Smart Biomanufacturing and Smarter Crowds:
A Process Systems Engineering + Artificial Intelligence Approach
Dr. Yu Luo
Chemical Engineering Department
University of Delaware
Advances in systems engineering and the increasingly affordable computing power enable a paradigm shift — from heuristic-based approaches to systematic, automatic, and model-based approaches — in decision-making across different domains and industries. We will discuss this transformation with two seemingly unrelated examples and examine past, present, and future applications of process systems engineering (PSE) and artificial intelligence (AI) to decision-making. PSE and AI share substantial overlap in data analytics and optimization, while branches such as process control and intelligent agents are unique to their respective disciplines.
The production and sale of monoclonal antibody (mAb) therapeutics are a 170-billion-dollar industry to treat cancer, immune disorder, cardiovascular disease, and inflammatory diseases. Yet, mAbs are currently produced commercially in fed-batch cultures using media recipes and protocols based primarily on heuristics. The main production objectives are maximizing mAb titer and achieving desired glycan distributions that result from glycosylation—an enzymatic, post-translational process that attaches sugar molecules (glycans) to mAbs—which is a critical product quality attribute because it can either promote or inhibit a mAb drug’s therapeutic effects. To achieve the smart manufacture of mAbs, one needs to go beyond heuristics and adjust the process inputs such as feeding schedule, dissolved oxygen, and pH to control mAb titer formation and glycosylation in an automatic, rational, and efficient manner. This task remains challenging because glycosylation is a complex, non-templated process that is difficult to model, on-line glycan sensors are not commercially available, and the manipulated variables are far fewer than the controlled variables. To address these issues, we developed a modular, data-driven process model that balances predictability and flexibility, sensor fusion algorithms to reconstruct glycan information, and a model-based control system to optimize production attributes with minimum human input, by adjusting control policy iteratively during the culture.
In a separate example to be discussed, suppose we have a group of intelligent decision-makers who are already “smart,” i.e., they are capable of evaluating, improving, and optimizing their decisions. Can approaches be developed to make them “smarter”—more efficient at solving problems—by providing useful feedback? Such a question applies to a variety of scenarios ranging from demand response in smart grid management to policymaking. We developed a control-theoretic opinion dynamics model, mathematically proved that a partial adjustment of decision toward the mean could potentially make a crowd “smarter,” and tested the theory using both agent-based simulations and human-subject experiments.
Dr. Yu Luo is a postdoctoral researcher in the Department of Chemical and Biomolecular Engineering at the University of Delaware (UD). He earned his Ph.D. in Chemical Engineering at Columbia University and B.Eng. in Chemical Engineering at the National University of Singapore. Dr. Luo’s doctoral training from Columbia with Prof. Venkat Venkatasubramanian and Prof. Garud Iyengar focused on understanding complex multi-agent systems and managing systemic risk through the lens of chemical engineering (PSE in particular) and AI. His current research at UD, with Prof. Babatunde A. Ogunnaike and Prof. Kelvin H. Lee, employs various systems techniques to design and control the upstream biopharmaceutical process of manufacturing therapeutic antibodies.
Thursday, January 16 at 2:00pm to 3:00pm
Earle Hall, 100
206 S. Palmetto Blvd., Clemson, SC 29634