Finding the optimal conformation for fluorescent labeled nu-cleosides is difficult for standard GAs. We investigated the structure of molecule and of the resulting search space itself and concluded that the problem is of the "needle in the haystack" type. Using niching and island models and a crossover operator adapted to the underlying problem, the GA was able to find the global optimum. Each of these improvements increased the probability to find the global optimum from 0% (standard GA) to -30%. The combination of either niching + island model, or graph crossover + island model, increased the probability up to 80%; the combination of all three methods up to 90%.
展开▼