The search for structural similarity among proteins can provide valuable insights into their functional mechanisms and their functional relationships. Though the protein 1D sequence contains the information of protein folding, the performance of predicting the 3D-structure directly from the sequence is still limited. As the increase of available protein structures, we can now conduct more precise and thorough studies of protein structures. Among many is the design of protein structural alphabet that can characterize protein local structures. We use the self-organizing map combined with the minimum spanning tree algorithm for visualization to determine the alphabet size and then apply the k-means algorithm to group protein fragments into clusters corresponding to the structural alphabet. The intra-cluster and inter-cluster analyses show the significant structural cohesiveness. A comparative study of our alphabet with one of the recently developed structural alphabets also demonstrated a competitive result.
展开▼
