Aggregate Selection in Multi-objective Biochemical Optimization via the Average Cuboid Volume Indicator

The identification of peptides that optimize several physiochemical properties is an important task in the drug design process. Multi-objective genetic algorithms are efficient and cost-effective methods to scan the highly complex search space for optimal candidate peptides. A multi-objective genetic algorithm called NGSA-II has been proposed in previous work with the aim of producing diverse high-quality peptides in a low number of generations. An important component of NSGA-II is the selection process that determines the high-quality individuals for the succeeding generation. This paper presents two kinds of selection strategies for NSGA-II to guide the search process towards high-quality peptides while maintaining diversity within the genetic material. The proposed selection strategies rely both on tournaments, and use a combination of fitness-proportionate selection and a discerning selection criterion, which is front-based in one case and indicator-based in the other case. The two strategies are compared to each other with respect to the search behavior on a generic three-dimensional molecular minimization problem.