Uso de redes neurais artificiais na simulação Monte Carlo aplicado ao problema de dobramento de proteínas

Abstract

This work proposes a new strategy to optimize the Monte Carlo method (MC) applied to the protein folding problem. This strategy is based on the information obtained from Artificial Neural Networks (ANNs), trained to predict the protein

secondary structure. The work presents, initially, background knowledge about proteins and their structure. Follows an introduction to the MC method, Neural Networks and to the prediction of secondary structure using PHD/PROF programs.

Then, a survey about tridimensional protein structure is presented. Other concepts,such as information gain in the context of hybrid systems, are also presented. Based

on state-of-the art results, a new method is proposed using the predictions produced by the PROF program, available on-line and with a performance higher than 76% for secundary structure prediction, for the reduction of the MC search space. The MC method is presented with the secondary structure prediction based on ANNs (MC-RNA) and applied to four diferent proteins obtained fro