A new method for clustering mult-domain protein sequences

A new method for clustering multi-domain protein sequences was proposed by revising preference value of classical affinity propagation (AP) algorithm combined by Silhouette index of clustering validity. At the same time, the classical substitution match similarity (SMS) between two protein sequences was generalized to meet the demand of clustering `twilight zone' protein sequences. Experimental results on four test datasets demonstrate that our method can acquire number of clusters more approximate to the family number of clusters classified by the phylogenetic trees, more consistence clustering structure for a given dataset of proteins, and the comparatively advantage in clustering multi-domain protein sequences.