A Method for Automatically Finding Structural Motifs in Proteins

A lot of work has gone into predicting the secondary (small scale) structure of proteins from their amino acid sequence. Current research indicates that there are limits on how well sec-ondary structure can be predicted from local sequence information. To further advance predic-tion, the interactions between elements of secondary structure which are inherently non-local, have to be better understood. This project studies a special case of secondary structure interaction, coupled helical motifs, consisting of two interacting helices. The underlying hypothesis of this work is that there are dierent types of coupled helical motifs, which can be characterized by dierent sets of rules governing the underlying amino acid sequence of the protein. In order to learn such rules, a classi cation of the coupled helical motifs needs to be introduced. This can be achieved by unsupervised learning methods such as clustering. We present a method to automatically extract structural motifs in proteins. The method uses hierarchical agglomerative clustering to nd structurally equivalent sets of motifs in proteins. These motifs can be used for study of the underlying amino acid sequence. We test the metho