Ensemble of Kernel Machines for Protein Classification



Machine learning algorithms have been successfully used in proteomics, filling a gap of information in fields like protein structure and function determination. Kernel Machines like the Support Vector Machine (SVM) can build state-of-the-art classifiers for protein membership prediction when the adequate kernel functions are used. The Relevance Vector Machine (RVM) is another Kernel Machine formulation that presents theoretical advantages over the SVM, but maybe because it is not as diffused as the latter there is a lack of studies concerning its application to some specific problems like protein remote homology detection. Results from a performance study with a benchmark data set show that the RVM can learn in this scenario. However, the very low complexity of the discriminative models has also a negative impact in learning when comparing to the SVM, which stills awarding the title of best classifier for this kind of task. Here we propose a multi-class system for protein classification founded on binary classifiers that use a collection of Kernel Machines. The architecture of the ensemble is presented as well as its efficiency.




Workshop Applications of Computational Intelligence, December 2009

Cited by

No citations found