Efficient remote homology detection

We propose an effcient multi-class classiffcation algorithm for remote homology detection (RHD). Unlike methods that treat RHD as a set of binary classiffcation tasks, our algorithm solves a single multiclass classiffcation problem by incorporating information about classwise correlations among the proteins using joint kernel functions. Furthermore,the proposed method leads to notable reduction in computational time compared to binary classiffcation algorithms. We evaluate our method on the Structural Classiffcation of Proteins database and show that performance is better or comparable to several state-of-theartalgorithms for protein classiffcation and remote homology detection.