(351as) Experimental Design Algorithm for Efficient Optimization of Culture Media

Design and analysis of culture media for bioprocesses often involves the use of many time-consuming experiments. However, because of the diversity of metabolites needed in many industrially relevant bioprocesses, and their nonlinear effects, design optimization is especially time-consuming and difficult to perform. This difficulty can be managed by combining a surrogate model with an optimization method to select response-optimal design parameters, often called response surface methods. Here, we use a radial basis function to learn metabolite-response interactions from benchtop scale experiments using C2C12 cells. This approximate model then feeds into a hybrid global/local stochastic optimization method to iteratively suggest optimal metabolite concentrations from a user-defined list of metabolites and objective function. This algorithm was used to design more efficient proliferation media for C2C12s in fewer experiments than traditional polynomial response surface methods, so can be used to reduce the cost of benchtop experiments in combinatorally complex experimental design problems.