Estudo de aplicação de um método híbrido para classificação de proteínas utilizando sequence coding by sliding window e redes neurais artificiais

Abstract

This work proposes the construction of a protein classifier for the public database COG (Clusters of Orthologous groups). The classifier is built using artificial neural networks (ANN). The major problem with this method is that ANNs accept only single sizes inputs, and the proteins in COG can have many different sizes, because of that the proteins need to be coded, and the coding method used in this work is the Sequence Coding By Sliding Window (SCSW), which uses sliding windows and generates unique size vectors for any entry. The downside with this coding is that the vectors generated can be ambiguous, causing different sequences to generate equal vectors, to avoid that an ideal size for the sliding window must be found. With the encoding ready the datasets for training and validation of ANN must be selected, this selection must be done because the data obtained from COG are too large; and this makes a selection of data required. The selection is done using the Fuzzy c-means method; it selects the points and creates sets for the creation of ANN. In the end, the results obtained with the ANN in the validation set are compared with the expected results, with these results the hit rate of the network is calculated.