Distinguishing medical drugs from a large set of side effects using a distributed genetic algorithm on a PC cluster

A Distributed Genetic Algorithm to compute minimal reducts is presented for a novel biomedical application to distinguish 50 medical drugs from 228 side effects. The results indicate that 15 side effects are sufficient to differentiate among all the 50 drugs. In fact, any one of 4 sets of 15 side effects can be used. The Distributed Genetic Algorithm is inherently parallel, uses a variable mutation rate and is efficiently implemented on a PC cluster using 5, 10 and 20 nodes each with a Message Passing Interface. Results show that the distributed algorithm with 20 nodes uses much less computation time than two sequential methods (savings of about a factor of 5).