A key issue in supervised protein classification is the representationof input sequences of amino acids. Recent work using string kernels for protein data has achieved state-of-the-artclassification performance. However, such representations are basedonly on labeled data --- examples with known 3D structures, organized intostructural classes ---while in practice, unlabeled data isfar more plentiful.In this work, we develop simple and scalable clusterkernel techniques for incorporating unlabeled data into therepresentation of protein sequences. We show that our methodsgreatly improve the classification performance of string kernels andoutperform standard approaches for using unlabeled data, such asadding close homologs of the positive examples to the training data.We achieve equal or superior performance to previously presented cluster kernelmethods while achieving far greater computational efficiency.
Semi-Supervised Protein Classification using Cluster Kernels.
Jason Weston, Christina Leslie, Dengyong Zhou, Andre Elisseeff and William Stafford Noble