An Approach for Clustering Protein Pockets into Similar Groups∗

In this work, we propose a network-based approach to cluster the protein pockets into similar groups in a database level. A pocket similarity network is constructed to describe structural similarity relationships among the pockets from the systematic perspective, which possesses the community structures and can can be utilized to develop a direct method to cluster the pockets by partitioning the network to small communities, which correspond to pocket groups individually. As a first step, we join the pockets into structurally similar pocket groups via a hierarchical process guided by maximizing a widely used modularity measurement Q. Then we analyze the functional similarity underlying every divided pocket groups. As a result most of the pockets in the same group are identified to share similar functions. These results show that our clustering method is effective and efficient to reveal biologically meaningful pocket groups regard to functional consistence.