Predição da função das proteí­nas sem alinhamentos usando máquinas de vetor de suporte.

Abstract

This thesis presents a new model to protein function prediction using support vector machines, a machine learning approach trained using structural parameters calculated from protein tertiary structure. The model is different from the others paradigms because it is not necessary to search for similarities against the others known proteins in public databases by alignments. In this way, the model is able to associate functional relationships among proteins with no similarities and it could be used when all other methods fail or when the user don t want to use the concept of similarity in function predictions. The proof that the model is valid was accomplished analyzing its performance with unknown proteins, i.e proteins not used in the training set. The validation approach used a set of binding proteins.