O uso de rede neural artificial MLP na predição de estruturas secundárias de proteínas
Carregando...
Data
2004-06-23
Autores
Ferreira, Fausto Roberto [UNESP]
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A predição de estruturas secundárias e terciárias pode contribuir para elucidar o problema de enovelamento de proteínas. Para isso, métodos de Redes Neurais Artificiais (RNAs) e Algoritmos Genéticos são utilizados a fim de predizê-las, a partir de determinadas seqüências primárias de aminoácidos. Neste sentido, esta pesquisa visa à utilização de três níveis de RNAs. O primeiro nível é composto por um vetor de entrada representando a seqüência primaria dos aminoácidos, com uma dimensão de 22.n, onde n é o tamanho da janela compreendida entre 7 a 23. O segundo nível possui a implementação dos resultados da primeira rede. Por fim o terceiro nível é composto por um júri de decisão. As RNAs são treinadas no Simulador MATLAB 5.0, um software composto de vários recursos para a sua implementação (Neural Network Toolbox). As RNAs implementadas são do tipo Multi Layer Perceptron (MLP), que utilizam o algoritmo backpropagation (RPROP) e a função de treinamento trainrp. Os dados obtidos são comparados com os preditores 'The Predict Protein Server Default' (www.emblheidelberg.de/predictprotein/submit_def.html), 'The PSA Protein Structure Prediction Server' (http//bmerc-www.bu.edu/psa/request.html) e 'The PSIPRED Protein Structure Prediction Server' (http://bioinf.cs.ucl.ac.uk/psipred/), a fim de se obter um modelo de predição.
The prediction of (secondaray and tertiary) structures of proteins can contribute to elucidadate the protein-folding problem. In oder to predict these structures we used methods of Artificial Neural Network (ANN) and genetic algorithms starting from the primary sequences of amino acids. The present work is composed of 3 networks levels. The first level is composed of ANNs of an input vector representing a segment of primary amino acid sequence. Since the encoding scheme uses a local window into the sequence, the input vector is a 22.n dimensional vector where n is the number of positions in the window (between 7 and 23). The outputs of level 1 are the inputs of the second level ANNs. The third level is the jury decision. The ANNs were trained with the Simulator MATLAB 5.0, software with several tools for its implementation (Neural Network Toolbox). The implemented ANNs are Multi Layer Perceptron (MLP) kind, which use the backpropagation algorithms (RPROP) together with training function trainrp. The obtained date are compared with the predictors 'The Predict Protein Server Default' (www.emblheidelberg.de/predictprotein/submit_def.html), 'The PSA Protein Structure Prediction Server' (http//bmerc-www.bu.edu/psa/request.html) e 'The PSIPRED Protein Structure Prediction Server' (http://bioinf.cs.ucl.ac.uk/psipred/) in order to heve an idea of the quality of the prediction.
The prediction of (secondaray and tertiary) structures of proteins can contribute to elucidadate the protein-folding problem. In oder to predict these structures we used methods of Artificial Neural Network (ANN) and genetic algorithms starting from the primary sequences of amino acids. The present work is composed of 3 networks levels. The first level is composed of ANNs of an input vector representing a segment of primary amino acid sequence. Since the encoding scheme uses a local window into the sequence, the input vector is a 22.n dimensional vector where n is the number of positions in the window (between 7 and 23). The outputs of level 1 are the inputs of the second level ANNs. The third level is the jury decision. The ANNs were trained with the Simulator MATLAB 5.0, software with several tools for its implementation (Neural Network Toolbox). The implemented ANNs are Multi Layer Perceptron (MLP) kind, which use the backpropagation algorithms (RPROP) together with training function trainrp. The obtained date are compared with the predictors 'The Predict Protein Server Default' (www.emblheidelberg.de/predictprotein/submit_def.html), 'The PSA Protein Structure Prediction Server' (http//bmerc-www.bu.edu/psa/request.html) e 'The PSIPRED Protein Structure Prediction Server' (http://bioinf.cs.ucl.ac.uk/psipred/) in order to heve an idea of the quality of the prediction.
Descrição
Palavras-chave
Proteínas - Estrutura, Redes neurais (Computação), Algoritmos genéticos, Bioinformática, Predição - Estrutura
Como citar
FERREIRA, Fausto Roberto. O uso de rede neural artificial MLP na predição de estruturas secundárias de proteínas. 2004. 83 f. Dissertação (mestrado) - Universidade Estadual Paulista, Instituto de Biociências, Letras e Ciências Exatas, 2004.