Tree-structured maximal common subgraph searching. An example of parallel computation with a single sequential processor

Abstract

A new maximal common subgraph search algorithm (MCS-1) is described and applied to selection of starting materials for synthesis of a target molecule. This algorithm achieves parallelism of up to 20-fold with a single processor by using a unique storage tree. The storage tree requires only 79% of the storage required by a sequential organization and the storage tree requires only 28% of the bonds to be represented. The algorithm continues to improve relative to the sequential organization as the number of compounds in the library increases. The efficiency of the algorithm was evaluated with four queries for ten different sized libraries, ordered in various ways. An execution trace of MCS-1 is included.