We describe an abstract data model of protein structures by representing the geometry of proteins using spatial data types and present a framework for fast structural similarity search based on the matching of topology strings using bipartite graph matching. The system has been implemented on top of the Oracle 9i spatial database management system. The performance evaluation was conducted on 36 proteins from the Chew and Kedem data set and also on a subset of the PDB40. Our method performs well in terms of the quality of matching whilst having the advantage of fast execution and being able to compute similarity search in polynomial time. Thus, this work shows that the pre-computed string representation of topological properties between secondary structure elements using spatial relationships of spatial database management system is practical for fast structural similarity search.
展开▼
