Improved Functional Prediction of Proteins by Learning Kernel Combinations in Multilabel Setting

Kernel methods have been successfully applied to a variety of biological data analysis problems. One problem of using kernels, however, is the lacking interpretability of the decision functions. It has been proposed to address this problem by using multiple kernels together with some combination rules, where each of the kernels measures different aspects of the data. Methods for learning sparse kernel combinations have the potential to extract relevant measurements for a given task.