Abstract:Graph clustering is a powerful methods to infer protein family classification from sequence only. To achieve good performance for a set of proteins that have complex intra- and inter-class relationships as in many protein superfamilies,two factors are essential:1) the similarity graph as input that contains enough information for classification and 2) a stable algorithm that can discover the obscure group structure hidden in the similarity graph. We tested a modularity optimization algorithm,called Contraction-Dilation(CD),on a set of sequences from the Pfam clan enolase with broad sequence diversity. The results show that CD outputs are in high agreement with the Pfam classification when the algorithm parameters and similarity graph are appropriately set. The fact that best performance can be achieved in a wide range around optimal settings shows the capability of this approach in general situation.
