Increasing demand for recombinant proteins (including monoclonal antibodies) where time to market is critical could benefit from the use of model-based optimization of cell viability and productivity. Owing to the complexity of metabolic regulation, unstructured models of animal cell cultures typically have built-in errors (structural and parametric uncertainty) which give rise to the need for obtaining relevant data through experimental design in modeling for optimization. A Bayesian optimization strategy which integrates tendency models with iterative policy learning is proposed. Parameter distributions in a probabilistic model of bioreactor performance are re-estimated using data from experiments designed for maximizing information content and productivity. Results obtained highlight that experimental design for run-to-run optimization using a probabilistic tendency model is effective to maximize biomass growth even though significant model uncertainty is present. A hybrid cybernetic model of a myeloma cell culture coconsuming glucose and glutamine is used to simulate data to demonstrate the efficacy of the proposed approach.