A method of improving the efficiency of mining sub-structures in molecular structure databases

Abstract

One problem exists in current substructure mining algorithms is that when the sizes of molecular structure databases increase, the costs in terms of both time and space increase to a level that normal PCs are not powerful enough to perform substructure data mining tasks. After examining a number of well known molecular structure databases, we found that there exist a large number of common loop substructures within molecular structure databases, and repeatedly mining these same substructures costs the system resources significantly. In this paper, we introduce a new method: (1) to treat these common loop substructures as some kinds of "atom" structures; (2) to maintain the links of the new "atom" structures with the rest of the molecular structures, and to reorganize the original molecular structures. Therefore we avoid repeat many same operations during mining process and produce less redundant results. We tested the method using four real molecular structure databases: AID2DA'99/CA, AID2DA'99/CM, AID2DA'99 and NCI'99. The results indicated that (1) the speed of substructure mining has been improved due to the reorganization; (2) the number of patterns obtained by mining has been reduced with less redundant information. © Springer-Verlag Berlin Heidelberg 2007.