This thesis is concerned with developing two-stage Bayesian optimum design procedure in nanotoxicology with simulation experiments. The proposed design method adopts suitable nonlinear dose response curve and non-constant variance model in experimental modeling, which more adequately represents the desirable properties of bioassay-experiment. A multiple objective genetic algorithm were used to select best set of non-dominated experiments on Pareto front. The proposed approach has been shown to provide better design than traditional experiment method in terms of D-optimum design and A-optimum design. |