Predição de proteínas exportadas por meio de redes neurais artificiais

Abstract

One of the main ways of fighting diseases is vaccination. One of the phases for producing a vaccine consists of choosing some proteins that will compose them. The central objective of this paper is to identify exported proteins. These proteins have more chances to be effective when utilized on vaccines against pathogenic bacteria. We utilized Artificial Intelligence techniques like Artificial Neural Networks to classify the proteins in exported or not exported. To achieve that goal, we constructed a database of proteins extracted from the genomes of 35 species of pathogenic bacteria, utilized to train the network. After that, we implemented an Artificial Neural Network with training by the backpropagation algorithm to classify the proteins database. We prepared multiple executions with differents architectures of the network and diverse training parameters. We compared the results of these executions from the implemented network with the results of Weka. We employed ten propensity indexes (PARJ860101, JOND750101, EISD840101, JURD98010, Basic, Acid, Polar, NonPolar, Mass and Molar mass). Then, the network had a hit rate of 82.1%; meanwhile, Weka had a hit rate of 87.2%. The results were promissory, analyzing that the artificial neural network implemented had an accuracy close to the Weka. Even without optimization, the network finalized it is training on time shorter than Weka's algorithm on some executions, taking 4.7% less time to train. We implemented the artificial neural network using the language C, and the code is available on https://github.com/tiagohugovitor/Rede-Neural-TCC.