutura de proteína é um processo na Biologia Molecular pelo qual a estrutura 3D de uma proteína é determinada com base em estruturas já conhecidas de outras proteínas. É um processo importante porque a estrutura de uma proteína é um fator determinante para sua função na célula. Conhecendo a estrutura de uma proteína, os cientistas podem descobrir que tipo de atividade a proteína realiza na célula e criar drogas para combater doenças. O processo de predição é baseado na similaridade entre as seqüências de aminoácidos que formam proteínas: a estrutura de uma proteína alvo é predita reusando conhecimento de proteínas cujas estruturas já foram determinadas e suas seqüências de aminoácidos são similares à seqüência alvo. Portanto, reuso de conhecimento ocorre no processo de predição de estrutura de proteína, mas sem a utilização de uma ontologia de domínio. Neste trabalho, nós aplicamos uma técnica de reuso de conhecimento baseado em ontologia na predição de estrutura de proteína com o objetivo de melhorar o processo de predição em sua eficiência e qualidade das estruturas obtidas. Um experimento foi realizado no qual a técnica foi aplicada para predizer a estrutura de 286 seqüências alvo. Houve melhorias e perdas de qualidade das estruturas preditas, ao passo que um ganho de performance (tempo de execução) foi observado em 38% das seqüências alvo.

With advancement in the development of the new laboratory instruments and experimental techniques, the protein data has an explosive increasing rate. Therefore how to efficiently store, retrieve and modify protein data is becoming a challenging issue that most biological scientists have to face and solve. Traditional data models such as relational database lack of support for complex data types, which is a big issue for protein data application. Hence many scientists switch to the object-oriented databases since object-oriented nature of life science data perfectly matches the architecture of object-oriented databases, but there are still a lot of problems that need to be solved in order to apply OODB methodologies to manage protein data. One major problem is that the general-purpose OODBs do not have any built-in data types for biological research and built-in biological domain-specific functional operations. In this dissertation, we present an application system with built-in data types and built-in biological domain-specific functional operations that extends the Object-Oriented Database (OODB) system by adding domain-specific additional layers Protein-QL, Protein Algebra Architecture and Protein-OODB above OODB to manage protein structure data. This system is composed of three parts: 1) Client API to provide easy usage for different users. 2) Middleware including Protein-QL, Protein Algebra Architecture and Protein-OODB is designed to implement protein domain specific query language and optimize the complex queries, also it capsulates the details of the implementation such that users can easily understand and master Protein-QL. 3) Data Storage is used to store our protein data. This system is for protein domain, but it can be easily extended into other biological domains to build a bio-OODBMS. In this system, protein, primary, secondary, and tertiary structures are defined as internal data types to simplify the queries in Protein-QL such that the domain scientists can easily master the query language and formulate data requests, and EyeDB is used as the underlying OODB to communicate with Protein-OODB. In addition, protein data is usually stored as PDB format and PDB format is old, ambiguous, and inadequate, therefore, PDB data curation will be discussed in detail in the dissertation.