Local Hierarchical Classification Techniques Analysis Using Attribute Selection for Protein Function Prediction

With the rapid advancement of researches in the genomics and proteomic areas, the growth of bases with biological data was inevitable, making the analysis of these data a Herculean task for the human beings. Thus, it was indispensable the intervention of informatics to fulfill this need. Bioinformatics is used to analyze information in the field of biology using computer techniques. One of the problems of this area is the prediction of the protein functions, which is not so common because the analysis is very laborious and complex to treat, especially when there are classes with hierarchy, that is, their classes organized in super classes that inherit Protein functions of subclasses, forming structures of trees or directed acyclic graphs. The method presented here is based on the hierarchical classification of the protein function using machine learning algorithms, thus performing the prediction of protein functions. The novelty of this work lies in the study of feature selection approaches applied to different local-model hierarchical classification approaches. The results were obtained by conducting the experiments using the hierarchical mean and standard deviation, calculated through the correct rates that the hierarchical classification algorithms obtained. From the results found, comparisons were made between the hierarchical classification methods with and without the selection of attributes, thus proving that in the prediction scenario of the protein function, which have their classes in the hierarchical format, become much more favorable with the local hierarchical ranking approach per layer and not using attribute selection.