Using extended phylogenetc profiles and support vector machines for protein family classification

We proposed a new approach to compare profiles when the correlations among attributes can be represented as a tree. To account for these correlations, the profile is extended with new bits corresponding to the internal nodes of the tree, which encode the correlations. An ad hoc scoring scheme is adopted for measuring the similarity among these extended profiles, and the scores thus obtained are then provided to a classifier -a support vector machine using a polynomial kernel function -for classification. The effectiveness of the proposed scoring scheme is assessed by the classifier’s improved accuracy. As an application the method is used to classify proteins into their functional families based on the phylogenetic profiles It is shown that the performance is much improved than using simple hamming distances and is also better than using a Bayesian based tree kernel.