Self-Adaptive Multi-objective Evolutionary Algorithm for Molecular Design

Self-adaptation is an efficient way to control the strategy parameters of an Evolutionary Algorithm automatically during optimization. It is based on implicit evolutionary search in the space of strategy parameters, and has been proven to work well as on-line parameter control method for a variety of strategy parameters, from local to global ones. Our proposed Self-Adaptive Multi-Objective Evolutionary Algorithm is a two level algorithm. The proposed solution is applied on the problem of de novo molecular design. The outer level is the algorithm that is responsible for the self adaptive techniques and is based on Multi-Objective Genetic Algorithm. The inner level is based on the elite Multi-Objective Evolutionary Graph Algorithm. Both the outer and inner algorithms are variations of our previously proposed Multi-Objective Evolutionary Graph Algorithm framework. The outer Multi-Objective Genetic Algorithm operates on a chromosome of elements, while the inner elite Multi-Objective Evolutionary Graph Algorithm operates on molecular graph chromosomes. In general, the proposed solution: (i) searches a larger space, (ii) generates far more solutions per iteration, (iii) evaluates different sets of parameter options for the given problem, and (iv) proposes the fittest parameter sets that should be used for the given problem.