The production of many biotherapeutics starts with an optimized production system, such as one based on Chinese hamster ovary (CHO) cells. Optimizing the culture media, though, can be complex and expensive. So, Michael Betenbaugh, PhD, professor of chemical and biomolecular engineering at Johns Hopkins University, and his colleagues applied Bayesian optimization to the process.
In a pre-proof for iScience, they reported that their method works with multiple parallel experiments, includes constraints based on thermodynamics, and “bridges machine-learning and physical modeling to create a more data-efficient process optimization strategy.”
Betenbaugh’s team used a combination of advanced machine learning techniques, thermodynamic modeling, and laboratory automation to improve how media is formulated.
The approach allowed many experiments to be run at the same time using special lab equipment and a smart algorithm called multi-scale multi-recommendation. Bayesian optimization helped the system quickly learn which nutrients and conditions lead to better outcomes, like higher protein production. The goal was to find the right levels of ingredients, such as amino acids, glucose, and vitamins.
However, there were challenges. Some amino acids can only dissolve up to a certain point before they start forming solid crystals that can ruin the media and result in failed production batches. Even worse, certain amino acids can affect the solubility of others—sometimes helping, sometimes hurting. To address this, the team combined a special solubility-prediction model with their machine learning system. That ensured that the recommended media recipes were both effective and physically possible.
The method was first tested using computer simulations of CHO cell metabolism. It was then successfully applied in real lab experiments using automated bioreactors.
Overall, the study showed that using a hybrid system of machine learning, solubility modeling, and robotics leads to better, faster, and more reliable media design. This method could transform how future biotherapeutics are developed—speeding up production and improving outcomes, especially in urgent situations like pandemics or complex disease treatment needs.
