A novel approach to protein substructure matching

The rapidly increasing volumes of structural data of proteins has led to need of algorithms which can rapidly predict functions for proteins based on structure. Similarity between protein structures can provide evidence of possible functional similarity. In this paper, an attempt is made to efficiently recognize similar protein structures in the protein database contain thousands of proteins. This paper gives an efficient heuristic algorithm for finding protein 3D substructures in a 3D protein structure that are similar to a given query 3D protein substructure. This algorithm can be used for searching a database of protein 3D structures. Our approach is to divide the protein structure into sub-structures of size of query structure and compare each sub structure with the query protein using Procrustes algorithm which is based on the root mean square distance between the structures. The division involves constructing a bounding box over both the query and protein structure and dividing the bigger box into sizes of the smaller box. The above algorithm is implemented in parallel using message passing interface. Experiments show that our algorithm can find similar 3D substructures in reasonable time. This paper also presents various statistics as how our algorithm performs against a sequential algorithm and how the algorithm performs with varying sizes of the query structure.