Assessing protein function using a combination of supervised and unsupervised learning

The determination of protein function using experimental techniques is time-consuming and expensive; the use of machine learning techniques to rapidly assess protein function may be useful in streamlining this process. The problem of assigning functional classes to proteins is complicated by the fact that a single protein can participate in several different pathways and thus can have multiple functions. It follows that the instances in the resulting classification problem can carry multiple class labels. We have developed a tree-based classifier that capable of handling multiply-labeled data: we call the resulting tree a recursive maximum-contrast tree (RMCT). The name derives from the way in which nodes in the tree are split; this is done by selecting the two training instances with maximum contrast (that is, the two training instances with maximum separation according to some distance measure) and using them as seeds in a clustering algorithm to form a partition of the training instances and hence of the feature space. We test our algorithm on protein phylogenetic profiles generated from 60 completely sequenced genomes, and we compare our results to those achieved using existing algorithms such as support vector machines and decision trees