(448b) Multi-Information Source / Objective Bayesian Optimization of Serum-Free Cell Culture Media for Cellular Agriculture

The emerging field of cellular agriculture, in which bioreactors are used to grow mammalian cells for human consumption in leu of carbon-intensive animal agriculture, is predicated on use of inexpensive and serum-free (no animal components) culture media. Designing such a media is a difficult task due to the number of metabolites in use, nonlinear component-wise interactions, and the difficulty in making measurements that reflect the complex dynamics of cell growth and survival. We address these issues by using a Bayesian optimization algorithm in which a multi-information source Gaussian process model explores the media design space while maximizing a multi-objective (expected hypervolume improvement) acquisition function to find optimal media component concentrations. Using this acquisition function the uncertainty in the Gaussian process model us iteratively reduced while the cell growth metric and component cost are maximized and minimized respect. Multiple cell growth assays are used in concert, increasing the quality of the model predictions. The result was a methodology that discovered a high-growth / minimized cost medium in few experiments while exploring the pareto trade-off curve between cell growth and medium cost.