Mining Calcium-binding Sites From Protein Structure Graphs
Identifying protein calcium-binding sites is an important problem in proteomics. To this end, we construct a graph containing only oxygen information to represent protein partial structures. In this graph, each vertex represents an oxygen atom. Edges are given to any two vertex-atoms based on a simple distance threshold between contact atoms. Applying a clique-finding algorithm to a set of graphs representing a group of calcium-binding proteins, we obtain several hundred oxygen clique-clusters with size four possibly around calcium-binding sites. We then use geometric and chemic properties of four co-spherical vertices to exclude some clique-clusters. We finally use support vector machines (SVM) to do binary classification with vertex-atom coordinates as the input variables for distinguishing calcium-binding clique-clusters and non calcium-binding clique-clusters. The results show the site selectivity reaches 80% with 91% site sensitivity. This new protein graph mining and geometric classification model can be used for rapid and automated annotation of protein function-calcium binding
Deng, Hai, Hui Liu, and Yanqing Zhang. "Mining calcium-binding sites from protein structure graphs." In 2005 International Conference on Neural Networks and Brain. IEEE, 2005.
Proceedings of 2005 International Conference on Neural Networks and Brain Proceedings, ICNNB'05