An Empirical Study on the Influence of Genetic Operators for Molecular Docking Optimization

Jorge Tavares (1) , Nouredine Melab (1, 2) , El-Ghazali Talbi (1, 2)

1 DOLPHIN - Parallel Cooperative Multi-criteria Optimization
2 LIFL - Laboratoire d'Informatique Fondamentale de Lille

Evolutionary approaches to molecular docking typically use a real-value encoding with standard genetic operators. Mutation is usually based on Gaussian and Cauchy distributions whereas for crossover no special considerations are made. The choice of operators is important for an efficient algorithm for this problem. We investigate their effect by performing a locality, heritability and heuristic bias analysis. Our investigation focus on encoding properties and how the different variation operators affect them. It is important to understand the behavior and influence of these components in order to design new and more efficient evolutionary algorithms for the molecular docking problem. Results confirm that high locality is important and explain the behavior of different crossover and mutation operators. In addition, the heritability and heuristic bias study provides some insights in how the different crossover operators perform. Optimization runs in different instances of the problem support the analysis findings. The performance and behavior of the variation operators are consistent on several molecules.

Domaines: Intelligence artificielle [cs.AI]

Jorge Tavares, Nouredine Melab, El-Ghazali Talbi. An Empirical Study on the Influence of Genetic Operators for Molecular Docking Optimization. [Research Report] RR-6660, INRIA. 2008. ⟨inria-00323625⟩